#include "OwnerWorldLoader.h"
#include "StreamingWorld.h"
#include "StandaloneCoreScheduler.h"
#include "../Core/src/StandaloneSnapshotAdapter.h"
#include "../Research/OwnerShapeWorld/gpu_field_port/GpuFarViewer.h"
#include "VoxelDensityProfile.h"

#ifdef TEXT
#undef TEXT
#endif
#ifndef NOMINMAX
#define NOMINMAX
#endif
#ifndef WIN32_LEAN_AND_MEAN
#define WIN32_LEAN_AND_MEAN
#endif

#include <d3d11.h>
#include <d3dcompiler.h>
#include <dxgi.h>
#include <windowsx.h>

#include <atomic>
#include <bit>
#include <complex>
#include <cstdio>
#include <cstdlib>
#include <deque>
#include <filesystem>
#include <fstream>
#include <iomanip>
#include <iostream>
#include <iterator>
#include <limits>
#include <map>
#include <numeric>
#include <optional>
#include <sstream>
#include <thread>
#include <unordered_map>
#include <unordered_set>

#pragma comment(lib,"d3d11.lib")
#pragma comment(lib,"dxgi.lib")
#pragma comment(lib,"d3dcompiler.lib")
#pragma comment(lib,"user32.lib")
#pragma comment(lib,"gdi32.lib")

namespace
{
    struct V3 { float X=0.0f,Y=0.0f,Z=0.0f; };
    V3 Add(V3 A,V3 B){return {A.X+B.X,A.Y+B.Y,A.Z+B.Z};}
    V3 Sub(V3 A,V3 B){return {A.X-B.X,A.Y-B.Y,A.Z-B.Z};}
    V3 Mul(V3 A,float S){return {A.X*S,A.Y*S,A.Z*S};}
    float Dot(V3 A,V3 B){return A.X*B.X+A.Y*B.Y+A.Z*B.Z;}
    V3 Cross(V3 A,V3 B){return {A.Y*B.Z-A.Z*B.Y,A.Z*B.X-A.X*B.Z,A.X*B.Y-A.Y*B.X};}
    V3 Normalize(V3 A){return Mul(A,1.0f/std::sqrt(std::max(1.0e-20f,Dot(A,A))));}

#include "FarSurvivalAudit.h"

    struct FMeshSheetAudit
    {
        uint64 Components=0;
        uint64 SmallComponents=0;
        uint64 SmallComponentsInView=0;
        uint64 BoundarySmallComponentsInView=0;
        uint64 ThinSheets=0;
        uint64 TrianglesInThinSheets=0;
        double MinimumThicknessVoxels=std::numeric_limits<double>::max();
    };

    struct FMeshArtifactAudit
    {
        uint64 Tiles=0,SurfaceVertices=0,EdgeVertices=0,MidpointVertices=0;
        uint64 SkirtTriangles=0,HiddenSameLodSkirtTriangles=0,ExposedSkirtTriangles=0,
            ReverseSideSolidSkirtTriangles=0,RepeatedOffsetPatches=0,FieldMismatchPatches=0,
            NormalMismatchPatches=0,GradientFallbackPatches=0,GradientFallbackSamples=0,
            DetailDisplacementSamples=0,HighDetailDisplacementSamples=0,DetailDisplacementPatches=0,
            CurvedEdgeChecks=0,CurvedEdgeRootPatches=0,CurvedEdgeRootBadVertices=0,
            DetailGradientSamples=0,DetailGradientOverLimit=0;
        double MaximumAirSkirtDensity=0.0,MaximumPatchDominance=0.0;
        double MaximumDetailDisplacement=0.0;
        double MaximumCurvedEdgeRootError=0.0;
        double MaximumDetailGradient=0.0;
        uint64 FieldEdgeChecks=0,AnalyticCrossingMisses=0,OldInt8RangeEdges=0,NormalChecks=0;
        double SumAbsSourceResidual=0.0,MaximumAbsSourceResidual=0.0,
            SumAnalyticEdgeShiftVoxels=0.0,MaximumAnalyticEdgeShiftVoxels=0.0,
            SumNormalAgreement=0.0,MinimumNormalAgreement=1.0;
        std::array<double,3> WorstShiftPosition{},WorstResidualPosition{};
        double WorstShiftFraction=0.0,WorstShiftD0=0.0,WorstShiftD1=0.0,
            WorstShiftExactD0=0.0,WorstShiftExactD1=0.0;
        int32 WorstShiftAxis=-1;
        std::array<uint64,16> OffsetHistogram{};
    };

    uint8 ComputeVisibleSkirtFaceMask(const FVoxelTileKey& Tile,
        const std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash>& Desired)
    {
        const FIntVector Origin=Tile.OriginVoxels();
        const int32 Extent=Tile.ExtentVoxels();
        const auto FloorDiv=[](int64 Value,int64 Divisor)
        {
            int64 Quotient=Value/Divisor;
            if(Value%Divisor<0)--Quotient;
            return static_cast<int32>(Quotient);
        };
        uint8 Mask=0;
        for(int32 Axis=0;Axis<3;++Axis)
        for(int32 Side=0;Side<2;++Side)
        {
            int64 Probe[3]={Origin.X+Extent/2,Origin.Y+Extent/2,Origin.Z+Extent/2};
            Probe[Axis]=Side?static_cast<int64>((&Origin.X)[Axis])+Extent
                :static_cast<int64>((&Origin.X)[Axis])-1;
            int32 NeighborLevel=-1;
            for(int32 Level=0;Level<=8;++Level)
            {
                const int64 NeighborExtent=static_cast<int64>(CHUNK_SIZE)<<Level;
                const FIntVector Coord(FloorDiv(Probe[0],NeighborExtent),
                    FloorDiv(Probe[1],NeighborExtent),FloorDiv(Probe[2],NeighborExtent));
                if(Desired.contains(FVoxelTileKey(Coord,Level)))
                {NeighborLevel=Level;break;}
            }
            if(NeighborLevel<0||NeighborLevel!=Tile.Level)
                Mask|=static_cast<uint8>(1u<<(Axis*2+Side));
        }
        return Mask;
    }

    FMeshArtifactAudit AuditMeshArtifacts(const FStandaloneCpuTile& Tile,
        const VoxelForgeCore::World& Shape,double SkirtDepthMeters,uint8 VisibleSkirtFaceMask,
        bool bPrintSkirts,bool bPrintPatches)
    {
        FMeshArtifactAudit Result;
        if(Tile.Key.Level<0||Tile.Key.Level>1||Tile.Vertices.empty()||Tile.Indices.size()<3)
            return Result;
        ++Result.Tiles;
        const size_t VertexCount=Tile.Vertices.size();
        const size_t SurfaceVertices=std::min<size_t>(Tile.SurfaceVertexCount,VertexCount);
        Result.SurfaceVertices=SurfaceVertices;
        const int32 Step=1<<Tile.Key.Level;
        const FIntVector Origin=Tile.Key.OriginVoxels();

        using FBlockKey=std::array<int32,4>; // interpolated axis, then the 8-voxel block xyz
        struct FFieldBlock
        {uint64 Vertices=0,BadVertices=0;double MaximumShift=0.0,MaximumResidual=0.0;
            std::array<double,3> Position{};};
        struct FNormalBlock
        {uint64 Vertices=0,BadVertices=0;double MinimumAgreement=1.0;
            std::array<double,3> Position{},MeshNormal{};};
        struct FGradientFallbackBlock
        {uint64 Vertices=0,BadVertices=0,BadSamples=0;double MaximumError=0.0;
            std::array<double,3> Position{};};
        struct FCurvedEdgeRootBlock
        {uint64 Vertices=0,BadVertices=0;double MaximumError=0.0;
            std::array<double,3> Position{};};
        std::map<FBlockKey,std::array<uint64,256>> OffsetBlocks;
        std::map<FBlockKey,FFieldBlock> FieldBlocks;
        std::map<FBlockKey,FNormalBlock> NormalBlocks;
        std::map<FBlockKey,FGradientFallbackBlock> GradientFallbackBlocks;
        std::map<FBlockKey,FCurvedEdgeRootBlock> CurvedEdgeRootBlocks;
        struct FDetailDisplacementBlock
        {uint64 Vertices=0,HighVertices=0;double Maximum=0.0,MaximumGradient=0.0;std::array<double,3> Position{};};
        std::map<std::array<int32,3>,FDetailDisplacementBlock> DetailDisplacementBlocks;
        std::map<FBlockKey,std::array<double,3>> FirstPositionByBlock;
        std::map<FBlockKey,std::array<double,3>> FirstNormalByBlock;
        uint64 RayTileResidualSamples=0,RayTileResidualGt01=0,
            RayTileResidualGt025=0,RayTileResidualGt05=0;
        double RayTileResidualSum=0.0,RayTileResidualMaximum=0.0;
        std::array<double,3> RayTileResidualMaximumPosition{};
        uint64 RayTileInteriorSamples=0,RayTileInteriorGt0125=0,
            RayTileInteriorGt025=0,RayTileInteriorGt05=0,RayTileInteriorGt1=0;
        double RayTileInteriorResidualSum=0.0,RayTileInteriorResidualMaximum=0.0;
        std::array<double,3> RayTileInteriorMaximumPosition{};
        for(size_t I=0;I<SurfaceVertices;++I)
        {
            const auto& Position=Tile.Vertices[I].Position;
            const double P[3]={Position[0]*4.0,Position[1]*4.0,Position[2]*4.0};
            double DetailGradientAtVertex=0.0;
            if(Tile.Key.Level==0)
            {
                const VoxelForgeCore::DetailFieldSample DetailSample=Shape.detailSampleAt(
                    {P[0],P[1],P[2]});
                if(DetailSample.caveInfluence>=0.60f&&DetailSample.floorInfluence>=0.60f)
                {
                    constexpr double GradientProbeOffset=0.5;
                    double DetailGradientSquared=0.0;
                    for(int32 Axis=0;Axis<3;++Axis)
                    {
                        double Minus[3]={P[0],P[1],P[2]};
                        double Plus[3]={P[0],P[1],P[2]};
                        Minus[Axis]-=GradientProbeOffset;
                        Plus[Axis]+=GradientProbeOffset;
                        const double DMinus=Shape.detailSampleAt(
                            {Minus[0],Minus[1],Minus[2]}).displacement;
                        const double DPlus=Shape.detailSampleAt(
                            {Plus[0],Plus[1],Plus[2]}).displacement;
                        const double Gradient=(DPlus-DMinus)/(2.0*GradientProbeOffset);
                        DetailGradientSquared+=Gradient*Gradient;
                    }
                    const double DetailGradient=std::sqrt(DetailGradientSquared);
                    DetailGradientAtVertex=DetailGradient;
                    ++Result.DetailGradientSamples;
                    Result.MaximumDetailGradient=std::max(
                        Result.MaximumDetailGradient,DetailGradient);
                    if(DetailGradient>0.55)++Result.DetailGradientOverLimit;
                    const double DetailDisplacement=std::abs(DetailSample.displacement);
                    ++Result.DetailDisplacementSamples;
                    Result.MaximumDetailDisplacement=std::max(
                        Result.MaximumDetailDisplacement,DetailDisplacement);
                    const std::array<int32,3> DetailKey{
                        static_cast<int32>(std::floor((P[0]-Origin.X)/8.0)),
                        static_cast<int32>(std::floor((P[1]-Origin.Y)/8.0)),
                        static_cast<int32>(std::floor((P[2]-Origin.Z)/8.0))};
                    FDetailDisplacementBlock& DetailBlock=DetailDisplacementBlocks[DetailKey];
                    ++DetailBlock.Vertices;
                    DetailBlock.MaximumGradient=std::max(DetailBlock.MaximumGradient,DetailGradientAtVertex);
                    if(DetailDisplacement>0.65)
                    {
                        ++DetailBlock.HighVertices;
                        if(DetailDisplacement>DetailBlock.Maximum)
                        {
                            DetailBlock.Maximum=DetailDisplacement;
                            DetailBlock.Position={P[0],P[1],P[2]};
                        }
                        ++Result.HighDetailDisplacementSamples;
                    }
                }
            }
            double Grid[3];int32 VariableAxis=-1;
            for(int32 Axis=0;Axis<3;++Axis)
            {
                Grid[Axis]=(P[Axis]-(&Origin.X)[Axis])/Step;
                if(std::abs(Grid[Axis]-std::round(Grid[Axis]))>0.003)
                {
                    if(VariableAxis>=0){VariableAxis=-2;break;}
                    VariableAxis=Axis;
                }
            }
            if(VariableAxis<0)continue;
            double Fraction=Grid[VariableAxis]-std::floor(Grid[VariableAxis]);
            if(Fraction<0.003||Fraction>0.997)continue;
            ++Result.EdgeVertices;
            const int32 CoarseBin=std::clamp(static_cast<int32>(Fraction*16.0),0,15);
            const int32 PreciseBin=std::clamp(static_cast<int32>(std::lround(Fraction*256.0)),0,255);
            ++Result.OffsetHistogram[CoarseBin];
            if(CoarseBin==7||CoarseBin==8)++Result.MidpointVertices;
            double EdgeA[3],EdgeB[3];
            for(int32 Axis=0;Axis<3;++Axis)
            {
                const int32 Cell=static_cast<int32>(std::floor(Grid[Axis]));
                EdgeA[Axis]=(&Origin.X)[Axis]+Cell*Step;
                EdgeB[Axis]=EdgeA[Axis];
            }
            EdgeB[VariableAxis]+=Step;
            const double D0=Shape.densityAt({EdgeA[0],EdgeA[1],EdgeA[2]});
            const double D1=Shape.densityAt({EdgeB[0],EdgeB[1],EdgeB[2]});
            const double ExactD0=Shape.unclampedDensityAt(
                {EdgeA[0]+0.013,EdgeA[1]+0.013,EdgeA[2]+0.013});
            const double ExactD1=Shape.unclampedDensityAt(
                {EdgeB[0]+0.013,EdgeB[1]+0.013,EdgeB[2]+0.013});
            const double ExactDenominator=ExactD1-ExactD0;
            ++Result.FieldEdgeChecks;
            if(std::abs(D0)>127.0/256.0||std::abs(D1)>127.0/256.0)
                ++Result.OldInt8RangeEdges;
            const bool bExactCrossing=(ExactD0<0.0&&ExactD1>0.0)
                ||(ExactD0>0.0&&ExactD1<0.0);
            double AnalyticShift=0.0;
            if(bExactCrossing&&std::abs(ExactDenominator)>1.0e-9)
            {
                const double ExactT=std::clamp(-ExactD0/ExactDenominator,0.0,1.0);
                AnalyticShift=std::abs(Fraction-ExactT)*Step;
                Result.SumAnalyticEdgeShiftVoxels+=AnalyticShift;
                if(AnalyticShift>Result.MaximumAnalyticEdgeShiftVoxels)
                {
                    Result.MaximumAnalyticEdgeShiftVoxels=AnalyticShift;
                    Result.WorstShiftPosition={P[0],P[1],P[2]};
                    Result.WorstShiftFraction=Fraction;Result.WorstShiftD0=D0;
                    Result.WorstShiftD1=D1;Result.WorstShiftExactD0=ExactD0;
                    Result.WorstShiftExactD1=ExactD1;Result.WorstShiftAxis=VariableAxis;
                }
            }
            else ++Result.AnalyticCrossingMisses;
            if(Tile.Key.Level==0&&Step==1&&bExactCrossing
                &&Fraction>=0.375&&Fraction<=0.625)
            {
                const double MidExact=Shape.unclampedDensityAt({
                    (EdgeA[0]+EdgeB[0])*0.5+0.013,
                    (EdgeA[1]+EdgeB[1])*0.5+0.013,
                    (EdgeA[2]+EdgeB[2])*0.5+0.013});
                double CurveRootT=0.5;
                if(std::abs(MidExact)>1.0e-8)
                {
                    if((MidExact<0.0)==(ExactD0<0.0))
                    {
                        const double Denominator=ExactD1-MidExact;
                        if(std::abs(Denominator)>1.0e-9)
                            CurveRootT=0.5+0.5*std::clamp(-MidExact/Denominator,0.0,1.0);
                    }
                    else
                    {
                        const double Denominator=MidExact-ExactD0;
                        if(std::abs(Denominator)>1.0e-9)
                            CurveRootT=0.5*std::clamp(-ExactD0/Denominator,0.0,1.0);
                    }
                }
                const double CurveError=std::abs(Fraction-CurveRootT);
                ++Result.CurvedEdgeChecks;
                Result.MaximumCurvedEdgeRootError=std::max(
                    Result.MaximumCurvedEdgeRootError,CurveError);
                FBlockKey RootKey{};RootKey[0]=VariableAxis;
                for(int32 Axis=0;Axis<3;++Axis)
                    RootKey[Axis+1]=static_cast<int32>(std::floor(
                        (P[Axis]-(&Origin.X)[Axis])/8.0));
                FCurvedEdgeRootBlock& RootBlock=CurvedEdgeRootBlocks[RootKey];
                ++RootBlock.Vertices;
                if(CurveError>0.08)
                {
                    ++RootBlock.BadVertices;
                    if(CurveError>RootBlock.MaximumError)
                    {
                        RootBlock.MaximumError=CurveError;
                        RootBlock.Position={P[0],P[1],P[2]};
                    }
                    ++Result.CurvedEdgeRootBadVertices;
                }
            }
            const double SourceResidual=std::abs(Shape.unclampedDensityAt({P[0],P[1],P[2]}));
            if(Tile.Key.Level==0&&Tile.Key.Coord.X==1
                &&Tile.Key.Coord.Y==-16&&Tile.Key.Coord.Z==-3)
            {
                ++RayTileResidualSamples;
                RayTileResidualSum+=SourceResidual;
                if(SourceResidual>0.1)++RayTileResidualGt01;
                if(SourceResidual>0.25)++RayTileResidualGt025;
                if(SourceResidual>0.5)++RayTileResidualGt05;
                if(SourceResidual>RayTileResidualMaximum)
                {
                    RayTileResidualMaximum=SourceResidual;
                    RayTileResidualMaximumPosition={P[0],P[1],P[2]};
                }
            }
            Result.SumAbsSourceResidual+=SourceResidual;
            if(SourceResidual>Result.MaximumAbsSourceResidual)
            {
                Result.MaximumAbsSourceResidual=SourceResidual;
                Result.WorstResidualPosition={P[0],P[1],P[2]};
            }
            const double H=0.5;
            const double GX=Shape.unclampedDensityAt({P[0]+H+0.013,P[1]+0.013,P[2]+0.013})
                -Shape.unclampedDensityAt({P[0]-H+0.013,P[1]+0.013,P[2]+0.013});
            const double GY=Shape.unclampedDensityAt({P[0]+0.013,P[1]+H+0.013,P[2]+0.013})
                -Shape.unclampedDensityAt({P[0]+0.013,P[1]-H+0.013,P[2]+0.013});
            const double GZ=Shape.unclampedDensityAt({P[0]+0.013,P[1]+0.013,P[2]+H+0.013})
                -Shape.unclampedDensityAt({P[0]+0.013,P[1]+0.013,P[2]-H+0.013});
            const double GL=std::sqrt(GX*GX+GY*GY+GZ*GZ);
            double NormalAgreement=1.0;
            if(Tile.Key.Level==0&&GL>1.0e-8)
            {
                const auto& N=Tile.Vertices[I].Normal;
                NormalAgreement=std::clamp((N[0]*GX+N[1]*GY+N[2]*GZ)/GL,-1.0,1.0);
                ++Result.NormalChecks;Result.SumNormalAgreement+=NormalAgreement;
                Result.MinimumNormalAgreement=std::min(Result.MinimumNormalAgreement,NormalAgreement);
            }
            FBlockKey Key{};Key[0]=VariableAxis;
            for(int32 Axis=0;Axis<3;++Axis)
                Key[Axis+1]=static_cast<int32>(std::floor((P[Axis]-(&Origin.X)[Axis])/8.0));
            if(Tile.Key.Level==0&&GL>1.0e-8)
            {
                FNormalBlock& NormalBlock=NormalBlocks[Key];
                ++NormalBlock.Vertices;
                if(NormalAgreement<NormalBlock.MinimumAgreement)
                {
                    NormalBlock.MinimumAgreement=NormalAgreement;
                    NormalBlock.Position={P[0],P[1],P[2]};
                    NormalBlock.MeshNormal={Tile.Vertices[I].Normal[0],
                        Tile.Vertices[I].Normal[1],Tile.Vertices[I].Normal[2]};
                }
                if(NormalAgreement<0.90)++NormalBlock.BadVertices;
            }
            if(Tile.Key.Level==0)
            {
                FFieldBlock& FieldBlock=FieldBlocks[Key];
                ++FieldBlock.Vertices;
                FieldBlock.MaximumShift=std::max(FieldBlock.MaximumShift,AnalyticShift);
                FieldBlock.MaximumResidual=std::max(FieldBlock.MaximumResidual,SourceResidual);
                const bool bBadFieldEdge=SourceResidual>0.5;
                if(bBadFieldEdge)
                {
                    ++FieldBlock.BadVertices;
                    if(FieldBlock.BadVertices==1)FieldBlock.Position={P[0],P[1],P[2]};
                }

                FGradientFallbackBlock& FallbackBlock=GradientFallbackBlocks[Key];
                ++FallbackBlock.Vertices;
                uint64 VertexFallbackSamples=0;
                for(int32 Endpoint=0;Endpoint<2;++Endpoint)
                {
                    const double* EdgePoint=Endpoint?EdgeB:EdgeA;
                    for(int32 Axis=0;Axis<3;++Axis)
                    for(int32 Side=-1;Side<=1;Side+=2)
                    {
                        double Probe[3]={EdgePoint[0],EdgePoint[1],EdgePoint[2]};
                        Probe[Axis]+=Side;
                        const double Packed=Shape.densityAt({Probe[0],Probe[1],Probe[2]});
                        if(std::abs(std::abs(Packed)-4.125)>0.001)continue;
                        double ExactProbe[3]={Probe[0]+0.013,Probe[1]+0.013,Probe[2]+0.013};
                        const double Exact=Shape.unclampedDensityAt(
                            {ExactProbe[0],ExactProbe[1],ExactProbe[2]});
                        const double Error=std::abs(Packed-Exact);
                        if(Error<=0.25)continue;
                        ++VertexFallbackSamples;
                        FallbackBlock.MaximumError=std::max(FallbackBlock.MaximumError,Error);
                    }
                }
                if(VertexFallbackSamples>0&&NormalAgreement<0.98)
                {
                    ++FallbackBlock.BadVertices;
                    FallbackBlock.BadSamples+=VertexFallbackSamples;
                    if(FallbackBlock.BadVertices==1)FallbackBlock.Position={P[0],P[1],P[2]};
                }
            }
            FirstPositionByBlock.try_emplace(Key,std::array<double,3>{P[0],P[1],P[2]});
            const auto& Normal=Tile.Vertices[I].Normal;
            FirstNormalByBlock.try_emplace(Key,std::array<double,3>{Normal[0],Normal[1],Normal[2]});
            ++OffsetBlocks[Key][PreciseBin];
        }

        struct FHighDetailBlock
        {std::array<int32,3> Key{};uint64 Vertices=0;double Maximum=0.0,MaximumGradient=0.0;
            std::array<double,3> Position{};};
        std::vector<FHighDetailBlock> HighDetailBlocks;
        for(const auto& [Key,Block]:DetailDisplacementBlocks)
        {
            if(Block.Vertices<12||Block.HighVertices<6
                ||static_cast<double>(Block.HighVertices)/Block.Vertices<0.25)continue;
            HighDetailBlocks.push_back({Key,Block.HighVertices,Block.Maximum,Block.MaximumGradient,Block.Position});
        }
        std::vector<uint8> HighDetailVisited(HighDetailBlocks.size(),0);
        for(size_t I=0;I<HighDetailBlocks.size();++I)
        {
            if(HighDetailVisited[I])continue;
            std::vector<size_t> Pending{I};HighDetailVisited[I]=1;
            uint64 BlockCount=0,VertexCountInPatch=0;double Maximum=0.0,MaximumGradient=0.0;
            size_t Representative=I;
            while(!Pending.empty())
            {
                const size_t Current=Pending.back();Pending.pop_back();
                const FHighDetailBlock& A=HighDetailBlocks[Current];
                ++BlockCount;VertexCountInPatch+=A.Vertices;
                MaximumGradient=std::max(MaximumGradient,A.MaximumGradient);
                if(A.Maximum>Maximum){Maximum=A.Maximum;Representative=Current;}
                for(size_t J=0;J<HighDetailBlocks.size();++J)
                {
                    if(HighDetailVisited[J])continue;
                    int32 Manhattan=0;
                    for(int32 Axis=0;Axis<3;++Axis)
                        Manhattan+=std::abs(A.Key[Axis]-HighDetailBlocks[J].Key[Axis]);
                    if(Manhattan==1){HighDetailVisited[J]=1;Pending.push_back(J);}
                }
            }
            // Large relief is valid when it changes smoothly over multiple cells. Flag
            // only connected high-amplitude patches whose measured slope is alias-prone.
            if(BlockCount<3||VertexCountInPatch<48||MaximumGradient<=0.55)continue;
            ++Result.DetailDisplacementPatches;
            if(bPrintPatches&&Result.DetailDisplacementPatches<=8)
            {
                const auto& Sample=HighDetailBlocks[Representative];
                std::printf("[MeshDetailDisplacementPatch] level=0 tile=(%d,%d,%d) connected_8vox_blocks=%llu vertices=%llu max_displacement=%.3f vox max_gradient=%.4f vox_per_vox sample=(%.3f,%.3f,%.3f)\n",
                    Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(BlockCount),
                    static_cast<unsigned long long>(VertexCountInPatch),Maximum,MaximumGradient,
                    Sample.Position[0],Sample.Position[1],Sample.Position[2]);
            }
        }

        struct FBadFieldBlock
        {FBlockKey Key{};uint64 Vertices=0,BadVertices=0;double Share=0.0,MaximumShift=0.0,
            MaximumResidual=0.0;std::array<double,3> Position{};};
        std::vector<FBadFieldBlock> BadFieldBlocks;
        for(const auto& [Key,Block]:FieldBlocks)
        {
            if(Block.Vertices<12||Block.BadVertices<6)continue;
            const double Share=static_cast<double>(Block.BadVertices)/Block.Vertices;
            if(Share<0.25)continue;
            BadFieldBlocks.push_back({Key,Block.Vertices,Block.BadVertices,Share,
                Block.MaximumShift,Block.MaximumResidual,Block.Position});
        }
        std::vector<uint8> BadFieldVisited(BadFieldBlocks.size(),0);
        for(size_t I=0;I<BadFieldBlocks.size();++I)
        {
            if(BadFieldVisited[I])continue;
            std::vector<size_t> Pending{I};BadFieldVisited[I]=1;
            uint64 BlockCount=0,BadVertexCount=0;double MaxShift=0.0,MaxResidual=0.0;
            size_t Representative=I;
            while(!Pending.empty())
            {
                const size_t Current=Pending.back();Pending.pop_back();
                const FBadFieldBlock& A=BadFieldBlocks[Current];
                ++BlockCount;BadVertexCount+=A.BadVertices;
                MaxShift=std::max(MaxShift,A.MaximumShift);
                MaxResidual=std::max(MaxResidual,A.MaximumResidual);
                if(A.MaximumShift>=MaxShift||A.MaximumResidual>=MaxResidual)Representative=Current;
                for(size_t J=0;J<BadFieldBlocks.size();++J)
                {
                    if(BadFieldVisited[J]||A.Key[0]!=BadFieldBlocks[J].Key[0])continue;
                    int32 Manhattan=0;
                    for(int32 Axis=1;Axis<4;++Axis)
                        Manhattan+=std::abs(A.Key[Axis]-BadFieldBlocks[J].Key[Axis]);
                    if(Manhattan==1){BadFieldVisited[J]=1;Pending.push_back(J);}
                }
            }
            if(BlockCount<2||BadVertexCount<24)continue;
            ++Result.FieldMismatchPatches;
            if(bPrintPatches&&Result.FieldMismatchPatches<=8)
            {
                const FBadFieldBlock& Sample=BadFieldBlocks[Representative];
                std::printf("[MeshFieldMismatchPatch] level=%d tile=(%d,%d,%d) connected_8vox_blocks=%llu bad_edges=%llu max_analytic_shift=%.3f vox max_source_residual=%.3f vox sample=(%.3f,%.3f,%.3f)\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(BlockCount),
                    static_cast<unsigned long long>(BadVertexCount),MaxShift,MaxResidual,
                    Sample.Position[0],Sample.Position[1],Sample.Position[2]);
            }
        }

        struct FBadGradientFallbackBlock
        {FBlockKey Key{};uint64 Vertices=0,BadVertices=0,BadSamples=0;
            double MaximumError=0.0;std::array<double,3> Position{};};
        std::vector<FBadGradientFallbackBlock> BadGradientFallbackBlocks;
        for(const auto& [Key,Block]:GradientFallbackBlocks)
        {
            if(Block.Vertices<12||Block.BadVertices<6)continue;
            if(static_cast<double>(Block.BadVertices)/Block.Vertices<0.25)continue;
            BadGradientFallbackBlocks.push_back({Key,Block.Vertices,Block.BadVertices,
                Block.BadSamples,Block.MaximumError,Block.Position});
        }
        std::vector<uint8> GradientFallbackVisited(BadGradientFallbackBlocks.size(),0);
        for(size_t I=0;I<BadGradientFallbackBlocks.size();++I)
        {
            if(GradientFallbackVisited[I])continue;
            std::vector<size_t> Pending{I};GradientFallbackVisited[I]=1;
            uint64 BlockCount=0,BadVertexCount=0,BadSamples=0;double MaximumError=0.0;
            size_t Representative=I;
            while(!Pending.empty())
            {
                const size_t Current=Pending.back();Pending.pop_back();
                const FBadGradientFallbackBlock& A=BadGradientFallbackBlocks[Current];
                ++BlockCount;BadVertexCount+=A.BadVertices;BadSamples+=A.BadSamples;
                MaximumError=std::max(MaximumError,A.MaximumError);
                if(A.MaximumError>=MaximumError)Representative=Current;
                for(size_t J=0;J<BadGradientFallbackBlocks.size();++J)
                {
                    if(GradientFallbackVisited[J]
                        ||A.Key[0]!=BadGradientFallbackBlocks[J].Key[0])continue;
                    int32 Manhattan=0;
                    for(int32 Axis=1;Axis<4;++Axis)
                        Manhattan+=std::abs(A.Key[Axis]-BadGradientFallbackBlocks[J].Key[Axis]);
                    if(Manhattan==1)
                    {GradientFallbackVisited[J]=1;Pending.push_back(J);}
                }
            }
            if(BlockCount<2||BadVertexCount<24)continue;
            ++Result.GradientFallbackPatches;Result.GradientFallbackSamples+=BadSamples;
            if(bPrintPatches&&Result.GradientFallbackPatches<=8)
            {
                const auto& Sample=BadGradientFallbackBlocks[Representative];
                std::printf("[MeshGradientFallbackPatch] level=%d tile=(%d,%d,%d) connected_8vox_blocks=%llu vertices=%llu fallback_gradient_samples=%llu max_distance_error=%.3f vox sample=(%.3f,%.3f,%.3f)\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(BlockCount),
                    static_cast<unsigned long long>(BadVertexCount),
                    static_cast<unsigned long long>(BadSamples),MaximumError,
                    Sample.Position[0],Sample.Position[1],Sample.Position[2]);
            }
        }

        struct FBadNormalBlock
        {FBlockKey Key{};uint64 Vertices=0,BadVertices=0;double MinimumAgreement=1.0;
            std::array<double,3> Position{},MeshNormal{};};
        std::vector<FBadNormalBlock> BadNormalBlocks;
        for(const auto& [Key,Block]:NormalBlocks)
        {
            if(Block.Vertices<12||Block.BadVertices<6)continue;
            if(static_cast<double>(Block.BadVertices)/Block.Vertices<0.25)continue;
            BadNormalBlocks.push_back({Key,Block.Vertices,Block.BadVertices,
                Block.MinimumAgreement,Block.Position,Block.MeshNormal});
        }
        std::vector<uint8> BadNormalVisited(BadNormalBlocks.size(),0);
        for(size_t I=0;I<BadNormalBlocks.size();++I)
        {
            if(BadNormalVisited[I])continue;
            std::vector<size_t> Pending{I};BadNormalVisited[I]=1;
            uint64 BlockCount=0,BadVertexCount=0;double MinAgreement=1.0;size_t Representative=I;
            while(!Pending.empty())
            {
                const size_t Current=Pending.back();Pending.pop_back();
                const FBadNormalBlock& A=BadNormalBlocks[Current];
                ++BlockCount;BadVertexCount+=A.BadVertices;
                MinAgreement=std::min(MinAgreement,A.MinimumAgreement);
                if(A.MinimumAgreement<BadNormalBlocks[Representative].MinimumAgreement)
                    Representative=Current;
                for(size_t J=0;J<BadNormalBlocks.size();++J)
                {
                    if(BadNormalVisited[J]||A.Key[0]!=BadNormalBlocks[J].Key[0])continue;
                    int32 Manhattan=0;
                    for(int32 Axis=1;Axis<4;++Axis)
                        Manhattan+=std::abs(A.Key[Axis]-BadNormalBlocks[J].Key[Axis]);
                    if(Manhattan==1){BadNormalVisited[J]=1;Pending.push_back(J);}
                }
            }
            if(BlockCount<2||BadVertexCount<24)continue;
            ++Result.NormalMismatchPatches;
            if(bPrintPatches&&Result.NormalMismatchPatches<=8)
            {
                const auto& Sample=BadNormalBlocks[Representative];
                std::printf("[MeshNormalMismatchPatch] level=%d tile=(%d,%d,%d) connected_8vox_blocks=%llu bad_normals=%llu min_field_normal_dot=%.3f sample=(%.3f,%.3f,%.3f)\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(BlockCount),
                    static_cast<unsigned long long>(BadVertexCount),MinAgreement,
                    Sample.Position[0],Sample.Position[1],Sample.Position[2]);
                double EdgeA[3],EdgeB[3];
                for(int32 Axis=0;Axis<3;++Axis)
                {
                    EdgeA[Axis]=(&Origin.X)[Axis]
                        +std::floor((Sample.Position[Axis]-(&Origin.X)[Axis])/Step)*Step;
                    EdgeB[Axis]=EdgeA[Axis];
                }
                EdgeB[Sample.Key[0]]+=Step;
                double GradPacked[2][3]{},GradExact[2][3]{};
                for(int32 Endpoint=0;Endpoint<2;++Endpoint)
                {
                    const double* P=Endpoint?EdgeB:EdgeA;
                    for(int32 Axis=0;Axis<3;++Axis)
                    {
                        double Minus[3]={P[0],P[1],P[2]},Plus[3]={P[0],P[1],P[2]};
                        Minus[Axis]-=1.0;Plus[Axis]+=1.0;
                        GradPacked[Endpoint][Axis]=Shape.densityAt({Plus[0],Plus[1],Plus[2]})
                            -Shape.densityAt({Minus[0],Minus[1],Minus[2]});
                        for(int32 A=0;A<3;++A){Minus[A]+=0.013;Plus[A]+=0.013;}
                        GradExact[Endpoint][Axis]=Shape.unclampedDensityAt(
                            {Plus[0],Plus[1],Plus[2]})
                            -Shape.unclampedDensityAt({Minus[0],Minus[1],Minus[2]});
                    }
                }
                const double T=std::clamp(Sample.Position[Sample.Key[0]]-EdgeA[Sample.Key[0]],0.0,1.0);
                double PackedInterpolated[3]{},ExactInterpolated[3]{};
                for(int32 Axis=0;Axis<3;++Axis)
                {
                    PackedInterpolated[Axis]=GradPacked[0][Axis]+T*(GradPacked[1][Axis]-GradPacked[0][Axis]);
                    ExactInterpolated[Axis]=GradExact[0][Axis]+T*(GradExact[1][Axis]-GradExact[0][Axis]);
                }
                const auto NormalDot=[](const double A[3],const double B[3])
                {
                    const double AL=std::sqrt(A[0]*A[0]+A[1]*A[1]+A[2]*A[2]);
                    const double BL=std::sqrt(B[0]*B[0]+B[1]*B[1]+B[2]*B[2]);
                    return AL>1.0e-10&&BL>1.0e-10
                        ?(A[0]*B[0]+A[1]*B[1]+A[2]*B[2])/(AL*BL):0.0;
                };
                std::printf("[MeshNormalProbe] t=%.3f edge_axis=%d mesh=(%.3f,%.3f,%.3f) packed_grid=(%.3f,%.3f,%.3f) exact_grid=(%.3f,%.3f,%.3f) packed_dot=%.3f exact_dot=%.3f endpoints=(%.0f,%.0f,%.0f)->(%.0f,%.0f,%.0f)\n",
                    T,Sample.Key[0],Sample.MeshNormal[0],Sample.MeshNormal[1],Sample.MeshNormal[2],
                    PackedInterpolated[0],PackedInterpolated[1],PackedInterpolated[2],
                    ExactInterpolated[0],ExactInterpolated[1],ExactInterpolated[2],
                    NormalDot(Sample.MeshNormal.data(),PackedInterpolated),
                    NormalDot(Sample.MeshNormal.data(),ExactInterpolated),
                    EdgeA[0],EdgeA[1],EdgeA[2],EdgeB[0],EdgeB[1],EdgeB[2]);
                for(int32 Endpoint=0;Endpoint<2;++Endpoint)
                for(int32 Axis=0;Axis<3;++Axis)
                {
                    const double* P=Endpoint?EdgeB:EdgeA;
                    for(int32 Side=-1;Side<=1;Side+=2)
                    {
                        double Q[3]={P[0],P[1],P[2]};Q[Axis]+=Side;
                        const double Packed=Shape.densityAt({Q[0],Q[1],Q[2]});
                        Q[0]+=0.013;Q[1]+=0.013;Q[2]+=0.013;
                        const double Exact=Shape.unclampedDensityAt({Q[0],Q[1],Q[2]});
                        if(std::abs(Packed-Exact)>0.02)
                            std::printf("[MeshNormalProbeNode] endpoint=%d axis=%d side=%d packed=%.4f exact=%.4f delta=%.4f\n",
                                Endpoint,Axis,Side,Packed,Exact,Packed-Exact);
                    }
                }
            }
        }

        if(Tile.Key.Level==0&&Tile.Key.Coord.X==1
            &&Tile.Key.Coord.Y==-16&&Tile.Key.Coord.Z==-3)
        {
            struct FConcentratedBlock
            {FBlockKey Key{};uint64 Count=0;double TopOne=0.0,TopThree=0.0;
                std::array<int32,3> Bins{};std::array<double,3> Position{};};
            std::vector<FConcentratedBlock> ConcentratedBlocks;
            for(const auto& [Key,Bins]:OffsetBlocks)
            {
                const uint64 Count=std::accumulate(Bins.begin(),Bins.end(),uint64{0});
                if(Count<12)continue;
                std::array<int32,256> Order{};
                std::iota(Order.begin(),Order.end(),0);
                std::partial_sort(Order.begin(),Order.begin()+3,Order.end(),
                    [&](int32 A,int32 B){return Bins[A]>Bins[B];});
                const uint64 TopThree=Bins[Order[0]]+Bins[Order[1]]+Bins[Order[2]];
                ConcentratedBlocks.push_back({Key,Count,
                    static_cast<double>(Bins[Order[0]])/Count,
                    static_cast<double>(TopThree)/Count,
                    {Order[0],Order[1],Order[2]},FirstPositionByBlock[Key]});
            }
            std::sort(ConcentratedBlocks.begin(),ConcentratedBlocks.end(),
                [](const auto& A,const auto& B){return A.TopThree>B.TopThree;});
            for(size_t I=0;I<std::min<size_t>(8,ConcentratedBlocks.size());++I)
            {
                const auto& B=ConcentratedBlocks[I];
                std::printf("[MeshOffsetConcentration] tile=(1,-16,-3) axis=%d block=(%d,%d,%d) vertices=%llu top1=%.3f top3=%.3f bins=(%d,%d,%d) sample=(%.3f,%.3f,%.3f)\n",
                    B.Key[0],B.Key[1],B.Key[2],B.Key[3],
                    static_cast<unsigned long long>(B.Count),B.TopOne,B.TopThree,
                    B.Bins[0],B.Bins[1],B.Bins[2],
                    B.Position[0],B.Position[1],B.Position[2]);
            }
        }

        struct FRepeatBlock
        {FBlockKey Key{};int32 Bin=0;uint64 Count=0;double Share=0.0;
            std::array<double,3> Position{},Normal{};};
        std::vector<FRepeatBlock> RepeatBlocks;
        for(const auto& [Key,Bins]:OffsetBlocks)
        {
            const uint64 Count=std::accumulate(Bins.begin(),Bins.end(),uint64{0});
            if(Count<12)continue;
            const auto Best=std::max_element(Bins.begin(),Bins.end());
            const int32 Bin=static_cast<int32>(Best-Bins.begin());
            const double Share=static_cast<double>(*Best)/Count;
            if(Bin<80||Bin>176||Share<0.75)continue;
            RepeatBlocks.push_back({Key,Bin,Count,Share,FirstPositionByBlock[Key],FirstNormalByBlock[Key]});
        }
        std::vector<uint8> Visited(RepeatBlocks.size(),0);
        for(size_t I=0;I<RepeatBlocks.size();++I)
        {
            if(Visited[I])continue;
            std::vector<size_t> Pending{I};Visited[I]=1;
            uint64 BlockCount=0,VertexCountInPatch=0;double Dominance=0.0;
            while(!Pending.empty())
            {
                const size_t Current=Pending.back();Pending.pop_back();
                const FRepeatBlock& A=RepeatBlocks[Current];
                ++BlockCount;VertexCountInPatch+=A.Count;Dominance=std::max(Dominance,A.Share);
                for(size_t J=0;J<RepeatBlocks.size();++J)
                {
                    if(Visited[J]||A.Key[0]!=RepeatBlocks[J].Key[0]
                        ||std::abs(A.Bin-RepeatBlocks[J].Bin)>1)continue;
                    int32 Manhattan=0;
                    for(int32 Axis=1;Axis<4;++Axis)
                        Manhattan+=std::abs(A.Key[Axis]-RepeatBlocks[J].Key[Axis]);
                    if(Manhattan==1){Visited[J]=1;Pending.push_back(J);}
                }
            }
            if(BlockCount<3||VertexCountInPatch<48)continue;
            ++Result.RepeatedOffsetPatches;
            Result.MaximumPatchDominance=std::max(Result.MaximumPatchDominance,Dominance);
            if(bPrintPatches&&Result.RepeatedOffsetPatches<=8)
            {
                const auto& P=RepeatBlocks[I].Position;const int32 Axis=RepeatBlocks[I].Key[0];
                double EdgeA[3],EdgeB[3];
                for(int32 A=0;A<3;++A)
                {
                    const int32 Cell=static_cast<int32>(std::floor((P[A]-(&Origin.X)[A])/Step));
                    EdgeA[A]=(&Origin.X)[A]+Cell*Step;EdgeB[A]=EdgeA[A];
                }
                EdgeB[Axis]+=Step;
                const double D0=Shape.densityAt({EdgeA[0],EdgeA[1],EdgeA[2]});
                const double D1=Shape.densityAt({EdgeB[0],EdgeB[1],EdgeB[2]});
                const double ExactD0=Shape.densityAt({EdgeA[0]+0.013,EdgeA[1]+0.013,EdgeA[2]+0.013});
                const double ExactD1=Shape.densityAt({EdgeB[0]+0.013,EdgeB[1]+0.013,EdgeB[2]+0.013});
                const double InterpolatedT=std::abs(D1-D0)>1.0e-9
                    ?std::clamp(-D0/(D1-D0),0.0,1.0):0.5;
                const double SourceDensity=Shape.densityAt({P[0],P[1],P[2]});
                const auto& N=RepeatBlocks[I].Normal;
                const double PlusNormal=Shape.densityAt({P[0]+N[0],P[1]+N[1],P[2]+N[2]});
                const double MinusNormal=Shape.densityAt({P[0]-N[0],P[1]-N[1],P[2]-N[2]});
                std::printf("[MeshLatticePatch] level=%d tile=(%d,%d,%d) connected_8vox_blocks=%llu edge_vertices=%llu dominant_share=%.3f midpoint_bin_cluster=1 axis=%d sample=(%.3f,%.3f,%.3f) offset=%.4f normal=(%.3f,%.3f,%.3f) endpoint_density=(%.4f,%.4f) analytic_probe=(%.4f,%.4f) predicted_t=%.4f source_density=%.4f normal_probe=(plus1=%.4f,minus1=%.4f)\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(BlockCount),
                    static_cast<unsigned long long>(VertexCountInPatch),Dominance,Axis,
                    P[0],P[1],P[2],(P[Axis]-EdgeA[Axis])/Step,N[0],N[1],N[2],D0,D1,
                    ExactD0,ExactD1,InterpolatedT,SourceDensity,PlusNormal,MinusNormal);
            }
        }

        const size_t TriangleCount=Tile.Indices.size()/3;
        for(size_t T=0;T<TriangleCount;++T)
        {
            const uint32 A=Tile.Indices[T*3],B=Tile.Indices[T*3+1],C=Tile.Indices[T*3+2];
            if(A>=VertexCount||B>=VertexCount||C>=VertexCount)continue;
            if(A<SurfaceVertices&&B<SurfaceVertices&&C<SurfaceVertices)
            {
                if(Tile.Key.Level==0&&Tile.Key.Coord.X==1
                    &&Tile.Key.Coord.Y==-16&&Tile.Key.Coord.Z==-3)
                {
                    const auto& PA=Tile.Vertices[A].Position;
                    const auto& PB=Tile.Vertices[B].Position;
                    const auto& PC=Tile.Vertices[C].Position;
                    const double V[3][3]={{PA[0]*4.0,PA[1]*4.0,PA[2]*4.0},
                        {PB[0]*4.0,PB[1]*4.0,PB[2]*4.0},
                        {PC[0]*4.0,PC[1]*4.0,PC[2]*4.0}};
                    const double Probe[4][3]={
                        {(V[0][0]+V[1][0]+V[2][0])/3.0,
                         (V[0][1]+V[1][1]+V[2][1])/3.0,
                         (V[0][2]+V[1][2]+V[2][2])/3.0},
                        {(V[0][0]+V[1][0])/2.0,(V[0][1]+V[1][1])/2.0,(V[0][2]+V[1][2])/2.0},
                        {(V[1][0]+V[2][0])/2.0,(V[1][1]+V[2][1])/2.0,(V[1][2]+V[2][2])/2.0},
                        {(V[2][0]+V[0][0])/2.0,(V[2][1]+V[0][1])/2.0,(V[2][2]+V[0][2])/2.0}};
                    for(const auto& Sample:Probe)
                    {
                        const double Residual=std::abs(Shape.unclampedDensityAt(
                            {Sample[0]+0.013,Sample[1]+0.013,Sample[2]+0.013}));
                        ++RayTileInteriorSamples;RayTileInteriorResidualSum+=Residual;
                        if(Residual>0.125)++RayTileInteriorGt0125;
                        if(Residual>0.25)++RayTileInteriorGt025;
                        if(Residual>0.5)++RayTileInteriorGt05;
                        if(Residual>1.0)++RayTileInteriorGt1;
                        if(Residual>RayTileInteriorResidualMaximum)
                        {
                            RayTileInteriorResidualMaximum=Residual;
                            RayTileInteriorMaximumPosition={Sample[0],Sample[1],Sample[2]};
                        }
                    }
                }
                continue;
            }
            const uint8 SkirtFaces=T<Tile.TriangleSkirtFaceMasks.size()
                ?Tile.TriangleSkirtFaceMasks[T]:0x3f;
            if((SkirtFaces&VisibleSkirtFaceMask)==0)
            {++Result.HiddenSameLodSkirtTriangles;continue;}
            ++Result.SkirtTriangles;
            const auto& PA=Tile.Vertices[A].Position;const auto& PB=Tile.Vertices[B].Position;
            const auto& PC=Tile.Vertices[C].Position;
            const VoxelForgeCore::Vec3d Center{
                (static_cast<double>(PA[0])+PB[0]+PC[0])*4.0/3.0,
                (static_cast<double>(PA[1])+PB[1]+PC[1])*4.0/3.0,
                (static_cast<double>(PA[2])+PB[2]+PC[2])*4.0/3.0};
            const double Density=Shape.densityAt(Center);
            double AlternativeCenter[3]={0.0,0.0,0.0};
            const uint32 TriangleIndices[3]={A,B,C};
            for(uint32 Index:TriangleIndices)
                for(int32 Axis=0;Axis<3;++Axis)
                {
                    double Position=Tile.Vertices[Index].Position[Axis];
                    if(Index>=SurfaceVertices)
                        Position+=2.0*SkirtDepthMeters*Tile.Vertices[Index].Normal[Axis];
                    AlternativeCenter[Axis]+=Position*4.0/3.0;
                }
            const double AlternativeDensity=Shape.densityAt({
                AlternativeCenter[0],AlternativeCenter[1],AlternativeCenter[2]});
            double Normal[3]={0.0,0.0,0.0};
            for(uint32 Index:TriangleIndices)if(Index>=SurfaceVertices)
                for(int32 Axis=0;Axis<3;++Axis)Normal[Axis]+=Tile.Vertices[Index].Normal[Axis];
            const double NormalLength=std::sqrt(Normal[0]*Normal[0]+Normal[1]*Normal[1]+Normal[2]*Normal[2]);
            if(NormalLength>1.0e-8)for(double& Value:Normal)Value/=NormalLength;
            const double Inward1=Shape.densityAt({Center.x-Normal[0],Center.y-Normal[1],Center.z-Normal[2]});
            const double Inward2=Shape.densityAt({Center.x-Normal[0]*2.0,Center.y-Normal[1]*2.0,Center.z-Normal[2]*2.0});
            double MaximumProbeDensity=Density;
            const double ProbeWeights[10][3]={
                {1.0,0.0,0.0},{0.0,1.0,0.0},{0.0,0.0,1.0},
                {1.0/3.0,1.0/3.0,1.0/3.0},
                {0.5,0.5,0.0},{0.0,0.5,0.5},{0.5,0.0,0.5},
                {0.75,0.125,0.125},{0.125,0.75,0.125},{0.125,0.125,0.75}};
            for(const auto& Weights:ProbeWeights)
            {
                const VoxelForgeCore::Vec3d Probe{
                    PA[0]*Weights[0]+PB[0]*Weights[1]+PC[0]*Weights[2],
                    PA[1]*Weights[0]+PB[1]*Weights[1]+PC[1]*Weights[2],
                    PA[2]*Weights[0]+PB[2]*Weights[1]+PC[2]*Weights[2]};
                MaximumProbeDensity=std::max(MaximumProbeDensity,static_cast<double>(
                    Shape.densityAt({Probe.x*4.0,Probe.y*4.0,Probe.z*4.0})));
            }
            if(MaximumProbeDensity>0.25)
            {
                ++Result.ExposedSkirtTriangles;
                Result.MaximumAirSkirtDensity=std::max(
                    Result.MaximumAirSkirtDensity,MaximumProbeDensity);
                if(AlternativeDensity<-0.25)++Result.ReverseSideSolidSkirtTriangles;
                if(bPrintSkirts&&Result.ExposedSkirtTriangles<=8)
                    std::printf("[ExposedTileSkirt] level=%d tile=(%d,%d,%d) center=(%.2f,%.2f,%.2f) center_density=%.3f max_probe_density=%.3f reverse_density=%.3f inward=(%.3f,%.3f) normal=(%.3f,%.3f,%.3f)\n",
                        Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                        Center.x,Center.y,Center.z,Density,MaximumProbeDensity,
                        AlternativeDensity,Inward1,Inward2,
                        Normal[0],Normal[1],Normal[2]);
            }
        }
        if(Tile.Key.Level==0&&Tile.Key.Coord.X==1
            &&Tile.Key.Coord.Y==-16&&Tile.Key.Coord.Z==-3)
        {
            std::printf("[MeshRayTileResidual] tile=(1,-16,-3) vertices=%llu mean=%.4f max=%.4f gt0.1=%llu gt0.25=%llu gt0.5=%llu max_vertex=(%.3f,%.3f,%.3f)\n",
                static_cast<unsigned long long>(RayTileResidualSamples),
                RayTileResidualSamples?RayTileResidualSum/RayTileResidualSamples:0.0,
                RayTileResidualMaximum,
                static_cast<unsigned long long>(RayTileResidualGt01),
                static_cast<unsigned long long>(RayTileResidualGt025),
                static_cast<unsigned long long>(RayTileResidualGt05),
                RayTileResidualMaximumPosition[0],RayTileResidualMaximumPosition[1],
                RayTileResidualMaximumPosition[2]);
            std::printf("[MeshRayTileInteriorResidual] tile=(1,-16,-3) samples=%llu mean=%.4f max=%.4f gt0.125=%llu gt0.25=%llu gt0.5=%llu gt1=%llu max_probe=(%.3f,%.3f,%.3f)\n",
                static_cast<unsigned long long>(RayTileInteriorSamples),
                RayTileInteriorSamples?RayTileInteriorResidualSum/RayTileInteriorSamples:0.0,
                RayTileInteriorResidualMaximum,
                static_cast<unsigned long long>(RayTileInteriorGt0125),
                static_cast<unsigned long long>(RayTileInteriorGt025),
                static_cast<unsigned long long>(RayTileInteriorGt05),
                static_cast<unsigned long long>(RayTileInteriorGt1),
                RayTileInteriorMaximumPosition[0],RayTileInteriorMaximumPosition[1],
                RayTileInteriorMaximumPosition[2]);
        }
        struct FBadCurvedEdgeRootBlock
        {FBlockKey Key{};uint64 Vertices=0,BadVertices=0;double Share=0.0,MaximumError=0.0;
            std::array<double,3> Position{};};
        std::vector<FBadCurvedEdgeRootBlock> BadCurvedEdgeRootBlocks;
        for(const auto& [Key,Block]:CurvedEdgeRootBlocks)
        {
            if(Block.Vertices<12||Block.BadVertices<6)continue;
            const double Share=static_cast<double>(Block.BadVertices)/Block.Vertices;
            if(Share<0.25)continue;
            BadCurvedEdgeRootBlocks.push_back({Key,Block.Vertices,Block.BadVertices,Share,
                Block.MaximumError,Block.Position});
        }
        std::vector<uint8> CurvedEdgeRootVisited(BadCurvedEdgeRootBlocks.size(),0);
        for(size_t I=0;I<BadCurvedEdgeRootBlocks.size();++I)
        {
            if(CurvedEdgeRootVisited[I])continue;
            std::vector<size_t> Pending{I};CurvedEdgeRootVisited[I]=1;
            uint64 BlockCount=0,BadVertexCount=0;double MaximumError=0.0;size_t Representative=I;
            while(!Pending.empty())
            {
                const size_t Current=Pending.back();Pending.pop_back();
                const FBadCurvedEdgeRootBlock& A=BadCurvedEdgeRootBlocks[Current];
                ++BlockCount;BadVertexCount+=A.BadVertices;
                if(A.MaximumError>MaximumError)
                {MaximumError=A.MaximumError;Representative=Current;}
                for(size_t J=0;J<BadCurvedEdgeRootBlocks.size();++J)
                {
                    if(CurvedEdgeRootVisited[J]
                        ||A.Key[0]!=BadCurvedEdgeRootBlocks[J].Key[0])continue;
                    int32 Manhattan=0;
                    for(int32 Axis=1;Axis<4;++Axis)
                        Manhattan+=std::abs(A.Key[Axis]-BadCurvedEdgeRootBlocks[J].Key[Axis]);
                    if(Manhattan==1)
                    {CurvedEdgeRootVisited[J]=1;Pending.push_back(J);}
                }
            }
            if(BlockCount<2||BadVertexCount<24)continue;
            ++Result.CurvedEdgeRootPatches;
            if(bPrintPatches&&Result.CurvedEdgeRootPatches<=8)
            {
                const FBadCurvedEdgeRootBlock& Sample=BadCurvedEdgeRootBlocks[Representative];
                std::printf("[MeshCurvedEdgeRootPatch] level=0 tile=(%d,%d,%d) connected_8vox_blocks=%llu bad_vertices=%llu max_root_error=%.4f vox sample=(%.3f,%.3f,%.3f)\n",
                    Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(BlockCount),
                    static_cast<unsigned long long>(BadVertexCount),MaximumError,
                    Sample.Position[0],Sample.Position[1],Sample.Position[2]);
            }
        }
        return Result;
    }

    FMeshSheetAudit AuditThinMeshSheets(const FStandaloneCpuTile& Tile,
        const FVector& CameraVoxels,float Yaw,float Pitch,bool bPrintCandidates)
    {
        FMeshSheetAudit Result;
        const size_t VertexCount=Tile.Vertices.size();
        if(VertexCount==0||Tile.Indices.size()<3)return Result;
        std::vector<uint32> Parent(VertexCount);
        std::vector<uint8> Rank(VertexCount,0);
        std::iota(Parent.begin(),Parent.end(),0u);
        auto Find=[&](uint32 Value)
        {
            uint32 Root=Value;
            while(Parent[Root]!=Root)Root=Parent[Root];
            while(Parent[Value]!=Value)
            {const uint32 Next=Parent[Value];Parent[Value]=Root;Value=Next;}
            return Root;
        };
        auto Union=[&](uint32 A,uint32 B)
        {
            A=Find(A);B=Find(B);if(A==B)return;
            if(Rank[A]<Rank[B])std::swap(A,B);
            Parent[B]=A;if(Rank[A]==Rank[B])++Rank[A];
        };
        const size_t TriangleCount=Tile.Indices.size()/3;
        for(size_t T=0;T<TriangleCount;++T)
        {
            const uint32 A=Tile.Indices[T*3],B=Tile.Indices[T*3+1],C=Tile.Indices[T*3+2];
            if(A>=VertexCount||B>=VertexCount||C>=VertexCount)continue;
            Union(A,B);Union(B,C);
        }
        struct FComponent
        {
            uint64 Triangles=0,Vertices=0;
            double Min[3]={std::numeric_limits<double>::max(),std::numeric_limits<double>::max(),std::numeric_limits<double>::max()};
            double Max[3]={-std::numeric_limits<double>::max(),-std::numeric_limits<double>::max(),-std::numeric_limits<double>::max()};
            double Sum[3]={0.0,0.0,0.0};
            double Cov[6]={0.0,0.0,0.0,0.0,0.0,0.0};
        };
        std::vector<FComponent> Components(VertexCount);
        for(size_t I=0;I<VertexCount;++I)
        {
            const uint32 Root=Find(static_cast<uint32>(I));
            FComponent& C=Components[Root];++C.Vertices;
            for(int32 A=0;A<3;++A)
            {
                const double P=Tile.Vertices[I].Position[A];
                C.Min[A]=std::min(C.Min[A],P);C.Max[A]=std::max(C.Max[A],P);C.Sum[A]+=P;
            }
        }
        for(size_t T=0;T<TriangleCount;++T)
        {
            const uint32 A=Tile.Indices[T*3],B=Tile.Indices[T*3+1],C=Tile.Indices[T*3+2];
            if(A>=VertexCount||B>=VertexCount||C>=VertexCount)continue;
            ++Components[Find(A)].Triangles;
        }
        for(size_t I=0;I<VertexCount;++I)
        {
            const uint32 Root=Find(static_cast<uint32>(I));
            FComponent& C=Components[Root];
            if(C.Vertices==0)continue;
            const double P[3]={Tile.Vertices[I].Position[0],Tile.Vertices[I].Position[1],Tile.Vertices[I].Position[2]};
            const double D[3]={P[0]-C.Sum[0]/C.Vertices,P[1]-C.Sum[1]/C.Vertices,P[2]-C.Sum[2]/C.Vertices};
            C.Cov[0]+=D[0]*D[0];C.Cov[1]+=D[1]*D[1];C.Cov[2]+=D[2]*D[2];
            C.Cov[3]+=D[0]*D[1];C.Cov[4]+=D[0]*D[2];C.Cov[5]+=D[1]*D[2];
        }
        const double Camera[3]={CameraVoxels.X*0.25,CameraVoxels.Y*0.25,CameraVoxels.Z*0.25};
        const double CP=std::cos(Pitch),Forward[3]={std::sin(Yaw)*CP,std::cos(Yaw)*CP,std::sin(Pitch)};
        const double Right[3]={std::cos(Yaw),-std::sin(Yaw),0.0};
        const double Up[3]={-std::sin(Yaw)*std::sin(Pitch),-std::cos(Yaw)*std::sin(Pitch),std::cos(Pitch)};
        for(size_t Root=0;Root<VertexCount;++Root)
        {
            FComponent& C=Components[Root];if(C.Triangles==0||C.Vertices==0)continue;
            ++Result.Components;
            const double Extent[3]={C.Max[0]-C.Min[0],C.Max[1]-C.Min[1],C.Max[2]-C.Min[2]};
            const double MaxExtent=std::max({Extent[0],Extent[1],Extent[2]})/0.25;
            const bool bSmall=C.Triangles<=256&&MaxExtent<=20.0;
            if(bSmall)++Result.SmallComponents;
            if(!bSmall||C.Vertices<4)continue;
            double A[3][3]={{C.Cov[0],C.Cov[3],C.Cov[4]},
                {C.Cov[3],C.Cov[1],C.Cov[5]},{C.Cov[4],C.Cov[5],C.Cov[2]}};
            for(int32 Sweep=0;Sweep<12;++Sweep)
            {
                int32 P=0,Q=1;
                if(std::abs(A[0][2])>std::abs(A[P][Q])){P=0;Q=2;}
                if(std::abs(A[1][2])>std::abs(A[P][Q])){P=1;Q=2;}
                if(std::abs(A[P][Q])<1.0e-16)break;
                const double Tau=(A[Q][Q]-A[P][P])/(2.0*A[P][Q]);
                const double T=std::copysign(1.0,Tau)/(std::abs(Tau)+std::sqrt(1.0+Tau*Tau));
                const double Cos=1.0/std::sqrt(1.0+T*T),Sin=T*Cos;
                const double App=A[P][P],Aqq=A[Q][Q],Apq=A[P][Q];
                A[P][P]=App-T*Apq;A[Q][Q]=Aqq+T*Apq;A[P][Q]=A[Q][P]=0.0;
                for(int32 K=0;K<3;++K)if(K!=P&&K!=Q)
                {
                    const double Akp=A[K][P],Akq=A[K][Q];
                    A[K][P]=A[P][K]=Cos*Akp-Sin*Akq;
                    A[K][Q]=A[Q][K]=Sin*Akp+Cos*Akq;
                }
            }
            const double Variance=std::max(0.0,std::min({A[0][0],A[1][1],A[2][2]})/C.Vertices);
            const double ThicknessVoxels=3.5*std::sqrt(Variance)/0.25;
            const double Center[3]={C.Sum[0]/C.Vertices,C.Sum[1]/C.Vertices,C.Sum[2]/C.Vertices};
            const double Relative[3]={Center[0]-Camera[0],Center[1]-Camera[1],Center[2]-Camera[2]};
            const double Depth=Relative[0]*Forward[0]+Relative[1]*Forward[1]+Relative[2]*Forward[2];
            const double ScreenX=Relative[0]*Right[0]+Relative[1]*Right[1]+Relative[2]*Right[2];
            const double ScreenY=Relative[0]*Up[0]+Relative[1]*Up[1]+Relative[2]*Up[2];
            const bool bInView=Depth>0.0&&Depth<=64.0&&std::abs(ScreenX)<0.924*Depth
                &&std::abs(ScreenY)<0.5774*Depth;
            if(!bInView)continue;
            const FIntVector TileOrigin=Tile.Key.OriginVoxels();
            const int32 TileMin[3]={TileOrigin.X,TileOrigin.Y,TileOrigin.Z};
            const int32 TileMax[3]={TileOrigin.X+Tile.Key.ExtentVoxels(),
                TileOrigin.Y+Tile.Key.ExtentVoxels(),TileOrigin.Z+Tile.Key.ExtentVoxels()};
            bool bTouchesTileBoundary=false;
            for(int32 Axis=0;Axis<3;++Axis)
            {
                const double MinVoxels=C.Min[Axis]*4.0,MaxVoxels=C.Max[Axis]*4.0;
                bTouchesTileBoundary|=MinVoxels<=TileMin[Axis]+0.5
                    ||MaxVoxels>=TileMax[Axis]-0.5;
            }
            if(bTouchesTileBoundary)++Result.BoundarySmallComponentsInView;
            else ++Result.SmallComponentsInView;
            if(bPrintCandidates&&!bTouchesTileBoundary&&Result.SmallComponentsInView<=8)
                std::printf("[SmallMeshComponent] level=%d tile=(%d,%d,%d) tris=%llu size=(%.2f,%.2f,%.2f) vox pca_thickness=%.3f vox center=(%.2f,%.2f,%.2f)m\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(C.Triangles),Extent[0]/0.25,Extent[1]/0.25,
                    Extent[2]/0.25,ThicknessVoxels,Center[0],Center[1],Center[2]);
            else if(bPrintCandidates&&bTouchesTileBoundary
                &&Result.BoundarySmallComponentsInView<=8)
                std::printf("[BoundaryMeshComponent] level=%d tile=(%d,%d,%d) tris=%llu size=(%.2f,%.2f,%.2f) vox pca_thickness=%.3f vox center=(%.2f,%.2f,%.2f)m\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(C.Triangles),Extent[0]/0.25,Extent[1]/0.25,
                    Extent[2]/0.25,ThicknessVoxels,Center[0],Center[1],Center[2]);
            if(ThicknessVoxels>=1.0)continue;
            ++Result.ThinSheets;Result.MinimumThicknessVoxels=std::min(Result.MinimumThicknessVoxels,ThicknessVoxels);
            Result.TrianglesInThinSheets+=C.Triangles;
            if(bPrintCandidates&&Result.ThinSheets==1)
                std::printf("[ThinMeshSheetCandidate] level=%d tile=(%d,%d,%d) tris=%llu size=(%.2f,%.2f,%.2f) vox pca_thickness=%.3f vox center=(%.2f,%.2f,%.2f)m\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(C.Triangles),Extent[0]/0.25,Extent[1]/0.25,Extent[2]/0.25,
                    ThicknessVoxels,Center[0],Center[1],Center[2]);
        }
        return Result;
    }

    struct FCoarsePanelAudit
    {
        uint64 Tiles=0,PlanarOpenComponents=0,OffSurfacePanels=0,Triangles=0;
        double MaximumSourceErrorVoxels=0.0;
    };

    FCoarsePanelAudit AuditCoarseSurfacePanels(const FStandaloneCpuTile& Tile,
        const VoxelForgeCore::World& Shape,const VoxelForgeCore::WorldConfiguration& Config,
        bool bPrintCandidates)
    {
        FCoarsePanelAudit Result;
        const size_t VertexCount=Tile.Vertices.size();
        const size_t TriangleCount=Tile.Indices.size()/3;
        if(Tile.Key.Level<1||VertexCount==0||TriangleCount==0)return Result;
        std::vector<uint32> Parent(VertexCount);
        std::vector<uint8> Rank(VertexCount,0);
        std::iota(Parent.begin(),Parent.end(),0u);
        const auto Find=[&](uint32 Value)
        {
            uint32 Root=Value;
            while(Parent[Root]!=Root)Root=Parent[Root];
            while(Parent[Value]!=Value)
            {const uint32 Next=Parent[Value];Parent[Value]=Root;Value=Next;}
            return Root;
        };
        const auto Union=[&](uint32 A,uint32 B)
        {
            A=Find(A);B=Find(B);if(A==B)return;
            if(Rank[A]<Rank[B])std::swap(A,B);
            Parent[B]=A;if(Rank[A]==Rank[B])++Rank[A];
        };
        const auto EdgeKey=[](uint32 A,uint32 B)
        {if(A>B)std::swap(A,B);return (static_cast<uint64>(A)<<32)|B;};
        std::unordered_map<uint64,uint8> EdgeUse;
        EdgeUse.reserve(TriangleCount*3/2+1);
        for(size_t T=0;T<TriangleCount;++T)
        {
            const uint32 A=Tile.Indices[T*3],B=Tile.Indices[T*3+1],C=Tile.Indices[T*3+2];
            if(A>=VertexCount||B>=VertexCount||C>=VertexCount)continue;
            Union(A,B);Union(B,C);
            for(const auto [U,V]:std::array<std::pair<uint32,uint32>,3>{{{A,B},{B,C},{C,A}}})
            {
                uint8& Count=EdgeUse[EdgeKey(U,V)];
                if(Count<255)++Count;
            }
        }
        struct FComponent
        {
            uint64 Triangles=0,InteriorFreeEdges=0,SourceSamples=0,OffSourceSamples=0;
            double Normal[3]={0,0,0},NormalWeight=0,MaximumSourceError=0;
            double WorstPosition[3]={0,0,0};
            bool bWorstTriangleCentroid=false;
        };
        std::vector<FComponent> Components(VertexCount);
        for(size_t T=0;T<TriangleCount;++T)
        {
            const uint32 A=Tile.Indices[T*3],B=Tile.Indices[T*3+1],C=Tile.Indices[T*3+2];
            if(A>=VertexCount||B>=VertexCount||C>=VertexCount)continue;
            const uint32 Root=Find(A);FComponent& Stats=Components[Root];++Stats.Triangles;
            const auto& PA=Tile.Vertices[A].Position;const auto& PB=Tile.Vertices[B].Position;
            const auto& PC=Tile.Vertices[C].Position;
            const double U[3]={PB[0]-PA[0],PB[1]-PA[1],PB[2]-PA[2]};
            const double V[3]={PC[0]-PA[0],PC[1]-PA[1],PC[2]-PA[2]};
            double N[3]={U[1]*V[2]-U[2]*V[1],U[2]*V[0]-U[0]*V[2],U[0]*V[1]-U[1]*V[0]};
            const double Area2=std::sqrt(N[0]*N[0]+N[1]*N[1]+N[2]*N[2]);
            if(Area2>1.0e-12)
            {
                for(int32 Axis=0;Axis<3;++Axis)Stats.Normal[Axis]+=N[Axis]/Area2*Area2;
                Stats.NormalWeight+=Area2;
            }
        }
        const FIntVector Origin=Tile.Key.OriginVoxels();
        const double TileMin[3]={static_cast<double>(Origin.X),static_cast<double>(Origin.Y),static_cast<double>(Origin.Z)};
        const double TileMax[3]={TileMin[0]+Tile.Key.ExtentVoxels(),TileMin[1]+Tile.Key.ExtentVoxels(),TileMin[2]+Tile.Key.ExtentVoxels()};
        const auto IsTileBoundaryEdge=[&](uint32 A,uint32 B)
        {
            for(int32 Axis=0;Axis<3;++Axis)
            {
                const double PA=Tile.Vertices[A].Position[Axis]*4.0;
                const double PB=Tile.Vertices[B].Position[Axis]*4.0;
                if((std::abs(PA-TileMin[Axis])<0.75&&std::abs(PB-TileMin[Axis])<0.75)
                    ||(std::abs(PA-TileMax[Axis])<0.75&&std::abs(PB-TileMax[Axis])<0.75))return true;
            }
            return false;
        };
        for(const auto& [Key,Count]:EdgeUse)
        {
            if(Count!=1)continue;
            const uint32 A=static_cast<uint32>(Key>>32),B=static_cast<uint32>(Key);
            if(A>=VertexCount||B>=VertexCount||IsTileBoundaryEdge(A,B))continue;
            ++Components[Find(A)].InteriorFreeEdges;
        }
        const int32 Extent=Tile.Key.ExtentVoxels();
        const int32 Cells=Tile.Key.Level<Config.fullResolutionLodCount
            ?CHUNK_SIZE:std::max(1,Config.coarseTileCells);
        const double Step=std::max(1,Extent/Cells);
        const double ResidualLimit=std::max(2.0,Step*0.10);
        std::vector<uint8> Candidate(VertexCount,0);
        for(size_t Root=0;Root<VertexCount;++Root)
        {
            FComponent& Stats=Components[Root];
            if(Stats.Triangles==0||Stats.Triangles>256||Stats.InteriorFreeEdges==0||Stats.NormalWeight<=0.0)continue;
            const double NormalLength=std::sqrt(Stats.Normal[0]*Stats.Normal[0]
                +Stats.Normal[1]*Stats.Normal[1]+Stats.Normal[2]*Stats.Normal[2]);
            const double Coherence=NormalLength/Stats.NormalWeight;
            if(Coherence<0.995)continue;
            Candidate[Root]=1;++Result.PlanarOpenComponents;
        }
        if(Result.PlanarOpenComponents==0)return Result;
        for(size_t Vertex=0;Vertex<VertexCount;++Vertex)
        {
            const uint32 Root=Find(static_cast<uint32>(Vertex));
            if(!Candidate[Root])continue;
            const auto& P=Tile.Vertices[Vertex].Position;
            const double Error=std::abs(Shape.densityAt({P[0]*4.0,P[1]*4.0,P[2]*4.0}));
            FComponent& Stats=Components[Root];++Stats.SourceSamples;
            if(Error>Stats.MaximumSourceError)
            {
                Stats.MaximumSourceError=Error;
                std::copy_n(P,3,Stats.WorstPosition);
                Stats.bWorstTriangleCentroid=false;
            }
            if(Error>ResidualLimit)++Stats.OffSourceSamples;
        }
        for(size_t T=0;T<TriangleCount;++T)
        {
            const uint32 A=Tile.Indices[T*3],B=Tile.Indices[T*3+1],C=Tile.Indices[T*3+2];
            if(A>=VertexCount||B>=VertexCount||C>=VertexCount)continue;
            const uint32 Root=Find(A);if(!Candidate[Root])continue;
            const auto& PA=Tile.Vertices[A].Position;const auto& PB=Tile.Vertices[B].Position;
            const auto& PC=Tile.Vertices[C].Position;
            const double Error=std::abs(Shape.densityAt({(PA[0]+PB[0]+PC[0])*4.0/3.0,
                (PA[1]+PB[1]+PC[1])*4.0/3.0,(PA[2]+PB[2]+PC[2])*4.0/3.0}));
            FComponent& Stats=Components[Root];++Stats.SourceSamples;
            if(Error>Stats.MaximumSourceError)
            {
                Stats.MaximumSourceError=Error;
                Stats.WorstPosition[0]=(PA[0]+PB[0]+PC[0])/3.0;
                Stats.WorstPosition[1]=(PA[1]+PB[1]+PC[1])/3.0;
                Stats.WorstPosition[2]=(PA[2]+PB[2]+PC[2])/3.0;
                Stats.bWorstTriangleCentroid=true;
            }
            if(Error>ResidualLimit)++Stats.OffSourceSamples;
        }
        for(size_t Root=0;Root<VertexCount;++Root)
        {
            const FComponent& Stats=Components[Root];if(!Candidate[Root]||Stats.SourceSamples==0)continue;
            const double OffShare=static_cast<double>(Stats.OffSourceSamples)/Stats.SourceSamples;
            Result.MaximumSourceErrorVoxels=std::max(Result.MaximumSourceErrorVoxels,Stats.MaximumSourceError);
            if(OffShare<0.25)continue;
            ++Result.OffSurfacePanels;Result.Triangles+=Stats.Triangles;
            if(bPrintCandidates&&Result.OffSurfacePanels<=8)
                std::printf("[CoarsePanelCandidate] lod=%d tile=(%d,%d,%d) triangles=%llu interior_free_edges=%llu source_error=%.3f vox off_share=%.1f%% limit=%.3f vox worst=(%.2f,%.2f,%.2f)m sample=%s\n",
                    Tile.Key.Level,Tile.Key.Coord.X,Tile.Key.Coord.Y,Tile.Key.Coord.Z,
                    static_cast<unsigned long long>(Stats.Triangles),
                    static_cast<unsigned long long>(Stats.InteriorFreeEdges),Stats.MaximumSourceError,
                    OffShare*100.0,ResidualLimit,
                    Stats.WorstPosition[0],Stats.WorstPosition[1],Stats.WorstPosition[2],
                    Stats.bWorstTriangleCentroid?"centroid":"vertex");
        }
        Result.Tiles=1;
        return Result;
    }

    struct M4 { float M[4][4]{}; };
    M4 Identity(){M4 R{};for(int I=0;I<4;++I)R.M[I][I]=1.0f;return R;}
    M4 Transpose(M4 A){M4 B{};for(int I=0;I<4;++I)for(int J=0;J<4;++J)B.M[I][J]=A.M[J][I];return B;}
    M4 Perspective(float Fov,float Aspect,float NearZ,float FarZ)
    {
        M4 R{};const float Y=1.0f/std::tan(Fov*0.5f);
        R.M[0][0]=Y/Aspect;R.M[1][1]=Y;R.M[2][2]=FarZ/(FarZ-NearZ);
        R.M[2][3]=1.0f;R.M[3][2]=-NearZ*FarZ/(FarZ-NearZ);return R;
    }
    M4 View(V3 Position,V3 Forward,V3 Right,V3 Up)
    {
        M4 R=Identity();
        R.M[0][0]=Right.X;R.M[1][0]=Right.Y;R.M[2][0]=Right.Z;R.M[3][0]=-Dot(Right,Position);
        R.M[0][1]=Up.X;R.M[1][1]=Up.Y;R.M[2][1]=Up.Z;R.M[3][1]=-Dot(Up,Position);
        R.M[0][2]=Forward.X;R.M[1][2]=Forward.Y;R.M[2][2]=Forward.Z;R.M[3][2]=-Dot(Forward,Position);
        return R;
    }

    struct FGpuVertex { float Position[3]; float Normal[3]; };
    struct FConstants
    {
        M4 ViewProjection;
        float Camera[3]; float FogStart;
        float Forward[3]; float FogEnd;
        float LightDirection[3]; float Ambient;
        float Padding[4];
    };

    constexpr int32 MaxLevels=8;
    struct FStandaloneBatchKey
    {
        int32 Level=0,X=0,Y=0,Z=0;
        bool operator==(const FStandaloneBatchKey& O)const
        {return Level==O.Level&&X==O.X&&Y==O.Y&&Z==O.Z;}
    };
    struct FStandaloneBatchKeyHash
    {
        size_t operator()(const FStandaloneBatchKey& K)const noexcept
        {
            uint64 H=static_cast<uint32>(K.Level);
            H=(H*0x9e3779b185ebca87ull)^static_cast<uint32>(K.X);
            H=(H*0x9e3779b185ebca87ull)^static_cast<uint32>(K.Y);
            H=(H*0x9e3779b185ebca87ull)^static_cast<uint32>(K.Z);
            return static_cast<size_t>(H^(H>>32));
        }
    };
    struct FResidentTileMesh
    {
        std::vector<FStandaloneCpuVertex> Vertices;
        std::vector<uint32> Indices;
        std::vector<uint8> TriangleSkirtFaceMasks;
        uint8 ActiveSkirtFaceMask=0x3f;
        uint64 Samples=0;
        double WorkerSeconds=0.0;
    };
    struct FStreamingGpuBatch
    {
        ID3D11Buffer* VertexBuffer=nullptr;
        ID3D11Buffer* IndexBuffer=nullptr;
        UINT IndexCount=0;
        UINT DrawOnlyIndexStart=0,DrawOnlyIndexCount=0;
        uint64 VertexCount=0;
    };
    struct FCrossingWork
    {
        uint64 SubmittedJobs=0;
        uint64 PendingJobs=0;
        uint64 FinishedJobs=0;
        uint64 CancelledQueuedJobs=0;
        uint64 StaleResults=0;
        uint64 Tiles=0;
        uint64 Samples=0;
        double WorkerSeconds=0.0;
        double CorePageSeconds=0.0;
        double CorePageWorkerSeconds=0.0;
        uint64 PageCoveredTiles=0;
        uint64 PageRegions=0;
        uint64 PageLatticeNodes=0;
        FVoxelTileKey NearestNewTile{};
        double NearestReadySeconds=-1.0;
        bool bHasNearestNewTile=false;
        uint64 TilesByLevel[MaxLevels+1]{};
        uint64 SamplesByLevel[MaxLevels+1]{};
        double WorkerSecondsByLevel[MaxLevels+1]{};
        double WallSeconds=0.0;
        std::chrono::steady_clock::time_point StartTime{};
        bool bReported=false;
    };
    struct FGpuFarTile
    {
        ID3D11Buffer* VertexBuffer=nullptr;
        ID3D11Buffer* IndexBuffer=nullptr;
        ID3D11Buffer* IndirectArgs=nullptr;
        ID3D11Buffer* ConnectorVertexBuffer=nullptr;
        ID3D11Buffer* ConnectorIndexBuffer=nullptr;
        ID3D11Query* CompletionEvent=nullptr;
        VoxelForgeGpuFarViewer::GpuTimingQueries TimingQueries{};
        uint64 VertexCapacityBytes=0;
        uint32 OutputSlot=0,IndirectArgsByteOffset=0,BatchIndex=UINT32_MAX;
        uint8 VisibleSkirtFaceMask=0;
        bool bSkirtDrawArgsAudited=false;
        double GpuDispatchMs=0.0;
        bool bGpuComplete=false;
        bool bTimingResolved=false;
    };
    struct FGpuFarBatch
    {
        ID3D11Query* CompletionEvent=nullptr;
        VoxelForgeGpuFarViewer::GpuTimingQueries TimingQueries{};
        uint32 TileCount=0;
        uint64 TimestampFrequency=0;
        double GpuDispatchMs=0.0;
        std::chrono::steady_clock::time_point SubmittedAt{};
        bool bGpuComplete=false,bTimingResolved=false;
    };

    uint64 HashStandaloneFile(const std::string& Path)
    {
        std::ifstream File(Path,std::ios::binary);
        if(!File)return 0;
        uint64 Hash=14695981039346656037ull;
        char Buffer[65536];
        while(File)
        {
            File.read(Buffer,sizeof(Buffer));
            const std::streamsize Count=File.gcount();
            for(std::streamsize I=0;I<Count;++I)
            {
                Hash^=static_cast<uint8>(Buffer[I]);
                Hash*=1099511628211ull;
            }
        }
        return Hash;
    }

    FStandaloneBatchKey BatchForTile(const FVoxelTileKey& Tile)
    {
        return {Tile.Level,VoxelClipmapDesiredTiles::FloorDiv(Tile.Coord.X,8),
            VoxelClipmapDesiredTiles::FloorDiv(Tile.Coord.Y,8),Tile.Coord.Z};
    }

    bool PrepareStandaloneDensityWorld(FStandaloneOwnerWorld& World,int32 Seed,
                                       bool bRoundTunnels,bool bEnhancedWalls,std::string& Error)
    {
        World.Generator.reset();World.Manager.reset();
        World.Settings->Seed=Seed;World.Seed=Seed;
        // Keep the owner's walkable rooms, tunnels, and strate passages intact. The research
        // round field composes as additional air after the owner field, so it can widen/connect
        // spaces but cannot remove the authored entrance-to-exit route.
        World.Manager=std::make_unique<UVoxelStrateManager>();
        if(!World.Manager->Initialize(World.Settings.get(),Seed))
        {Error="standalone copied strate manager failed while applying viewer settings";return false;}
        World.Generator=std::make_unique<UVoxelGenerator>();
        World.Generator->InitializeSettings(World.Settings.get());
        World.Generator->SetStrateManager(World.Manager.get());
        World.Generator->bStandaloneResearchRoundTunnels=bRoundTunnels;
        World.Generator->bStandaloneResearchEnhancedWalls=bEnhancedWalls;
        return true;
    }

    struct FApp
    {
        HWND Window=nullptr;
        ID3D11Device* Device=nullptr;
        ID3D11DeviceContext* Context=nullptr;
        IDXGISwapChain* SwapChain=nullptr;
        ID3D11RenderTargetView* RenderTarget=nullptr;
        ID3D11DepthStencilView* DepthView=nullptr;
        ID3D11VertexShader* VertexShader=nullptr;
        ID3D11PixelShader* PixelShader=nullptr;
        ID3D11InputLayout* InputLayout=nullptr;
        ID3D11Buffer* Constants=nullptr;
        ID3D11RasterizerState* SolidRasterizer=nullptr;
        ID3D11RasterizerState* WireRasterizer=nullptr;
        std::shared_ptr<FStandaloneOwnerWorld> World;
        VoxelForgeCore::ResolvedWorldSnapshot Snapshot;
        std::unique_ptr<VoxelForgeCore::World> ShapeWorld;
        std::unique_ptr<VoxelForgeCore::World> PendingFullShapeWorld;
        FStandaloneCoreScheduler CoreScheduler;
        std::unique_ptr<FStandaloneStreamingGenerator> Streamer;
        std::thread FullShapeBuildThread;
        std::atomic<bool> bFullShapeBuildFinished{false};
        bool bFullShapeBuildSucceeded=false;
        std::string FullShapeBuildError;
        std::chrono::steady_clock::time_point FullShapeBuildStart{};
        FStandaloneMeshCache MeshCache;
        std::string DataPath;
        VFJson::Value OwnerJsonRoot;
        bool bHasCameraOverride=false;
        V3 CameraPositionOverride{};
        float CameraYawOverride=0.0f,CameraPitchOverride=0.0f;
        std::unordered_map<FVoxelTileKey,FResidentTileMesh,FStandaloneTileKeyHash> ResidentTiles;
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> DeferredLeaverTiles;
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> DesiredTileSet;
        std::unordered_map<FStandaloneBatchKey,std::vector<FVoxelTileKey>,FStandaloneBatchKeyHash> BatchMembers;
        std::unordered_map<FStandaloneBatchKey,FStreamingGpuBatch,FStandaloneBatchKeyHash> StreamingBatches;
        std::unordered_map<FVoxelTileKey,FGpuFarTile,FStandaloneTileKeyHash> GpuFarTiles;
        std::vector<FGpuFarBatch> GpuFarBatches;
        std::deque<FVoxelTileKey> GpuFarQueue;
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> GpuFarQueued;
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> GpuFarEligibleTiles;
        std::unordered_set<FStandaloneBatchKey,FStandaloneBatchKeyHash> DirtyBatchSet;
        std::unordered_map<uint64,FCrossingWork> CrossingWork;
        std::vector<FVoxelTileKey> DesiredTiles;
        std::vector<FVoxelTileKey> InitialNeighborhoodTiles;
        int32 DesiredByLevel[MaxLevels+1]{};
        uint64 GeneratedByLevel[MaxLevels+1]{};
        uint64 SamplesByLevel[MaxLevels+1]{};
        double WorkerSecondsByLevel[MaxLevels+1]{};
        uint64 ProfileCyclesByLevel[MaxLevels+1][VoxelDensityProfile::BucketCount]{};
        std::vector<double> TileSecondsByLevel[MaxLevels+1];
        uint64 CrossingIndex=0;
        bool bGenerationRunning=false;
        bool bRebuildPending=false;
        bool bParityChecked=false;
        bool bRoundTunnels=false;
        bool bRoundTunnelEnhanced=true;
        bool bShapeWorld=false;
        bool bShapePipelineActive=false;
        bool bNeighborhoodReadyReported=false;
        bool bShapeAuditOnly=false;
        bool bShapeCompare=false;
        bool bFarShapeStagePending=false;
        bool bGpuFarEnabled=false;
        bool bGpuFarAuditEnabled=false;
        bool bGpuFarShaderPrewarm=false,bDisableGpuWarmupPresentHold=false;
        bool bGpuFarWarmupResourcesOnly=false;
        bool bGpuFarWarmupNormalPriority=false,bGpuFarWarmupBeforeWorkers=false;
        bool bGpuFarWarmupOnly=false;
        std::string GpuWarmupEntryFilter;
        std::thread GpuShaderWarmupThread;
        bool bGpuShaderWarmupQueueSucceeded=false;
        std::string GpuShaderWarmupQueueError;
        bool bGpuFarAuditReported=false;
        bool bGpuSkirtEmitEnabled=true,bGpuConnectorRepairEnabled=true;
        bool bDisableVSync=false;
        bool bForceBlockingStartupPresent=false;
        bool bPresentCpuFrameBeforeGpu=false;
        bool bCpuReadyFramePresentedBeforeGpu=false;
        bool bHoldCpuFrameDuringGpuWarmup=false,bGpuWarmupHoldReported=false;
        bool bGpuWarmupReleaseReported=false;
        uint64 GpuWarmupHeldPresents=0;
        bool bAuditGpuSkirtDrawArgs=false,bGpuSkirtDrawAuditReported=false;
        ID3D11Buffer* GpuSkirtArgsStaging=nullptr;
        uint64 GpuSkirtDrawAuditTiles=0,GpuSkirtDrawAuditZeroCounts=0,
            GpuSkirtDrawAuditOverflow=0,GpuSkirtDrawAuditBadStartIndex=0;
        uint64 GpuSkirtDrawAuditSumIndices=0,GpuSkirtDrawAuditMaxIndices=0;
        std::unique_ptr<FFarSurvivalAudit> GpuFarAudit;
        bool bGpuFarFirstTileReported=false;
        std::filesystem::path GpuFarShaderPath;
        double GpuFarEnqueueWallSeconds=0.0;
        uint64 GpuFarTilesSubmitted=0;
        uint64 PresentCallCount=0;
        uint64 StartupNonBlockingPresentCalls=0,StartupNonBlockingPresentSkips=0;
        double GpuFarGpuDispatchMs=0.0;
        double GpuFarGpuDispatchMaxMs=0.0;
        double GpuFarSubmissionSpanMs=0.0,GpuFarSubmissionSpanMaxMs=0.0;
        uint64 GpuFarLocalMemoryPeakBytes=0,GpuFarLocalMemoryBudgetBytes=0;
        double GpuFarPoolReserveMs=0.0;
        uint64 GpuFarSubmissionBatches=0;
        uint32 GpuFarMaxTilesPerSubmissionBatch=0;
        uint32 GpuFarOutputCapacitySlots=0,GpuFarNextOutputSlot=0;
        std::vector<uint32> GpuFarFreeOutputSlots;
        std::unordered_map<FVoxelTileKey,VoxelForgeCore::FarFieldTileInput,FStandaloneTileKeyHash> GpuFarPreparedInputs;
        std::unordered_map<FVoxelTileKey,std::string,FStandaloneTileKeyHash> GpuFarPreparationErrors;
        double GpuFarPreparationWallMs=0.0,GpuFarPreparationWorkerMs=0.0;
        uint64 GpuFarPumpCalls=0,GpuFarCompletionPolls=0;
        double GpuFarFirstPumpMs=-1.0,GpuFarLastPumpMs=-1.0;
        double GpuFarFirstSubmitMs=-1.0,GpuFarLastSubmitMs=-1.0;
        double GpuFarFirstCompletionPollMs=-1.0,GpuFarLastCompletionPollMs=-1.0;
        double GpuFarPumpWallMs=0.0,GpuFarPumpMaxWallMs=0.0;
        double GpuFarPumpGapMs=0.0,GpuFarPumpGapMaxMs=0.0;
        std::chrono::steady_clock::time_point LastGpuFarPumpAt{};
        std::chrono::steady_clock::time_point LastDrawEndAt{};
        VoxelForgeCore::DetailVariant ShapeDetailVariant=VoxelForgeCore::DetailVariant::coherent;
        bool bWireframe=false;
        bool bCaptured=false;
        bool bAutoCaptureWritten=false;
        bool bFullDetailReported=false;
        bool bInitialClipmapReported=false;
        bool bWholeViewPresented=false;
        bool bGpuFirstBackedPresentReported=false;
        bool bProfileAttribution=false;
        bool bProfileDetailBand=false;
        bool bCompareEvaluators=false;
        bool bParityOnly=false;
        bool bDensityHash=false;
        int32 FlightTargetCrossings=0;
        int32 FlightObservedCrossings=0;
        uint64 FlightStartCrossingIndex=0;
        float FlightSpeedMps=16.0f;
        bool bFlightStarted=false;
        bool bFlightFinished=false;
        bool bFlightClosePosted=false;
        bool bFlightCrossingsValid=true;
        V3 FlightStartPosition{};
        std::chrono::steady_clock::time_point FlightStartTime{};
        bool bCaptureAfterClipmap=false;
        bool bQuitAfterClipmap=false;
        bool bPrintedClipOverride=false;
        bool bForceGenerate=false;
        std::string CapturePath;
        std::string MeshCachePath;
        std::chrono::steady_clock::time_point ColdProcessStart{};
        std::chrono::steady_clock::time_point GenerationStart{};
        bool bShowLevel[MaxLevels+1]={true,true,true,true,true,true,true,true,true};
        bool Keys[256]{};
        int32 Seed=0;
        int32 WorkerCount=8;
        int32 ClipRadius=3;
        int32 MaxClipLevelOverride=-1;
        int32 CaptureTileLimit=0;
        int32 CaptureOnlyLod=-1;
        bool bCaptureOnlyTile=false;
        int32 CaptureOnlyTileX=0,CaptureOnlyTileY=0,CaptureOnlyTileZ=0,CaptureOnlyTileLod=0;
        bool bDrawOnlyTile=false;
        int32 DrawOnlyTileX=0,DrawOnlyTileY=0,DrawOnlyTileZ=0,DrawOnlyTileLod=0;
        int32 CaptureOutputSlot=-1;
        int32 Width=1280,Height=800;
        float Yaw=0.0f,Pitch=0.0f,MoveSpeed=8.0f;
        V3 Position{};
        POINT LastMouse{};
        std::chrono::steady_clock::time_point PreviousFrame=std::chrono::steady_clock::now();
        double FrameMs=0.0;
        uint64 TotalVertices=0;
        uint64 TotalIndices=0;
        FIntVector BuildTargetChunk=FIntVector::ZeroValue;
        int32 MaxBatchUploadsPerFrame=4;
        double BatchUploadBudgetMilliseconds=2.0;

        ~FApp(){
            if(GpuShaderWarmupThread.joinable())GpuShaderWarmupThread.join();
            StopGeneration(true);ReleaseMeshes();
            if(SolidRasterizer)SolidRasterizer->Release();if(WireRasterizer)WireRasterizer->Release();
            if(Constants)Constants->Release();if(InputLayout)InputLayout->Release();
            if(GpuSkirtArgsStaging)GpuSkirtArgsStaging->Release();
            if(VertexShader)VertexShader->Release();if(PixelShader)PixelShader->Release();
            if(RenderTarget)RenderTarget->Release();if(DepthView)DepthView->Release();
            if(SwapChain)SwapChain->Release();if(Context)Context->Release();if(Device)Device->Release();
        }

        static void ReleaseGpuFarTile(FGpuFarTile& Tile)
        {
            if(Tile.VertexBuffer)Tile.VertexBuffer->Release();
            if(Tile.IndexBuffer)Tile.IndexBuffer->Release();
            if(Tile.IndirectArgs)Tile.IndirectArgs->Release();
            if(Tile.ConnectorVertexBuffer)Tile.ConnectorVertexBuffer->Release();
            if(Tile.ConnectorIndexBuffer)Tile.ConnectorIndexBuffer->Release();
            if(Tile.CompletionEvent)Tile.CompletionEvent->Release();
            if(Tile.TimingQueries.start)Tile.TimingQueries.start->Release();
            if(Tile.TimingQueries.end)Tile.TimingQueries.end->Release();
            if(Tile.TimingQueries.disjoint)Tile.TimingQueries.disjoint->Release();
            Tile={};
        }

        static void ReleaseGpuFarBatch(FGpuFarBatch& Batch)
        {
            if(Batch.CompletionEvent)Batch.CompletionEvent->Release();
            if(Batch.TimingQueries.start)Batch.TimingQueries.start->Release();
            if(Batch.TimingQueries.end)Batch.TimingQueries.end->Release();
            if(Batch.TimingQueries.disjoint)Batch.TimingQueries.disjoint->Release();
            Batch={};
        }

        void SampleGpuFarMemory()
        {
            if(!bGpuFarEnabled||!Device||!Context)return;
            uint64 Usage=0,Budget=0;std::string Error;
            if(VoxelForgeGpuFarViewer::QueryFarGpuMemory(Device,Context,GpuFarShaderPath,
                Usage,Budget,Error))
            {
                GpuFarLocalMemoryPeakBytes=std::max(GpuFarLocalMemoryPeakBytes,Usage);
                GpuFarLocalMemoryBudgetBytes=Budget;
            }
        }

        void ReleaseMeshes()
        {
            for(auto& Pair:StreamingBatches)
            {
                if(Pair.second.VertexBuffer)Pair.second.VertexBuffer->Release();
                if(Pair.second.IndexBuffer)Pair.second.IndexBuffer->Release();
            }
            StreamingBatches.clear();BatchMembers.clear();ResidentTiles.clear();
            for(auto& Pair:GpuFarTiles)
                ReleaseGpuFarTile(Pair.second);
            GpuFarTiles.clear();
            for(FGpuFarBatch& Batch:GpuFarBatches)ReleaseGpuFarBatch(Batch);
            GpuFarBatches.clear();GpuFarQueue.clear();GpuFarQueued.clear();GpuFarEligibleTiles.clear();
            GpuFarFreeOutputSlots.clear();GpuFarNextOutputSlot=0;
            GpuFarPreparedInputs.clear();GpuFarPreparationErrors.clear();
            GpuFarPreparationWallMs=0.0;GpuFarPreparationWorkerMs=0.0;
            GpuFarPumpCalls=0;GpuFarCompletionPolls=0;
            GpuFarFirstPumpMs=GpuFarLastPumpMs=GpuFarFirstSubmitMs=GpuFarLastSubmitMs=-1.0;
            GpuFarFirstCompletionPollMs=GpuFarLastCompletionPollMs=-1.0;
            GpuFarPumpWallMs=GpuFarPumpMaxWallMs=GpuFarPumpGapMs=GpuFarPumpGapMaxMs=0.0;
            LastGpuFarPumpAt={};
            LastDrawEndAt={};
            GpuFarAudit.reset();bGpuFarAuditReported=false;
            GpuFarTilesSubmitted=0;GpuFarEnqueueWallSeconds=0.0;GpuFarGpuDispatchMs=0.0;
            GpuFarGpuDispatchMaxMs=0.0;GpuFarSubmissionSpanMs=0.0;GpuFarSubmissionSpanMaxMs=0.0;
            GpuFarLocalMemoryPeakBytes=0;GpuFarLocalMemoryBudgetBytes=0;
            bGpuFarFirstTileReported=false;
            GpuFarPoolReserveMs=0.0;
            GpuFarSubmissionBatches=0;GpuFarMaxTilesPerSubmissionBatch=0;
            DirtyBatchSet.clear();DesiredTiles.clear();DesiredTileSet.clear();CrossingWork.clear();
            DeferredLeaverTiles.clear();
            for(int32 I=0;I<=MaxLevels;++I)
            {DesiredByLevel[I]=0;GeneratedByLevel[I]=0;SamplesByLevel[I]=0;WorkerSecondsByLevel[I]=0.0;
             std::fill(std::begin(ProfileCyclesByLevel[I]),std::end(ProfileCyclesByLevel[I]),0);
             TileSecondsByLevel[I].clear();}
            TotalVertices=0;TotalIndices=0;
        }

        void StopGeneration(bool bCancel=false)
        {
            (void)bCancel;
            if(Streamer)Streamer->Stop();
            Streamer.reset();
            if(FullShapeBuildThread.joinable())FullShapeBuildThread.join();
            PendingFullShapeWorld.reset();
            bFullShapeBuildFinished.store(false,std::memory_order_relaxed);
            bFullShapeBuildSucceeded=false;
            FullShapeBuildError.clear();
            InitialNeighborhoodTiles.clear();
            bShapePipelineActive=false;
            bNeighborhoodReadyReported=false;
            bFarShapeStagePending=false;
            bGenerationRunning=false;
        }

        V3 Forward() const
        {return Normalize({std::sin(Yaw)*std::cos(Pitch),std::cos(Yaw)*std::cos(Pitch),std::sin(Pitch)});}

        bool IsDrawOnlyTile(const FVoxelTileKey& Key) const
        {
            return !bDrawOnlyTile||(Key.Level==DrawOnlyTileLod&&Key.Coord.X==DrawOnlyTileX
                &&Key.Coord.Y==DrawOnlyTileY&&Key.Coord.Z==DrawOnlyTileZ);
        }

        FIntVector CameraChunk() const
        {return StandaloneWorldMesher::CameraChunk(FVector(Position.X*4.0,Position.Y*4.0,Position.Z*4.0));}

        bool NeedsRecenter(const FIntVector& Center) const
        {
            return CameraChunk()!=Center;
        }

        bool MakeBuffer(ID3D11Buffer** Out,const void* Data,size_t Bytes,UINT Bind)
        {
            if(Bytes==0||Bytes>std::numeric_limits<UINT>::max())return false;
            D3D11_BUFFER_DESC Desc{};Desc.ByteWidth=static_cast<UINT>(Bytes);
            Desc.Usage=D3D11_USAGE_DEFAULT;Desc.BindFlags=Bind;
            D3D11_SUBRESOURCE_DATA Initial{};Initial.pSysMem=Data;
            return SUCCEEDED(Device->CreateBuffer(&Desc,&Initial,Out));
        }

        static void AppendPngU32(std::vector<uint8>& Out,uint32 Value)
        {
            Out.push_back(static_cast<uint8>(Value >> 24));
            Out.push_back(static_cast<uint8>(Value >> 16));
            Out.push_back(static_cast<uint8>(Value >> 8));
            Out.push_back(static_cast<uint8>(Value));
        }

        static uint32 PngCrc32(const uint8* Data,size_t Count)
        {
            uint32 Crc=0xffffffffu;
            for(size_t I=0;I<Count;++I)
            {
                Crc^=Data[I];
                for(int32 Bit=0;Bit<8;++Bit)Crc=(Crc>>1)^((Crc&1)?0xedb88320u:0u);
            }
            return Crc^0xffffffffu;
        }

        static void AppendPngChunk(std::vector<uint8>& Png,const char Type[4],const std::vector<uint8>& Data)
        {
            AppendPngU32(Png,static_cast<uint32>(Data.size()));
            const size_t CrcStart=Png.size();
            Png.insert(Png.end(),Type,Type+4);
            Png.insert(Png.end(),Data.begin(),Data.end());
            AppendPngU32(Png,PngCrc32(Png.data()+CrcStart,Png.size()-CrcStart));
        }

        bool SaveBackBufferPng(const std::string& Path)
        {
            ID3D11Texture2D* BackBuffer=nullptr;
            if(FAILED(SwapChain->GetBuffer(0,IID_PPV_ARGS(&BackBuffer))))return false;
            D3D11_TEXTURE2D_DESC Desc{};BackBuffer->GetDesc(&Desc);
            Desc.Usage=D3D11_USAGE_STAGING;Desc.BindFlags=0;Desc.CPUAccessFlags=D3D11_CPU_ACCESS_READ;Desc.MiscFlags=0;
            ID3D11Texture2D* Staging=nullptr;
            const HRESULT CreateResult=Device->CreateTexture2D(&Desc,nullptr,&Staging);
            if(FAILED(CreateResult)){BackBuffer->Release();return false;}
            Context->CopyResource(Staging,BackBuffer);BackBuffer->Release();
            D3D11_MAPPED_SUBRESOURCE Mapped{};
            if(FAILED(Context->Map(Staging,0,D3D11_MAP_READ,0,&Mapped))){Staging->Release();return false;}

            const size_t RowBytes=static_cast<size_t>(Desc.Width)*4;
            std::vector<uint8> Raw((RowBytes+1)*Desc.Height);
            for(UINT Y=0;Y<Desc.Height;++Y)
            {
                uint8* Row=Raw.data()+static_cast<size_t>(Y)*(RowBytes+1);
                Row[0]=0;
                std::memcpy(Row+1,static_cast<const uint8*>(Mapped.pData)+static_cast<size_t>(Y)*Mapped.RowPitch,RowBytes);
            }
            Context->Unmap(Staging,0);Staging->Release();

            std::vector<uint8> Zlib{0x78,0x01};
            uint32 AdlerA=1,AdlerB=0;
            for(const uint8 Byte:Raw){AdlerA=(AdlerA+Byte)%65521u;AdlerB=(AdlerB+AdlerA)%65521u;}
            for(size_t Offset=0;Offset<Raw.size();)
            {
                const uint16 Block=static_cast<uint16>(std::min<size_t>(65535,Raw.size()-Offset));
                const bool bFinal=Offset+Block==Raw.size();
                Zlib.push_back(bFinal?1:0);
                Zlib.push_back(static_cast<uint8>(Block));Zlib.push_back(static_cast<uint8>(Block>>8));
                const uint16 NotBlock=static_cast<uint16>(~Block);
                Zlib.push_back(static_cast<uint8>(NotBlock));Zlib.push_back(static_cast<uint8>(NotBlock>>8));
                Zlib.insert(Zlib.end(),Raw.begin()+Offset,Raw.begin()+Offset+Block);
                Offset+=Block;
            }
            AppendPngU32(Zlib,(AdlerB<<16)|AdlerA);

            std::vector<uint8> Png{137,80,78,71,13,10,26,10};
            std::vector<uint8> Ihdr;AppendPngU32(Ihdr,Desc.Width);AppendPngU32(Ihdr,Desc.Height);
            Ihdr.insert(Ihdr.end(),{8,6,0,0,0});AppendPngChunk(Png,"IHDR",Ihdr);
            AppendPngChunk(Png,"IDAT",Zlib);AppendPngChunk(Png,"IEND",{});
            std::ofstream File(Path,std::ios::binary);
            if(!File)return false;
            File.write(reinterpret_cast<const char*>(Png.data()),static_cast<std::streamsize>(Png.size()));
            return File.good();
        }

        bool Resize(UINT NewWidth,UINT NewHeight)
        {
            if(!Device||NewWidth==0||NewHeight==0)return false;
            Width=static_cast<int32>(NewWidth);Height=static_cast<int32>(NewHeight);
            if(RenderTarget){RenderTarget->Release();RenderTarget=nullptr;}
            if(DepthView){DepthView->Release();DepthView=nullptr;}
            if(SwapChain->ResizeBuffers(0,NewWidth,NewHeight,DXGI_FORMAT_UNKNOWN,0)!=S_OK)return false;
            ID3D11Texture2D* Back=nullptr;
            if(FAILED(SwapChain->GetBuffer(0,IID_PPV_ARGS(&Back))))return false;
            const HRESULT RtvResult=Device->CreateRenderTargetView(Back,nullptr,&RenderTarget);Back->Release();
            if(FAILED(RtvResult))return false;
            D3D11_TEXTURE2D_DESC DepthDesc{};DepthDesc.Width=NewWidth;DepthDesc.Height=NewHeight;
            DepthDesc.MipLevels=1;DepthDesc.ArraySize=1;DepthDesc.Format=DXGI_FORMAT_D24_UNORM_S8_UINT;
            DepthDesc.SampleDesc.Count=1;DepthDesc.Usage=D3D11_USAGE_DEFAULT;DepthDesc.BindFlags=D3D11_BIND_DEPTH_STENCIL;
            ID3D11Texture2D* Depth=nullptr;
            if(FAILED(Device->CreateTexture2D(&DepthDesc,nullptr,&Depth)))return false;
            const HRESULT DsvResult=Device->CreateDepthStencilView(Depth,nullptr,&DepthView);Depth->Release();
            return SUCCEEDED(DsvResult);
        }

        bool InitializeDevice()
        {
            DXGI_SWAP_CHAIN_DESC Desc{};Desc.BufferCount=1;Desc.BufferDesc.Format=DXGI_FORMAT_R8G8B8A8_UNORM;
            Desc.BufferUsage=DXGI_USAGE_RENDER_TARGET_OUTPUT;Desc.OutputWindow=Window;Desc.SampleDesc.Count=1;
            Desc.Windowed=TRUE;Desc.SwapEffect=DXGI_SWAP_EFFECT_DISCARD;
            D3D_FEATURE_LEVEL Feature{};
            if(FAILED(D3D11CreateDeviceAndSwapChain(nullptr,D3D_DRIVER_TYPE_HARDWARE,nullptr,0,nullptr,0,
                D3D11_SDK_VERSION,&Desc,&SwapChain,&Device,&Feature,&Context)))return false;
            const char* ShaderSource=R"(
cbuffer C:register(b0){float4x4 VP;float3 Camera;float FogStart;float3 Forward;float FogEnd;float3 LightDir;float Ambient;float4 Pad;}
struct I{float3 p:POSITION;float3 n:NORMAL;};
struct O{float4 p:SV_POSITION;float3 world:TEXCOORD0;float3 n:NORMAL;};
 O VS(I i){O o;float3 p=i.p*Pad.x;o.p=mul(float4(p,1),VP);o.world=p;o.n=normalize(i.n);return o;}
float4 PS(O i):SV_TARGET{
 float3 N=normalize(i.n);float3 V=normalize(Camera-i.world);float3 L=normalize(-LightDir);
 float Diffuse=saturate(dot(N,L));float Head=saturate(dot(N,V));
 float Dist=length(Camera-i.world);float Atten=1.0-saturate(Dist/95.0);
 float3 Rock=lerp(float3(0.30,0.22,0.16),float3(0.48,0.43,0.34),saturate(N.z*0.5+0.5));
// Keep the camera fill broad and low-amplitude.  The old squared, high-intensity
// lobe made the 25 cm density-lattice normal variation read as horizontal bands.
Head=Head*Atten;
float3 Lit=Rock*(Ambient+0.88*Diffuse)+float3(0.50,0.69,0.82)*Head*0.75;
 float Fog=saturate((Dist-FogStart)/(FogEnd-FogStart));
 Lit=lerp(Lit,float3(0.10,0.14,0.19),Fog);return float4(Lit,1);
})";
            ID3DBlob* VsCode=nullptr;ID3DBlob* PsCode=nullptr;ID3DBlob* Errors=nullptr;
            HRESULT Hr=D3DCompile(ShaderSource,std::strlen(ShaderSource),nullptr,nullptr,nullptr,"VS","vs_5_0",0,0,&VsCode,&Errors);
            if(FAILED(Hr)){if(Errors)std::fprintf(stderr,"Vertex shader compile: %s\n",static_cast<const char*>(Errors->GetBufferPointer()));if(Errors)Errors->Release();return false;}
            Hr=D3DCompile(ShaderSource,std::strlen(ShaderSource),nullptr,nullptr,nullptr,"PS","ps_5_0",0,0,&PsCode,&Errors);
            if(FAILED(Hr)){if(Errors)std::fprintf(stderr,"Pixel shader compile: %s\n",static_cast<const char*>(Errors->GetBufferPointer()));if(Errors)Errors->Release();VsCode->Release();return false;}
            Hr=Device->CreateVertexShader(VsCode->GetBufferPointer(),VsCode->GetBufferSize(),nullptr,&VertexShader);
            if(SUCCEEDED(Hr))Hr=Device->CreatePixelShader(PsCode->GetBufferPointer(),PsCode->GetBufferSize(),nullptr,&PixelShader);
            D3D11_INPUT_ELEMENT_DESC Elements[]={{"POSITION",0,DXGI_FORMAT_R32G32B32_FLOAT,0,0,D3D11_INPUT_PER_VERTEX_DATA,0},
                {"NORMAL",0,DXGI_FORMAT_R32G32B32_FLOAT,0,12,D3D11_INPUT_PER_VERTEX_DATA,0}};
            if(SUCCEEDED(Hr))Hr=Device->CreateInputLayout(Elements,2,VsCode->GetBufferPointer(),VsCode->GetBufferSize(),&InputLayout);
            VsCode->Release();PsCode->Release();if(FAILED(Hr))return false;
            D3D11_BUFFER_DESC ConstantDesc{};ConstantDesc.ByteWidth=sizeof(FConstants);ConstantDesc.Usage=D3D11_USAGE_DEFAULT;
            ConstantDesc.BindFlags=D3D11_BIND_CONSTANT_BUFFER;
            if(FAILED(Device->CreateBuffer(&ConstantDesc,nullptr,&Constants)))return false;
            D3D11_RASTERIZER_DESC Raster{};Raster.FillMode=D3D11_FILL_SOLID;Raster.CullMode=D3D11_CULL_NONE;
            Raster.DepthClipEnable=TRUE;if(FAILED(Device->CreateRasterizerState(&Raster,&SolidRasterizer)))return false;
            Raster.FillMode=D3D11_FILL_WIREFRAME;if(FAILED(Device->CreateRasterizerState(&Raster,&WireRasterizer)))return false;
            if(!Resize(Width,Height))return false;
            if(bGpuFarEnabled)
            {
                std::string GpuInitError;
                const auto GpuPipelineStart=std::chrono::steady_clock::now();
                if(!VoxelForgeGpuFarViewer::InitializeFarPipeline(Device,Context,GpuFarShaderPath,
                    GpuInitError))
                {
                    std::fprintf(stderr,"[GpuPipelineInitError] %s\n",GpuInitError.c_str());
                    return false;
                }
                std::printf("[ColdTimeline] stage=gpu_pipeline_init end_ms=%.3f duration_ms=%.3f prewarm=%s\n",
                    std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count(),
                    std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-GpuPipelineStart).count(),
                    bGpuFarShaderPrewarm?(bGpuFarWarmupResourcesOnly?"async_resources_only":"async_dispatch")
                        :"disabled");
            }
            return true;
        }

        void StartGpuShaderWarmupAsync()
        {
            if(!bGpuFarShaderPrewarm||GpuShaderWarmupThread.joinable())return;
            bGpuShaderWarmupQueueSucceeded=false;GpuShaderWarmupQueueError.clear();
            GpuShaderWarmupThread=std::thread([this]()
            {
                bool bNormalBasePriority=false;
                if(bGpuFarWarmupNormalPriority)
                {
                    // The process stays BelowNormal; HIGHEST relative priority raises this
                    // worker thread to a normal base priority on Windows.
                    bNormalBasePriority=SetThreadPriority(GetCurrentThread(),THREAD_PRIORITY_HIGHEST)!=0;
                    std::printf("[GpuShaderWarmupThread] priority=%s process=BelowNormal thread_relative=Highest\n",
                        bNormalBasePriority?"Normal":"set_failed");
                    std::fflush(stdout);
                }
                bGpuShaderWarmupQueueSucceeded=VoxelForgeGpuFarViewer::QueueFarShaderWarmup(
                    Device,Context,GpuFarShaderPath,!bGpuFarWarmupResourcesOnly,GpuWarmupEntryFilter,
                    GpuShaderWarmupQueueError);
            });
            std::printf("[GpuShaderWarmup] queue_thread=started overlap=%s\n",
                bGpuFarWarmupBeforeWorkers?"none_before_generation_workers":"CPU_shape_compile");
            std::fflush(stdout);
        }

        size_t PrintParity()
        {
            if(bParityChecked||!World)return 0;
            const auto ParityStart=std::chrono::steady_clock::now();
            bParityChecked=true;size_t MismatchCount=0;
            std::map<std::string,size_t> Kinds;
            for(size_t I=0;I<World->DensitySamples.size();++I)
            {
                const FStandaloneParitySample& S=World->DensitySamples[I];
                const float Actual=World->Generator->GetDensityAt(static_cast<float>(S.PositionVoxels.X),
                    static_cast<float>(S.PositionVoxels.Y),static_cast<float>(S.PositionVoxels.Z));
                ++Kinds[S.Kind];
                if(std::bit_cast<uint32>(Actual)!=std::bit_cast<uint32>(S.PluginDensity))
                {
                    ++MismatchCount;
                    std::fprintf(stderr,"PARITY MISMATCH %s strate=%d voxels=(%.6f,%.6f,%.6f) plugin=%.9g standalone=%.9g\n",
                        S.Kind.c_str(),S.StrateIndex,S.PositionVoxels.X,S.PositionVoxels.Y,S.PositionVoxels.Z,S.PluginDensity,Actual);
                }
            }
            std::printf("Density parity: %zu/%zu bit-identical; mismatches=%zu\n",
                World->DensitySamples.size()-MismatchCount,World->DensitySamples.size(),MismatchCount);
            for(const auto& [Kind,Count]:Kinds)std::printf("  %-24s %zu\n",Kind.c_str(),Count);
            std::printf("[DiagnosticProfile] stage=density_parity samples=%zu duration_ms=%.3f\n",
                World->DensitySamples.size(),std::chrono::duration<double,std::milli>(
                    std::chrono::steady_clock::now()-ParityStart).count());
            std::fflush(stdout);
            return MismatchCount;
        }

        bool ConfigureWorld(std::string& Error)
        {
            const auto ConfigureStart=std::chrono::steady_clock::now();
            ShapeWorld.reset();
            Snapshot={};
            World=std::make_shared<FStandaloneOwnerWorld>();
            const auto ResolveStart=std::chrono::steady_clock::now();
            if(!StandaloneOwnerLoader::Resolve(OwnerJsonRoot,*World,Error))return false;
            const auto ResolveEnd=std::chrono::steady_clock::now();
            std::printf("[ConfigureProfile] stage=owner_resolve duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(ResolveEnd-ResolveStart).count());
            if(!bParityChecked&&Seed==World->Seed&&!bRoundTunnels)
                std::printf("[DiagnosticDeferred] density_parity=--parity-only samples=%zu\n",
                    World->DensitySamples.size());
            const auto ClipOverrideStart=std::chrono::steady_clock::now();
            const int32 ExportClipRadius=World->ClipRadius;
            World->ClipRadius=ClipRadius;
            World->Settings->ClipRadius=ClipRadius;
            if(!bPrintedClipOverride)
            {
                bPrintedClipOverride=true;
                const double FullDetailWidth=static_cast<double>(2*ClipRadius+2)
                    *CHUNK_SIZE*VOXEL_SIZE/100.0;
                std::printf("Standalone clip override: exported radius=%d (112 m cube) -> radius=%d (%.0f m LOD0 cube), voxel size=25 cm.\n",
                    ExportClipRadius,ClipRadius,FullDetailWidth);
            }
            std::printf("[ConfigureProfile] stage=clip_override duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ClipOverrideStart).count());
            const auto DensitySetupStart=std::chrono::steady_clock::now();
            if(!PrepareStandaloneDensityWorld(*World,Seed,bRoundTunnels,
                bRoundTunnelEnhanced,Error))return false;
            std::printf("[ConfigureProfile] stage=density_world_setup duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-DensitySetupStart).count());
            if(MaxClipLevelOverride>=0)
            {
                World->MaxClipLevel=MaxClipLevelOverride;
                World->Settings->MaxClipLevel=MaxClipLevelOverride;
                std::printf("Standalone max clip level override applied after world reload: max_level=%d.\n",
                    MaxClipLevelOverride);
            }
            // Compile one world field before streaming. The streamer already prioritizes the
            // player's neighborhood, so a second near-only field would only duplicate work.
            const auto ShapeSetupStart=std::chrono::steady_clock::now();
            if(bShapeWorld&&!BuildShapeWorldForCamera(Error,true,false,true))return false;
            std::printf("[ConfigureProfile] stage=shape_world_request duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ShapeSetupStart).count());
            const auto CacheStart=std::chrono::steady_clock::now();
            LoadMeshCache();
            std::printf("[ConfigureProfile] stage=mesh_cache duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-CacheStart).count());
            std::printf("[ConfigureProfile] stage=configure_total duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ConfigureStart).count());
            return true;
        }

        bool BuildShapeWorldForCamera(std::string& Error,bool bIncludeFarWorld=false,
                                      bool bNeighborhoodOnly=false,
                                      bool bDeferWalkabilityValidation=false)
        {
            const auto RequestSetupStart=std::chrono::steady_clock::now();
            const FVector CameraVoxels(Position.X*4.0,Position.Y*4.0,Position.Z*4.0);
            const FStandaloneDesiredTileSet Desired=StandaloneWorldMesher::BuildDesiredSet(*World,CameraVoxels);
            FIntVector Min(INT32_MAX,INT32_MAX,INT32_MAX),Max(INT32_MIN,INT32_MIN,INT32_MIN);
            FIntVector ClipMin(INT32_MAX,INT32_MAX,INT32_MAX),ClipMax(INT32_MIN,INT32_MIN,INT32_MIN);
            uint64 FineTiles=0;
            InitialNeighborhoodTiles.clear();
            const FIntVector CenterChunk=Desired.CenterChunk;
            for(const FVoxelTileKey& Tile:Desired.Tiles)
            {
                const FIntVector Origin=Tile.OriginVoxels();
                const int32 Extent=Tile.ExtentVoxels();
                ClipMin.X=std::min(ClipMin.X,Origin.X);ClipMin.Y=std::min(ClipMin.Y,Origin.Y);ClipMin.Z=std::min(ClipMin.Z,Origin.Z);
                ClipMax.X=std::max(ClipMax.X,Origin.X+Extent);ClipMax.Y=std::max(ClipMax.Y,Origin.Y+Extent);ClipMax.Z=std::max(ClipMax.Z,Origin.Z+Extent);
                if(bNeighborhoodOnly)
                {
                    if(Tile.Level!=0||std::abs(Tile.Coord.X-CenterChunk.X)>1
                        ||std::abs(Tile.Coord.Y-CenterChunk.Y)>1
                        ||std::abs(Tile.Coord.Z-CenterChunk.Z)>1)continue;
                    InitialNeighborhoodTiles.push_back(Tile);
                }
                else if(Tile.Level>=World->FullResClipLevels)continue;
                Min.X=std::min(Min.X,Origin.X);Min.Y=std::min(Min.Y,Origin.Y);Min.Z=std::min(Min.Z,Origin.Z);
                Max.X=std::max(Max.X,Origin.X+Extent);Max.Y=std::max(Max.Y,Origin.Y+Extent);Max.Z=std::max(Max.Z,Origin.Z+Extent);
                ++FineTiles;
            }
            if(!FineTiles){Error="no full-detail tiles are available for shape compilation";return false;}
            if(bNeighborhoodOnly)
            {
                if(InitialNeighborhoodTiles.size()!=27)
                {Error="the 3x3x3 player neighborhood does not contain 27 LOD0 tiles";return false;}
                constexpr int32 NearBuildPad=64;
                Min.X-=NearBuildPad;Min.Y-=NearBuildPad;Min.Z-=NearBuildPad;
                Max.X+=NearBuildPad;Max.Y+=NearBuildPad;Max.Z+=NearBuildPad;
            }
            const auto DesiredSetupEnd=std::chrono::steady_clock::now();
            std::printf("[ConfigureProfile] stage=desired_tiles_and_bounds tiles=%llu duration_ms=%.3f\n",
                static_cast<unsigned long long>(FineTiles),std::chrono::duration<double,std::milli>(
                    DesiredSetupEnd-RequestSetupStart).count());
            const auto SnapshotStart=std::chrono::steady_clock::now();
            Snapshot=VoxelForgeCore::MakeStandaloneSnapshotFromParsedJson(OwnerJsonRoot,Seed,ClipRadius,
                MaxClipLevelOverride,bRoundTunnels,bRoundTunnelEnhanced,Error);
            if(!Snapshot)
            {if(Error.empty())Error="the standalone adapter could not make a resolved core snapshot";return false;}
            ShapeWorld=std::make_unique<VoxelForgeCore::World>(Snapshot);
            const int32 ShapeWorkers=bNeighborhoodOnly?(WorkerCount>=8?2:1):WorkerCount;
            VoxelForgeCore::ShapeBuildRequest Request;
            Request.minVoxel={Min.X,Min.Y,Min.Z};Request.maxVoxel={Max.X,Max.Y,Max.Z};
            Request.clipMinVoxel={ClipMin.X,ClipMin.Y,ClipMin.Z};
            Request.clipMaxVoxel={ClipMax.X,ClipMax.Y,ClipMax.Z};
            Request.workerLimit=static_cast<uint32>(ShapeWorkers);Request.detail=ShapeDetailVariant;
            Request.includeFarWorld=bIncludeFarWorld;Request.roundTunnels=bRoundTunnels;
            Request.enhancedTunnelWalls=bRoundTunnelEnhanced;
            Request.profileDetailBand=bProfileDetailBand;
            Request.validatePassageRoutes=!bDeferWalkabilityValidation;
            if(bDeferWalkabilityValidation)
                std::printf("[DiagnosticDeferred] walkability=full_headless_or_route_audit workers=8 samples=5642\n");
            const auto ShapeCompileStart=std::chrono::steady_clock::now();
            std::printf("[ConfigureProfile] stage=snapshot_and_build_request duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(ShapeCompileStart-SnapshotStart).count());
            std::printf("[ColdTimeline] stage=shape_compile begin_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(ShapeCompileStart-ColdProcessStart).count());
            if(!ShapeWorld->compile(Request,CoreScheduler,Error))
            {ShapeWorld.reset();return false;}
            const auto ShapeCompileEnd=std::chrono::steady_clock::now();
            std::printf("[ColdTimeline] stage=shape_compile end_ms=%.3f duration_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(ShapeCompileEnd-ColdProcessStart).count(),
                std::chrono::duration<double,std::milli>(ShapeCompileEnd-ShapeCompileStart).count());
            bShapePipelineActive=bNeighborhoodOnly;
            bNeighborhoodReadyReported=false;
            bFarShapeStagePending=bShapeWorld&&!bNeighborhoodOnly&&!ShapeWorld->hasFarWorld();
            std::printf("[ShapeRegion] fine_tiles=%llu bounds=(%d,%d,%d)..(%d,%d,%d) clip=(%d,%d,%d)..(%d,%d,%d)\n",
                static_cast<unsigned long long>(FineTiles),Min.X,Min.Y,Min.Z,Max.X,Max.Y,Max.Z,
                ClipMin.X,ClipMin.Y,ClipMin.Z,ClipMax.X,ClipMax.Y,ClipMax.Z);
            return true;
        }

        bool StartFullShapeBuild(const std::vector<FVoxelTileKey>& FullDesired,std::string& Error)
        {
            FIntVector FineMin(INT32_MAX,INT32_MAX,INT32_MAX),FineMax(INT32_MIN,INT32_MIN,INT32_MIN);
            FIntVector ClipMin(INT32_MAX,INT32_MAX,INT32_MAX),ClipMax(INT32_MIN,INT32_MIN,INT32_MIN);
            for(const FVoxelTileKey& Tile:FullDesired)
            {
                const FIntVector Origin=Tile.OriginVoxels();const int32 Extent=Tile.ExtentVoxels();
                ClipMin.X=std::min(ClipMin.X,Origin.X);ClipMin.Y=std::min(ClipMin.Y,Origin.Y);ClipMin.Z=std::min(ClipMin.Z,Origin.Z);
                ClipMax.X=std::max(ClipMax.X,Origin.X+Extent);ClipMax.Y=std::max(ClipMax.Y,Origin.Y+Extent);ClipMax.Z=std::max(ClipMax.Z,Origin.Z+Extent);
                if(Tile.Level>=World->FullResClipLevels)continue;
                FineMin.X=std::min(FineMin.X,Origin.X);FineMin.Y=std::min(FineMin.Y,Origin.Y);FineMin.Z=std::min(FineMin.Z,Origin.Z);
                FineMax.X=std::max(FineMax.X,Origin.X+Extent);FineMax.Y=std::max(FineMax.Y,Origin.Y+Extent);FineMax.Z=std::max(FineMax.Z,Origin.Z+Extent);
            }
            if(FineMin.X==INT32_MAX||ClipMin.X==INT32_MAX)
            {Error="the full shape build has no desired detail or clipmap tiles";return false;}
            PendingFullShapeWorld=std::make_unique<VoxelForgeCore::World>(Snapshot);
            FullShapeBuildError.clear();bFullShapeBuildSucceeded=false;
            bFullShapeBuildFinished.store(false,std::memory_order_relaxed);
            const int32 FullWorkers=WorkerCount>=8?6:WorkerCount;
            FullShapeBuildStart=std::chrono::steady_clock::now();
            FullShapeBuildThread=std::thread([this,FineMin,FineMax,ClipMin,ClipMax,FullWorkers]()
            {
                VoxelForgeCore::ShapeBuildRequest Request;
                Request.minVoxel={FineMin.X,FineMin.Y,FineMin.Z};
                Request.maxVoxel={FineMax.X,FineMax.Y,FineMax.Z};
                Request.clipMinVoxel={ClipMin.X,ClipMin.Y,ClipMin.Z};
                Request.clipMaxVoxel={ClipMax.X,ClipMax.Y,ClipMax.Z};
                Request.workerLimit=static_cast<uint32>(FullWorkers);Request.detail=ShapeDetailVariant;
                Request.includeFarWorld=true;Request.roundTunnels=bRoundTunnels;
                Request.enhancedTunnelWalls=bRoundTunnelEnhanced;
                Request.profileDetailBand=bProfileDetailBand;
                Request.validatePassageRoutes=false;
                bFullShapeBuildSucceeded=PendingFullShapeWorld->compile(
                    Request,CoreScheduler,FullShapeBuildError);
                bFullShapeBuildFinished.store(true,std::memory_order_release);
            });
            std::printf("[ShapePipelineBuildStart] neighborhood_tiles=%zu near_mesh_workers=%d full_shape_workers=%d\n",
                InitialNeighborhoodTiles.size(),WorkerCount>=8?2:1,FullWorkers);
            std::fflush(stdout);
            return true;
        }

        FStandaloneMeshCacheMetadata MakeMeshCacheMetadata()const
        {
            FStandaloneMeshCacheMetadata Metadata;
            Metadata.SourceDataHash=HashStandaloneFile(DataPath);
            Metadata.Seed=Seed;Metadata.ClipRadius=ClipRadius;
            Metadata.FullResClipLevels=World->FullResClipLevels;
            Metadata.CoarseTileCells=World->CoarseTileCells;
            Metadata.StrateContentCutMinLevel=World->StrateContentCutMinLevel;
            Metadata.SkirtCells=World->SkirtCells;
            Metadata.LODOctaveDrop=World->LODOctaveDrop;
            Metadata.bGenerateSkirts=World->bGenerateSkirts?1:0;
            Metadata.AlgorithmVersion=FStandaloneMeshCache::CurrentVersion;
            Metadata.TunnelMode=bRoundTunnels?(bRoundTunnelEnhanced?2u:1u):0u;
            return Metadata;
        }

        void LoadMeshCache()
        {
            MeshCache.Clear();
            if(bForceGenerate||bShapeWorld||!World)return;
            const char* ModeSuffix=!bRoundTunnels?"_base":
                (bRoundTunnelEnhanced?"_round_detailed":"_round_legacy");
            const std::filesystem::path CachePath=std::filesystem::path(DataPath).parent_path()
                /(std::string("VoxelForge_seed")+std::to_string(Seed)+"_r"
                    +std::to_string(ClipRadius)+ModeSuffix+".vfmesh");
            MeshCachePath=CachePath.string();
            const auto Begin=std::chrono::steady_clock::now();
            std::string CacheError;
            if(MeshCache.Load(MeshCachePath,MakeMeshCacheMetadata(),CacheError))
            {
                const double Seconds=std::chrono::duration<double>(
                    std::chrono::steady_clock::now()-Begin).count();
                std::printf("[MeshCache] loaded %zu tiles, %.2f MiB in %.3f s from %s\n",
                    MeshCache.Num(),static_cast<double>(MeshCache.GetBytesLoaded())/1048576.0,
                    Seconds,MeshCachePath.c_str());
                std::fflush(stdout);
            }
            else if(CacheError!="cache file is missing or unreadable")
                std::printf("[MeshCache] ignored %s: %s\n",MeshCachePath.c_str(),CacheError.c_str());
        }

        void StartGeneration()
        {
            StopGeneration();
            ReleaseMeshes();
            GenerationStart=std::chrono::steady_clock::now();
            std::printf("[ColdTimeline] stage=generation_setup begin_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(GenerationStart-ColdProcessStart).count());
            std::string Error;
            if(!ConfigureWorld(Error))
            {std::fprintf(stderr,"World load failed: %s\n",Error.c_str());MessageBoxA(Window,Error.c_str(),"VoxelForge viewer",MB_ICONERROR);return;}
            std::printf("[ColdTimeline] stage=generation_setup configured_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count());
            const FVector CameraVoxels(Position.X*4.0,Position.Y*4.0,Position.Z*4.0);
            BuildTargetChunk=StandaloneWorldMesher::CameraChunk(CameraVoxels);
            bFullDetailReported=false;bInitialClipmapReported=false;bWholeViewPresented=false;bAutoCaptureWritten=false;
            if(bProfileAttribution)
            {
                VoxelDensityProfile::SetMode(VoxelDensityProfile::EMode::Attribution,128);
                VoxelDensityProfile::CalibrateTimer();
                VoxelDensityProfile::Reset();
                std::printf("Density stage profiling enabled (Attribution, sample interval 128).\n");
            }
            const FStandaloneDesiredTileSet Desired=StandaloneWorldMesher::BuildDesiredSet(*World,CameraVoxels);
            DesiredTiles=Desired.Tiles;
            if(ShapeWorld)
            {
                const auto CoreTiles=ShapeWorld->desiredTiles({CameraVoxels.X,CameraVoxels.Y,CameraVoxels.Z});
                DesiredTiles.clear();DesiredTiles.reserve(CoreTiles.size());
                for(const auto& Key:CoreTiles)
                    DesiredTiles.emplace_back(FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
            }
            if(bCaptureOnlyTile)
            {
                const size_t Before=DesiredTiles.size();
                DesiredTiles.erase(std::remove_if(DesiredTiles.begin(),DesiredTiles.end(),[&](const FVoxelTileKey& Key)
                {
                    return Key.Level!=CaptureOnlyTileLod||Key.Coord.X!=CaptureOnlyTileX
                        ||Key.Coord.Y!=CaptureOnlyTileY||Key.Coord.Z!=CaptureOnlyTileZ;
                }),DesiredTiles.end());
                std::printf("[CaptureOnlyTile] key=(%d,%d,%d,lod%d) before=%zu selected=%zu\n",
                    CaptureOnlyTileX,CaptureOnlyTileY,CaptureOnlyTileZ,CaptureOnlyTileLod,
                    Before,DesiredTiles.size());
                if(DesiredTiles.empty())
                {
                    Error="the requested capture-only tile is not in the desired clipmap";
                    std::fprintf(stderr,"World load failed: %s\n",Error.c_str());
                    return;
                }
            }
            if(CaptureTileLimit>0&&CapturePath.size())
            {
                const size_t Before=DesiredTiles.size();
                const V3 F=Forward(),R=Normalize(Cross(F,{0,0,1})),U=Normalize(Cross(R,F));
                const double TanHalfFov=std::tan(1.04719755*0.5);
                const double Aspect=static_cast<double>(Width)/std::max(1,Height);
                struct FCaptureTileRank { FVoxelTileKey Tile; double Depth=0.0; };
                std::vector<FCaptureTileRank> Visible;Visible.reserve(DesiredTiles.size());
                for(const FVoxelTileKey& Tile:DesiredTiles)
                {
                    const FIntVector Origin=Tile.OriginVoxels();
                    const double HalfExtent=Tile.ExtentVoxels()*0.8660254037844386;
                    const double DX=Origin.X+Tile.ExtentVoxels()*0.5-CameraVoxels.X;
                    const double DY=Origin.Y+Tile.ExtentVoxels()*0.5-CameraVoxels.Y;
                    const double DZ=Origin.Z+Tile.ExtentVoxels()*0.5-CameraVoxels.Z;
                    const double Depth=DX*F.X+DY*F.Y+DZ*F.Z;
                    if(Depth+HalfExtent<0.0)continue;
                    const double LateralX=std::abs(DX*R.X+DY*R.Y+DZ*R.Z);
                    const double LateralY=std::abs(DX*U.X+DY*U.Y+DZ*U.Z);
                    const double ForwardReach=std::max(1.0,Depth+HalfExtent);
                    if(LateralX>ForwardReach*TanHalfFov*Aspect+HalfExtent
                        ||LateralY>ForwardReach*TanHalfFov+HalfExtent)continue;
                    Visible.push_back({Tile,std::max(0.0,Depth)});
                }
                std::stable_sort(Visible.begin(),Visible.end(),[](const auto& A,const auto& B)
                    {return A.Depth<B.Depth;});
                if(Visible.size()>static_cast<size_t>(CaptureTileLimit))
                    Visible.resize(static_cast<size_t>(CaptureTileLimit));
                DesiredTiles.clear();DesiredTiles.reserve(Visible.size());
                for(const FCaptureTileRank& Ranked:Visible)DesiredTiles.push_back(Ranked.Tile);
                std::printf("[CaptureTileLimit] frustum_tiles=%zu selected=%zu original_desired=%zu limit=%d; reference capture only\n",
                    Visible.size(),DesiredTiles.size(),Before,CaptureTileLimit);
            }
            DesiredTileSet.clear();DesiredTileSet.reserve(DesiredTiles.size()*2+1);
            for(const FVoxelTileKey& Key:DesiredTiles)DesiredTileSet.insert(Key);
            if(bDrawOnlyTile)
            {
                const FVoxelTileKey DrawKey(FIntVector(DrawOnlyTileX,DrawOnlyTileY,DrawOnlyTileZ),
                    DrawOnlyTileLod);
                if(!DesiredTileSet.contains(DrawKey))
                {
                    Error="the requested draw-only tile is not in the full desired clipmap";
                    std::fprintf(stderr,"World load failed: %s\n",Error.c_str());
                    PostMessageA(Window,WM_CLOSE,0,0);
                    return;
                }
                std::printf("[DrawOnlyTile] key=(%d,%d,%d,lod%d) desired_targets=%zu filter=draw_only\n",
                    DrawOnlyTileX,DrawOnlyTileY,DrawOnlyTileZ,DrawOnlyTileLod,DesiredTiles.size());
            }
            for(int32 I=0;I<=MaxLevels;++I)DesiredByLevel[I]=0;
            for(const FVoxelTileKey& Tile:DesiredTiles)++DesiredByLevel[Tile.Level];
            const int32 StreamWorkers=bShapePipelineActive?(WorkerCount>=8?2:1):WorkerCount;
            Streamer=std::make_unique<FStandaloneStreamingGenerator>(*World,StreamWorkers,false,
                (!bShapeWorld&&MeshCache.IsLoaded())?&MeshCache:nullptr,
                bFarShapeStagePending?World->FullResClipLevels-1:-1,
                false,false,false,false,ShapeWorld.get());
            if(bGpuFarEnabled&&ShapeWorld&&ShapeWorld->hasFarWorld())
            {
                ScheduleGpuFarTiles(DesiredTiles);
                if(bGpuFarAuditEnabled)
                {
                    std::vector<FVoxelTileKey> AuditKeys;AuditKeys.reserve(GpuFarEligibleTiles.size());
                    for(const FVoxelTileKey& Key:DesiredTiles)
                        if(Key.Level>=6&&GpuFarEligibleTiles.contains(Key))AuditKeys.push_back(Key);
                    GpuFarAudit=std::make_unique<FFarSurvivalAudit>(*World,*ShapeWorld,AuditKeys,false);
                }
            }
            if(bShapePipelineActive)Streamer->SetDesiredTileFilter(InitialNeighborhoodTiles);
            else if((CaptureTileLimit>0&&CapturePath.size())||bCaptureOnlyTile)
                Streamer->SetDesiredTileFilter(DesiredTiles);
            const FStandaloneStreamingUpdate Update=Streamer->Start(CameraVoxels,{},CrossingIndex);
            if(bShapePipelineActive&&Update.DesiredTiles!=static_cast<int32>(InitialNeighborhoodTiles.size()))
            {
                StopGeneration();
                Error="the initial neighborhood streamer did not accept all 27 tiles";
                std::fprintf(stderr,"World load failed: %s\n",Error.c_str());
                return;
            }
            FCrossingWork& InitialWork=CrossingWork[CrossingIndex];
            InitialWork.SubmittedJobs=Update.NewJobs;InitialWork.PendingJobs=Update.NewJobs;
            InitialWork.StartTime=GenerationStart;
            bGenerationRunning=true;bRebuildPending=false;
            std::printf("[ColdTimeline] stage=cpu_stream_start submitted_ms=%.3f cpu_jobs=%d gpu_queued=%zu\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count(),
                Update.NewJobs,GpuFarQueue.size());
            const char* TunnelMode=!bRoundTunnels?"":(bRoundTunnelEnhanced
                ?" with detailed round/winding tunnels":" with legacy round/winding tunnels");
            std::printf("Streaming seed %d%s%s: %d desired tiles, %d full-detail tiles, %d workers; upload budget %d batches / %.1f ms per frame.\n",
                Seed,TunnelMode,bShapeWorld?" [shape intervals]":"",Update.DesiredTiles,
                Update.FullResTiles,Streamer->GetWorkerCount(),MaxBatchUploadsPerFrame,BatchUploadBudgetMilliseconds);
            if(bShapePipelineActive)
            {
                std::string ShapeError;
                if(!StartFullShapeBuild(DesiredTiles,ShapeError))
                {
                    StopGeneration();
                    std::fprintf(stderr,"World load failed: %s\n",ShapeError.c_str());
                    return;
                }
            }
            std::fflush(stdout);
        }

        void RequestSeed(int32 Delta)
        {
            const uint32 Next=static_cast<uint32>(Seed)+static_cast<uint32>(Delta);
            Seed=static_cast<int32>(Next);bRebuildPending=true;
            std::printf("Seed changed to %d; replacing the current streaming world.\n",Seed);
        }

        void StartFarShapeGeneration()
        {
            if(!bFarShapeStagePending||!ShapeWorld||ShapeWorld->hasFarWorld()||!Streamer)return;
            ShapeWorld->buildFarWorld();
            bFarShapeStagePending=false;
            Streamer->Stop();
            std::vector<FVoxelTileKey> ResidentKeys;ResidentKeys.reserve(ResidentTiles.size());
            for(const auto& Pair:ResidentTiles)ResidentKeys.push_back(Pair.first);
            Streamer.reset();
            Streamer=std::make_unique<FStandaloneStreamingGenerator>(*World,WorkerCount,false,nullptr,-1,
                false,false,false,false,ShapeWorld.get());
            if(bGpuFarEnabled)
            {
                ScheduleGpuFarTiles(DesiredTiles);
                if(bGpuFarAuditEnabled&&World)
                {
                    std::vector<FVoxelTileKey> AuditKeys;AuditKeys.reserve(GpuFarEligibleTiles.size());
                    for(const FVoxelTileKey& Key:DesiredTiles)
                        if(Key.Level>=6&&GpuFarEligibleTiles.contains(Key))AuditKeys.push_back(Key);
                    GpuFarAudit=std::make_unique<FFarSurvivalAudit>(*World,*ShapeWorld,AuditKeys,false);
                }
            }
            const FVector CameraVoxels(Position.X*4.0,Position.Y*4.0,Position.Z*4.0);
            ++CrossingIndex;
            const FStandaloneStreamingUpdate Update=Streamer->Start(CameraVoxels,ResidentKeys,CrossingIndex);
            FCrossingWork& Work=CrossingWork[CrossingIndex];
            Work.SubmittedJobs=Update.NewJobs;Work.PendingJobs=Update.NewJobs;
            Work.StartTime=std::chrono::steady_clock::now();
            bGenerationRunning=true;
            std::printf("[FarStreamStart] desired=%d retained=%d new_jobs=%d workers=%d\n",
                Update.DesiredTiles,Update.RetainedResidentTiles,Update.NewJobs,WorkerCount);
            std::fflush(stdout);
        }

        void RecordGpuCrossingTileReady(const FVoxelTileKey& Key)
        {
            const auto ReadyAt=std::chrono::steady_clock::now();
            for(auto& Pair:CrossingWork)
            {
                FCrossingWork& Work=Pair.second;
                if(Work.bHasNearestNewTile&&Work.NearestNewTile==Key
                    &&Work.NearestReadySeconds<0.0&&Work.StartTime.time_since_epoch().count()!=0)
                    Work.NearestReadySeconds=std::chrono::duration<double>(ReadyAt-Work.StartTime).count();
            }
        }

        bool IsRenderedTileReady(const FVoxelTileKey& Key)
        {
            if(bGpuFarEnabled&&ShapeWorld&&ShapeWorld->hasFarWorld()
                &&GpuFarEligibleTiles.contains(Key))
            {
                auto It=GpuFarTiles.find(Key);
                if(It==GpuFarTiles.end())return false;
                FGpuFarTile& Tile=It->second;
                if(Tile.BatchIndex!=UINT32_MAX&&Tile.BatchIndex<GpuFarBatches.size())
                {
                    FGpuFarBatch& Batch=GpuFarBatches[Tile.BatchIndex];
                    if(!Batch.bGpuComplete&&Batch.CompletionEvent&&Context
                        &&Context->GetData(Batch.CompletionEvent,nullptr,0,D3D11_ASYNC_GETDATA_DONOTFLUSH)==S_OK)
                    {
                        Batch.bGpuComplete=true;
                        const auto PollAt=std::chrono::steady_clock::now();
                        const double PollMs=std::chrono::duration<double,std::milli>(PollAt-ColdProcessStart).count();
                        if(GpuFarFirstCompletionPollMs<0.0)GpuFarFirstCompletionPollMs=PollMs;
                        GpuFarLastCompletionPollMs=PollMs;++GpuFarCompletionPolls;
                        Batch.CompletionEvent->Release();Batch.CompletionEvent=nullptr;
                    }
                    if(Batch.bGpuComplete&&!Batch.bTimingResolved)
                    {
                        if(VoxelForgeGpuFarViewer::ResolveGpuTiming(Context,Batch.TimingQueries,
                            Batch.GpuDispatchMs,&Batch.TimestampFrequency))
                        {
                            Batch.bTimingResolved=true;
                            GpuFarSubmissionSpanMs+=Batch.GpuDispatchMs;
                            GpuFarSubmissionSpanMaxMs=std::max(GpuFarSubmissionSpanMaxMs,Batch.GpuDispatchMs);
                        }
                    }
                    if(Batch.bGpuComplete&&Batch.bTimingResolved&&!Tile.bTimingResolved
                        &&Batch.TimestampFrequency)
                    {
                        double TileDispatchMs=0.0;
                        if(VoxelForgeGpuFarViewer::ResolveGpuTimestampPair(Context,Tile.TimingQueries,
                            Batch.TimestampFrequency,TileDispatchMs))
                        {
                            Tile.GpuDispatchMs=TileDispatchMs;Tile.bTimingResolved=true;
                            GpuFarGpuDispatchMs+=TileDispatchMs;
                            GpuFarGpuDispatchMaxMs=std::max(GpuFarGpuDispatchMaxMs,TileDispatchMs);
                        }
                    }
                    if(Batch.bGpuComplete)RecordGpuCrossingTileReady(Key);
                    return Batch.bGpuComplete;
                }
                if(!Tile.bGpuComplete&&Tile.CompletionEvent&&Context
                    &&Context->GetData(Tile.CompletionEvent,nullptr,0,D3D11_ASYNC_GETDATA_DONOTFLUSH)==S_OK)
                {
                    Tile.bGpuComplete=true;
                    Tile.CompletionEvent->Release();Tile.CompletionEvent=nullptr;
                }
                if(Tile.bGpuComplete&&!Tile.bTimingResolved)
                {
                    double AsyncDispatchMs=0.0;
                    if(VoxelForgeGpuFarViewer::ResolveGpuTiming(Context,Tile.TimingQueries,AsyncDispatchMs))
                    {
                        if(Tile.GpuDispatchMs==0.0)
                        {
                            Tile.GpuDispatchMs=AsyncDispatchMs;
                            GpuFarGpuDispatchMs+=AsyncDispatchMs;
                            GpuFarGpuDispatchMaxMs=std::max(GpuFarGpuDispatchMaxMs,AsyncDispatchMs);
                        }
                        Tile.bTimingResolved=true;
                    }
                }
                if(Tile.bGpuComplete)RecordGpuCrossingTileReady(Key);
                return Tile.bGpuComplete;
            }
            return ResidentTiles.contains(Key);
        }

        void PrepareGpuFarInputs(std::vector<FVoxelTileKey> NeedPreparation)
        {
            if(NeedPreparation.empty())return;
            const auto PrepareStart=std::chrono::steady_clock::now();
            std::printf("[ColdTimeline] stage=gpu_input_export begin_ms=%.3f tiles=%zu\n",
                std::chrono::duration<double,std::milli>(PrepareStart-ColdProcessStart).count(),NeedPreparation.size());
            std::vector<std::optional<VoxelForgeCore::FarFieldTileInput>> Inputs(NeedPreparation.size());
            std::vector<std::string> Errors(NeedPreparation.size());
            const std::size_t ThreadCount=std::min<std::size_t>(8,NeedPreparation.size());
            std::atomic<std::size_t> Next{0};
            std::vector<double> WorkerElapsed(ThreadCount,0.0);
            std::vector<std::thread> Workers;Workers.reserve(ThreadCount);
            for(std::size_t Worker=0;Worker<ThreadCount;++Worker)
            {
                Workers.emplace_back([&,Worker]
                {
                    const auto WorkerStart=std::chrono::steady_clock::now();
                    for(;;)
                    {
                        const std::size_t Index=Next.fetch_add(1,std::memory_order_relaxed);
                        if(Index>=NeedPreparation.size())break;
                        const FVoxelTileKey& Key=NeedPreparation[Index];
                        const VoxelForgeCore::TileKey CoreKey{{Key.Coord.X,Key.Coord.Y,Key.Coord.Z},Key.Level};
                        try
                        {
                            Inputs[Index]=Key.Level<World->FullResClipLevels
                                ?ShapeWorld->exportGpuFieldTile(CoreKey,Errors[Index])
                                :ShapeWorld->exportFarFieldTile(CoreKey,Errors[Index]);
                            if(Inputs[Index]&&Key.Level>=World->FullResClipLevels)
                            {
                                Inputs[Index]->visibleSkirtFaceMask=World->bGenerateSkirts
                                    &&bGpuSkirtEmitEnabled
                                    ?ComputeVisibleSkirtFaceMask(Key,DesiredTileSet):0;
                                Inputs[Index]->skirtDepthVoxels=std::max(1.0f,
                                    static_cast<float>(World->SkirtCells))*Key.StepVoxels();
                                if(!bGpuConnectorRepairEnabled)
                                {
                                    Inputs[Index]->connectorPanels.clear();
                                    Inputs[Index]->connectorRingCount=0;
                                    Inputs[Index]->connectorVertexCount=0;
                                }
                            }
                        }
                        catch(const std::exception& Error) { Errors[Index]=Error.what(); }
                    }
                    WorkerElapsed[Worker]=std::chrono::duration<double,std::milli>(
                        std::chrono::steady_clock::now()-WorkerStart).count();
                });
            }
            for(std::thread& Worker:Workers)Worker.join();
            const double PreparationMs=std::chrono::duration<double,std::milli>(
                std::chrono::steady_clock::now()-PrepareStart).count();
            double WorkerSumMs=0.0;for(const double Elapsed:WorkerElapsed)WorkerSumMs+=Elapsed;
            GpuFarPreparationWallMs+=PreparationMs;GpuFarPreparationWorkerMs+=WorkerSumMs;
            std::size_t Failed=0;
            for(std::size_t Index=0;Index<NeedPreparation.size();++Index)
            {
                if(Inputs[Index])GpuFarPreparedInputs.emplace(NeedPreparation[Index],std::move(*Inputs[Index]));
                else
                {
                    ++Failed;
                    GpuFarPreparationErrors.emplace(NeedPreparation[Index],Errors[Index].empty()
                        ?"far-field input export failed":Errors[Index]);
                }
            }
            std::printf("[GpuFarInputPrepare] tiles=%zu workers=%zu wall_ms=%.3f worker_sum_ms=%.3f failed=%zu\n",
                NeedPreparation.size(),ThreadCount,PreparationMs,WorkerSumMs,Failed);
            std::printf("[ColdTimeline] stage=gpu_input_export end_ms=%.3f duration_ms=%.3f worker_sum_ms=%.3f\n",
                std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count(),
                PreparationMs,WorkerSumMs);
            std::fflush(stdout);
        }

        void ScheduleGpuFarTiles(const std::vector<FVoxelTileKey>& Keys)
        {
            GpuFarQueue.clear();GpuFarQueued.clear();GpuFarEligibleTiles.clear();
            GpuFarPreparedInputs.clear();GpuFarPreparationErrors.clear();
            if(!bGpuFarEnabled||!ShapeWorld||!ShapeWorld->hasFarWorld())return;
            if(bGpuFarShaderPrewarm)
            {
                if(GpuShaderWarmupThread.joinable())GpuShaderWarmupThread.join();
                if(!bGpuShaderWarmupQueueSucceeded)
                {
                    std::fprintf(stderr,"[GpuShaderWarmupError] queue: %s\n",
                        GpuShaderWarmupQueueError.empty()?"warmup queue failed":GpuShaderWarmupQueueError.c_str());
                    bGpuFarEnabled=false;
                    return;
                }
                std::string WarmupError;
                if(!VoxelForgeGpuFarViewer::CompleteFarShaderWarmup(Device,Context,
                    GpuFarShaderPath,WarmupError))
                {
                    std::fprintf(stderr,"[GpuShaderWarmupError] %s\n",WarmupError.c_str());
                    bGpuFarEnabled=false;
                    return;
                }
            }
            std::vector<FVoxelTileKey> CpuTiles;CpuTiles.reserve(Keys.size());
            std::vector<FVoxelTileKey> Eligible;Eligible.reserve(Keys.size());
            uint64 SkirtTransitionFallbackTiles=0;
            uint64 RemoteLod3CpuFallbackTiles=0,RemoteLod4CpuFallbackTiles=0;
            for(const FVoxelTileKey& Key:Keys)
            {
                const VoxelForgeCore::TileKey CoreKey{{Key.Coord.X,Key.Coord.Y,Key.Coord.Z},Key.Level};
                const bool NearGpuEligible=Key.Level<World->FullResClipLevels;
                if(Key.Level==3||Key.Level==4)
                {
                    CpuTiles.push_back(Key);
                    if(Key.Level==3)++RemoteLod3CpuFallbackTiles;
                    else ++RemoteLod4CpuFallbackTiles;
                    if(bDrawOnlyTile&&IsDrawOnlyTile(Key))
                        std::printf("[DrawOnlyTileRoute] key=(%d,%d,%d,lod%d) route=cpu_remote_band_fallback\n",
                            Key.Coord.X,Key.Coord.Y,Key.Coord.Z,Key.Level);
                    continue;
                }
                if(NearGpuEligible||ShapeWorld->isFarFieldTileEligible(CoreKey))
                {
                    const uint8 VisibleSkirtFaces=World->bGenerateSkirts
                        ?ComputeVisibleSkirtFaceMask(Key,DesiredTileSet):0;
                    if(Key.Level>=World->FullResClipLevels)
                    {
                        const auto Existing=GpuFarTiles.find(Key);
                        if(Existing!=GpuFarTiles.end()
                            &&Existing->second.VisibleSkirtFaceMask!=VisibleSkirtFaces)
                        {
                            const uint32 ReusedSlot=Existing->second.OutputSlot;
                            ReleaseGpuFarTile(Existing->second);
                            GpuFarTiles.erase(Existing);
                            GpuFarFreeOutputSlots.push_back(ReusedSlot);
                        }
                    }
                    if(VisibleSkirtFaces!=0&&NearGpuEligible)
                    {
                        // Keep fine tiles on the CPU reference path. Far tiles carry the
                        // CPU-decided visible face mask to the deterministic GPU skirt pass.
                        CpuTiles.push_back(Key);
                        ++SkirtTransitionFallbackTiles;
                    }
                    else Eligible.push_back(Key);
                }
                else CpuTiles.push_back(Key);
                if(bDrawOnlyTile&&IsDrawOnlyTile(Key))
                    std::printf("[DrawOnlyTileRoute] key=(%d,%d,%d,lod%d) route=%s\n",
                        Key.Coord.X,Key.Coord.Y,Key.Coord.Z,Key.Level,
                        Eligible.empty()||Eligible.back()!=Key?"cpu_fallback":"gpu_eligible");
            }
            if(RemoteLod3CpuFallbackTiles+RemoteLod4CpuFallbackTiles>0)
                std::printf("[GpuRemoteBandFallback] lod3_cpu_tiles=%llu lod4_cpu_tiles=%llu reason=remote_capture_pixel_coverage\n",
                    static_cast<unsigned long long>(RemoteLod3CpuFallbackTiles),
                    static_cast<unsigned long long>(RemoteLod4CpuFallbackTiles));
            if(Eligible.size()>GpuFarOutputCapacitySlots)
            {
                for(auto& Pair:GpuFarTiles)ReleaseGpuFarTile(Pair.second);
                GpuFarTiles.clear();GpuFarFreeOutputSlots.clear();GpuFarNextOutputSlot=0;
                GpuFarOutputCapacitySlots=static_cast<uint32>(Eligible.size());
            }
            if(!Eligible.empty()&&GpuFarOutputCapacitySlots==0)
                GpuFarOutputCapacitySlots=static_cast<uint32>(Eligible.size());
            if(bCaptureOnlyTile&&CaptureOutputSlot>=0)
            {
                GpuFarOutputCapacitySlots=std::max(GpuFarOutputCapacitySlots,
                    static_cast<uint32>(CaptureOutputSlot+1));
                GpuFarNextOutputSlot=static_cast<uint32>(CaptureOutputSlot);
                std::printf("[CaptureOutputSlotProbe] slot=%d pool_slots=%u\n",
                    CaptureOutputSlot,GpuFarOutputCapacitySlots);
            }
            if(!Eligible.empty())
            {
                std::string PoolError;const auto ReserveStart=std::chrono::steady_clock::now();
                std::printf("[ColdTimeline] stage=gpu_output_pool begin_ms=%.3f slots=%u\n",
                    std::chrono::duration<double,std::milli>(ReserveStart-ColdProcessStart).count(),
                    GpuFarOutputCapacitySlots);
                if(!VoxelForgeGpuFarViewer::ReserveFarOutputPool(Device,Context,GpuFarShaderPath,
                    GpuFarOutputCapacitySlots,static_cast<uint32>(World->CoarseTileCells),PoolError))
                {
                    std::fprintf(stderr,"[GpuFarPoolError] %s\n",PoolError.c_str());
                    bGpuFarEnabled=false;GpuFarOutputCapacitySlots=0;
                    CpuTiles=Keys;Eligible.clear();GpuFarTiles.clear();
                }
                GpuFarPoolReserveMs+=std::chrono::duration<double,std::milli>(
                    std::chrono::steady_clock::now()-ReserveStart).count();
                std::printf("[ColdTimeline] stage=gpu_output_pool end_ms=%.3f duration_ms=%.3f\n",
                    std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count(),
                    GpuFarPoolReserveMs);
                SampleGpuFarMemory();
            }
            std::vector<FVoxelTileKey> NeedPreparation;NeedPreparation.reserve(Eligible.size());
            for(const FVoxelTileKey& Key:Eligible)
                if(!GpuFarTiles.contains(Key))NeedPreparation.push_back(Key);
            if(!NeedPreparation.empty())
                PrepareGpuFarInputs(std::move(NeedPreparation));
            std::vector<FVoxelTileKey> PreparedEligible;PreparedEligible.reserve(Eligible.size());
            for(const FVoxelTileKey& Key:Eligible)
            {
                if(!GpuFarTiles.contains(Key)&&!GpuFarPreparedInputs.contains(Key))
                {
                    CpuTiles.push_back(Key);
                    const auto Error=GpuFarPreparationErrors.find(Key);
                    std::fprintf(stderr,"[GpuFieldInputFallback] key=(%d,%d,%d,lod%d) reason=%s\n",
                        Key.Coord.X,Key.Coord.Y,Key.Coord.Z,Key.Level,
                        Error==GpuFarPreparationErrors.end()?"missing prepared input":Error->second.c_str());
                    continue;
                }
                GpuFarEligibleTiles.insert(Key);
                if(!GpuFarTiles.contains(Key))
                {GpuFarQueue.push_back(Key);GpuFarQueued.insert(Key);}
                PreparedEligible.push_back(Key);
            }
            Eligible.swap(PreparedEligible);
            std::sort(GpuFarQueue.begin(),GpuFarQueue.end(),[&](const FVoxelTileKey& A,const FVoxelTileKey& B)
            {
                if(A.Level!=B.Level)return A.Level<B.Level;
                const FIntVector AO=A.OriginVoxels(),BO=B.OriginVoxels();
                const double AC=(AO.X+0.5*A.ExtentVoxels()-Position.X*4.0)*(AO.X+0.5*A.ExtentVoxels()-Position.X*4.0)
                    +(AO.Y+0.5*A.ExtentVoxels()-Position.Y*4.0)*(AO.Y+0.5*A.ExtentVoxels()-Position.Y*4.0)
                    +(AO.Z+0.5*A.ExtentVoxels()-Position.Z*4.0)*(AO.Z+0.5*A.ExtentVoxels()-Position.Z*4.0);
                const double BC=(BO.X+0.5*B.ExtentVoxels()-Position.X*4.0)*(BO.X+0.5*B.ExtentVoxels()-Position.X*4.0)
                    +(BO.Y+0.5*B.ExtentVoxels()-Position.Y*4.0)*(BO.Y+0.5*B.ExtentVoxels()-Position.Y*4.0)
                    +(BO.Z+0.5*B.ExtentVoxels()-Position.Z*4.0)*(BO.Z+0.5*B.ExtentVoxels()-Position.Z*4.0);
                return AC<BC;
            });
            GpuFarQueued.clear();for(const FVoxelTileKey& Key:GpuFarQueue)GpuFarQueued.insert(Key);
            if(Streamer)Streamer->SetDesiredTileFilter(CpuTiles);
            const std::size_t NearGpuTiles=static_cast<std::size_t>(std::count_if(
                GpuFarEligibleTiles.begin(),GpuFarEligibleTiles.end(),[&](const FVoxelTileKey& Key)
                    {return Key.Level<World->FullResClipLevels;}));
            std::printf("[GpuFieldEligibility] eligible_gpu_tiles=%zu near_gpu_tiles=%zu cpu_fallback_tiles=%zu\n",
                GpuFarEligibleTiles.size(),NearGpuTiles,Keys.size()-GpuFarEligibleTiles.size());
            std::printf("[GpuSkirtSeamFallback] near_tiles=%llu reason=fine_visible_lod_skirts gpu_far_skirts=active\n",
                static_cast<unsigned long long>(SkirtTransitionFallbackTiles));
        }

        void PumpGpuFarGeneration()
        {
            if(!bGpuFarEnabled||!ShapeWorld||!ShapeWorld->hasFarWorld()||!Device||!Context
                ||GpuFarQueue.empty())return;
            const auto PumpStart=std::chrono::steady_clock::now();
            const double PumpMs=std::chrono::duration<double,std::milli>(PumpStart-ColdProcessStart).count();
            if(GpuFarFirstPumpMs<0.0)GpuFarFirstPumpMs=PumpMs;
            if(LastGpuFarPumpAt.time_since_epoch().count()!=0)
            {
                const double GapMs=std::chrono::duration<double,std::milli>(PumpStart-LastGpuFarPumpAt).count();
                GpuFarPumpGapMs+=GapMs;GpuFarPumpGapMaxMs=std::max(GpuFarPumpGapMaxMs,GapMs);
            }
            LastGpuFarPumpAt=PumpStart;GpuFarLastPumpMs=PumpMs;++GpuFarPumpCalls;
            const bool bDeferredBatch=!bGpuFarAuditEnabled;
            if(bDeferredBatch)
            {
                std::string BatchError;
                if(!VoxelForgeGpuFarViewer::BeginFarTileBatch(Device,Context,GpuFarShaderPath,BatchError))
                {
                    std::fprintf(stderr,"[GpuFarBatchError] begin: %s\n",BatchError.c_str());
                    bGpuFarEnabled=false;return;
                }
            }
            const auto Start=std::chrono::steady_clock::now();
            int32 Built=0;
            std::vector<FVoxelTileKey> BuiltKeys;BuiltKeys.reserve(32);
            do
            {
                const FVoxelTileKey Key=GpuFarQueue.front();GpuFarQueue.pop_front();GpuFarQueued.erase(Key);
                if(!DesiredTileSet.contains(Key)||GpuFarTiles.contains(Key))continue;
                FGpuFarTile Tile;
                if(!GpuFarFreeOutputSlots.empty())
                {Tile.OutputSlot=GpuFarFreeOutputSlots.back();GpuFarFreeOutputSlots.pop_back();}
                else if(GpuFarNextOutputSlot<GpuFarOutputCapacitySlots)
                    Tile.OutputSlot=GpuFarNextOutputSlot++;
                else
                {
                    std::fprintf(stderr,"[GpuFarPoolError] no free far-output slot for queued tile\n");
                    bGpuFarEnabled=false;break;
                }
                VoxelForgeGpuFarViewer::CpuMesh AuditMesh;
                std::string Error;double EnqueueMs=0.0,GpuDispatchMs=0.0;
                const VoxelForgeCore::TileKey CoreKey{{Key.Coord.X,Key.Coord.Y,Key.Coord.Z},Key.Level};
                const auto PreparedInputIt=GpuFarPreparedInputs.find(Key);
                const VoxelForgeCore::FarFieldTileInput* PreparedInput=PreparedInputIt==GpuFarPreparedInputs.end()
                    ?nullptr:&PreparedInputIt->second;
                if(!PreparedInput)
                {
                    const auto PrepError=GpuFarPreparationErrors.find(Key);
                    std::fprintf(stderr,"[GpuFarInputError] key=(%d,%d,%d,lod%d): %s\n",
                        Key.Coord.X,Key.Coord.Y,Key.Coord.Z,Key.Level,
                        PrepError==GpuFarPreparationErrors.end()?"prepared far input is missing":PrepError->second.c_str());
                    GpuFarFreeOutputSlots.push_back(Tile.OutputSlot);
                    bGpuFarEnabled=false;
                    break;
                }
                const uint8 PreparedSkirtFaceMask=PreparedInput->visibleSkirtFaceMask;
                if(!VoxelForgeGpuFarViewer::BuildFarTile(Device,Context,*ShapeWorld,CoreKey,
                    PreparedInput,
                    GpuFarShaderPath,&Tile.VertexBuffer,&Tile.IndexBuffer,&Tile.IndirectArgs,
                    &Tile.ConnectorVertexBuffer,&Tile.ConnectorIndexBuffer,
                    Tile.OutputSlot,GpuFarOutputCapacitySlots,&Tile.IndirectArgsByteOffset,
                    bDeferredBatch?nullptr:&Tile.CompletionEvent,
                    &Tile.TimingQueries,
                    (bGpuFarAuditEnabled&&Key.Level>=6)?&AuditMesh:nullptr,
                    &EnqueueMs,&GpuDispatchMs,Error))
                {
                    GpuFarFreeOutputSlots.push_back(Tile.OutputSlot);
                    std::fprintf(stderr,"[GpuFarTileError] key=(%d,%d,%d,lod%d): %s\n",
                        Key.Coord.X,Key.Coord.Y,Key.Coord.Z,Key.Level,Error.c_str());
                    if(PreparedInputIt!=GpuFarPreparedInputs.end())GpuFarPreparedInputs.erase(PreparedInputIt);
                    bGpuFarEnabled=false;
                    break;
                }
                if(PreparedInputIt!=GpuFarPreparedInputs.end())GpuFarPreparedInputs.erase(PreparedInputIt);
                if(bGpuFarAuditEnabled&&GpuFarAudit&&Key.Level>=6)
                {
                    FStandaloneCpuTile AuditTile;AuditTile.Key=Key;
                    AuditTile.SurfaceVertexCount=static_cast<uint32>(AuditMesh.vertices.size());
                    AuditTile.Vertices.reserve(AuditMesh.vertices.size());
                    for(const auto& Vertex:AuditMesh.vertices)
                        AuditTile.Vertices.push_back({
                            {Vertex.position[0]*0.01f,Vertex.position[1]*0.01f,Vertex.position[2]*0.01f},
                            {Vertex.normal[0],Vertex.normal[1],Vertex.normal[2]}});
                    AuditTile.Indices=std::move(AuditMesh.indices);
                    GpuFarAudit->AuditTile(AuditTile);
                }
                Tile.VertexCapacityBytes=static_cast<uint64>(
                    VoxelForgeGpuFarViewer::PooledOutputVertexCapacityPerTile)*32u;
                Tile.VisibleSkirtFaceMask=PreparedSkirtFaceMask;
                Tile.GpuDispatchMs=GpuDispatchMs;
                GpuFarEnqueueWallSeconds+=EnqueueMs/1000.0;
                if(GpuDispatchMs>0.0)
                {
                    GpuFarGpuDispatchMs+=GpuDispatchMs;
                    GpuFarGpuDispatchMaxMs=std::max(GpuFarGpuDispatchMaxMs,GpuDispatchMs);
                    Tile.bTimingResolved=true;
                }
                ++GpuFarTilesSubmitted;++Built;
                GpuFarTiles.emplace(Key,Tile);
                BuiltKeys.push_back(Key);
                if(!bGpuFarFirstTileReported)
                {
                    bGpuFarFirstTileReported=true;
                    std::printf("[GpuFarFirstTile] key=(%d,%d,%d,lod%d) cpu_export_upload_enqueue_ms=%.3f vertex_capacity_mb=%.3f mode=direct-indirect-draw\n",
                        Key.Coord.X,Key.Coord.Y,Key.Coord.Z,Key.Level,EnqueueMs,
                        Tile.VertexCapacityBytes/(1024.0*1024.0));
                    std::fflush(stdout);
                }
            }
            while(Built<32&&!GpuFarQueue.empty()
                &&std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-Start).count()<4.0);
            if(Built>0)
            {
                if(bDeferredBatch)
                {
                    FGpuFarBatch Batch;Batch.TileCount=static_cast<uint32>(Built);
                    std::string BatchError;
                    if(!VoxelForgeGpuFarViewer::SubmitFarTileBatch(Device,Context,GpuFarShaderPath,
                        &Batch.CompletionEvent,&Batch.TimingQueries,BatchError))
                    {
                        std::fprintf(stderr,"[GpuFarBatchError] submit: %s\n",BatchError.c_str());
                        bGpuFarEnabled=false;
                    }
                    else
                    {
                        Batch.SubmittedAt=std::chrono::steady_clock::now();
                        const double SubmitMs=std::chrono::duration<double,std::milli>(
                            Batch.SubmittedAt-ColdProcessStart).count();
                        if(GpuFarFirstSubmitMs<0.0)GpuFarFirstSubmitMs=SubmitMs;
                        GpuFarLastSubmitMs=SubmitMs;
                        const uint32 BatchIndex=static_cast<uint32>(GpuFarBatches.size());
                        GpuFarBatches.push_back(Batch);
                        for(const FVoxelTileKey& Key:BuiltKeys)
                        {
                            auto It=GpuFarTiles.find(Key);
                            if(It!=GpuFarTiles.end())It->second.BatchIndex=BatchIndex;
                        }
                        ++GpuFarSubmissionBatches;
                        GpuFarMaxTilesPerSubmissionBatch=std::max<uint32>(GpuFarMaxTilesPerSubmissionBatch,
                            static_cast<uint32>(Built));
                    }
                }
                else
                {
                    ++GpuFarSubmissionBatches;
                    GpuFarMaxTilesPerSubmissionBatch=std::max<uint32>(GpuFarMaxTilesPerSubmissionBatch,
                        static_cast<uint32>(Built));
                }
            }
            SampleGpuFarMemory();
            const double PumpWallMs=std::chrono::duration<double,std::milli>(
                std::chrono::steady_clock::now()-PumpStart).count();
            GpuFarPumpWallMs+=PumpWallMs;GpuFarPumpMaxWallMs=std::max(GpuFarPumpMaxWallMs,PumpWallMs);
        }

        void ToggleFarRenderer()
        {
            if(!bShapeWorld||!ShapeWorld||!ShapeWorld->hasFarWorld())
            {
                bGpuFarEnabled=!bGpuFarEnabled;
                std::printf("[FarRenderer] GPU far %s; it will take effect when far shapes are ready.\n",
                    bGpuFarEnabled?"enabled":"disabled");
                return;
            }
            bGpuFarEnabled=!bGpuFarEnabled;
            if(bGpuFarEnabled)
            {
                bPresentCpuFrameBeforeGpu=true;
                bHoldCpuFrameDuringGpuWarmup=true;
                bCpuReadyFramePresentedBeforeGpu=false;
                ScheduleGpuFarTiles(DesiredTiles);
                std::vector<FVoxelTileKey> CpuFar;
                for(const FVoxelTileKey& Key:DesiredTiles)
                    if(GpuFarEligibleTiles.contains(Key)&&ResidentTiles.contains(Key))CpuFar.push_back(Key);
                for(const FVoxelTileKey& Key:CpuFar)RemoveResidentTile(Key);
            }
            else
            {
                GpuFarQueue.clear();GpuFarQueued.clear();GpuFarEligibleTiles.clear();
                Streamer->SetDesiredTileFilter(DesiredTiles);
            }
            std::vector<FVoxelTileKey> ResidentKeys;ResidentKeys.reserve(ResidentTiles.size());
            for(const auto& Pair:ResidentTiles)ResidentKeys.push_back(Pair.first);
            const FVector CameraVoxels(Position.X*4.0,Position.Y*4.0,Position.Z*4.0);
            ++CrossingIndex;
            const FStandaloneStreamingUpdate Update=Streamer->UpdateDesired(
                CameraVoxels,ResidentKeys,CrossingIndex);
            FCrossingWork& Work=CrossingWork[CrossingIndex];
            Work.SubmittedJobs=Update.NewJobs;Work.PendingJobs=Update.NewJobs;
            Work.StartTime=std::chrono::steady_clock::now();
            std::printf("[FarRenderer] mode=%s cpu_field_jobs=%d gpu_field_ready=%zu gpu_field_queued=%zu (H toggles)\n",
                bGpuFarEnabled?"GPU":"CPU",Update.NewJobs,GpuFarTiles.size(),GpuFarQueue.size());
        }

        void TryCompleteShapePipeline()
        {
            if(!bShapePipelineActive||InitialNeighborhoodTiles.empty())return;
            bool bNeighborhoodUploaded=true;
            for(const FVoxelTileKey& Key:InitialNeighborhoodTiles)
                if(!ResidentTiles.contains(Key)){bNeighborhoodUploaded=false;break;}
            bNeighborhoodUploaded &= DirtyBatchSet.empty();
            if(bNeighborhoodUploaded&&!bNeighborhoodReadyReported)
            {
                bNeighborhoodReadyReported=true;
                const double Elapsed=std::chrono::duration<double>(
                    std::chrono::steady_clock::now()-ColdProcessStart).count();
                std::printf("[NeighborhoodReady] %.3fs after cold process start; tiles=%zu/%zu vertices=%llu indices=%llu\n",
                    Elapsed,InitialNeighborhoodTiles.size(),InitialNeighborhoodTiles.size(),
                    static_cast<unsigned long long>(TotalVertices),
                    static_cast<unsigned long long>(TotalIndices));
                std::fflush(stdout);
            }
            if(!bNeighborhoodReadyReported
                ||!bFullShapeBuildFinished.load(std::memory_order_acquire))return;
            if(FullShapeBuildThread.joinable())FullShapeBuildThread.join();
            if(!bFullShapeBuildSucceeded||!PendingFullShapeWorld)
            {
                bShapePipelineActive=false;bGenerationRunning=false;
                std::fprintf(stderr,"[ShapePipeline] full shape build failed: %s\n",
                    FullShapeBuildError.c_str());
                PostMessageA(Window,WM_CLOSE,0,0);
                return;
            }

            const double FullShapeSeconds=std::chrono::duration<double>(
                std::chrono::steady_clock::now()-FullShapeBuildStart).count();
            if(Streamer)Streamer->Stop();
            Streamer.reset();
            ShapeWorld=std::move(PendingFullShapeWorld);
            bFarShapeStagePending=false;bShapePipelineActive=false;

            const FVector CameraVoxels(Position.X*4.0,Position.Y*4.0,Position.Z*4.0);
            Streamer=std::make_unique<FStandaloneStreamingGenerator>(*World,WorkerCount,false,nullptr,-1,
                false,false,false,false,ShapeWorld.get());
            if(bGpuFarEnabled&&ShapeWorld->hasFarWorld())
            {
                ScheduleGpuFarTiles(DesiredTiles);
                if(bGpuFarAuditEnabled)
                {
                    std::vector<FVoxelTileKey> AuditKeys;AuditKeys.reserve(GpuFarEligibleTiles.size());
                    for(const FVoxelTileKey& Key:DesiredTiles)
                        if(Key.Level>=6&&GpuFarEligibleTiles.contains(Key))AuditKeys.push_back(Key);
                    GpuFarAudit=std::make_unique<FFarSurvivalAudit>(*World,*ShapeWorld,AuditKeys,false);
                }
                std::vector<FVoxelTileKey> CpuResidentNear;
                for(const auto& Pair:ResidentTiles)
                    if(Pair.first.Level<World->FullResClipLevels
                        &&GpuFarEligibleTiles.contains(Pair.first))CpuResidentNear.push_back(Pair.first);
                for(const FVoxelTileKey& Key:CpuResidentNear)RemoveResidentTile(Key);
            }
            std::vector<FVoxelTileKey> ResidentKeys;ResidentKeys.reserve(ResidentTiles.size());
            for(const auto& Pair:ResidentTiles)ResidentKeys.push_back(Pair.first);
            ++CrossingIndex;
            const FStandaloneStreamingUpdate Update=Streamer->Start(CameraVoxels,ResidentKeys,CrossingIndex);
            if(Update.DesiredTiles!=static_cast<int32>(DesiredTiles.size()))
            {
                Streamer->Stop();Streamer.reset();bGenerationRunning=false;
                std::fprintf(stderr,"[ShapePipeline] full streamer accepted %d of %zu desired tiles.\n",
                    Update.DesiredTiles,DesiredTiles.size());
                PostMessageA(Window,WM_CLOSE,0,0);
                return;
            }
            FCrossingWork& Work=CrossingWork[CrossingIndex];
            Work.SubmittedJobs=Update.NewJobs;Work.PendingJobs=Update.NewJobs;
            Work.StartTime=std::chrono::steady_clock::now();
            bGenerationRunning=true;
            std::printf("[ShapePipelineFullReady] full_shape=%.3fs retained=%d new_jobs=%d workers=%d; filling outward\n",
                FullShapeSeconds,Update.RetainedResidentTiles,Update.NewJobs,WorkerCount);
            std::fflush(stdout);
        }

        void MarkBatchDirty(const FStandaloneBatchKey& Key)
        {DirtyBatchSet.insert(Key);}

        void RefreshResidentSkirtMasks()
        {
            for(auto& [Key,Resident]:ResidentTiles)
            {
                if(!DesiredTileSet.contains(Key))continue;
                const uint8 NewMask=ComputeVisibleSkirtFaceMask(Key,DesiredTileSet);
                if(NewMask==Resident.ActiveSkirtFaceMask)continue;
                Resident.ActiveSkirtFaceMask=NewMask;
                MarkBatchDirty(BatchForTile(Key));
            }
        }

        static bool TileKeyLess(const FVoxelTileKey& A,const FVoxelTileKey& B)
        {
            if(A.Level!=B.Level)return A.Level<B.Level;
            if(A.Coord.Z!=B.Coord.Z)return A.Coord.Z<B.Coord.Z;
            if(A.Coord.Y!=B.Coord.Y)return A.Coord.Y<B.Coord.Y;
            return A.Coord.X<B.Coord.X;
        }

        void RemoveResidentTile(const FVoxelTileKey& Key)
        {
            auto Resident=ResidentTiles.find(Key);
            if(Resident==ResidentTiles.end())return;
            const FStandaloneBatchKey Batch=BatchForTile(Key);
            auto Members=BatchMembers.find(Batch);
            if(Members!=BatchMembers.end())
            {
                auto& Items=Members->second;
                Items.erase(std::remove(Items.begin(),Items.end(),Key),Items.end());
                if(Items.empty())BatchMembers.erase(Members);
            }
            MarkBatchDirty(Batch);
            ResidentTiles.erase(Resident);
        }

        void RecenterStreaming()
        {
            if(!World||!Streamer)return;
            const FVector CameraVoxels(Position.X*4.0,Position.Y*4.0,Position.Z*4.0);
            FStandaloneDesiredTileSet NewDesired=StandaloneWorldMesher::BuildDesiredSet(*World,CameraVoxels);
            if(bShapeWorld&&ShapeWorld)
            {
                const auto CoreTiles=ShapeWorld->desiredTiles(
                    {CameraVoxels.X,CameraVoxels.Y,CameraVoxels.Z});
                NewDesired.Tiles.clear();NewDesired.Tiles.reserve(CoreTiles.size());
                for(const auto& Key:CoreTiles)
                    NewDesired.Tiles.emplace_back(
                        FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
            }
            if(NewDesired.CenterChunk==BuildTargetChunk)return;
            const FIntVector Previous=BuildTargetChunk;
            const FIntVector CenterDelta(NewDesired.CenterChunk.X-Previous.X,
                NewDesired.CenterChunk.Y-Previous.Y,NewDesired.CenterChunk.Z-Previous.Z);
            if(FlightTargetCrossings>0&&bFlightStarted&&!bFlightFinished)
            {
                ++FlightObservedCrossings;
                if(CenterDelta.X!=0||CenterDelta.Y!=1||CenterDelta.Z!=0)
                    bFlightCrossingsValid=false;
            }
            std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> NewSet;
            NewSet.reserve(NewDesired.Tiles.size()*2+1);
            for(const FVoxelTileKey& Key:NewDesired.Tiles)NewSet.insert(Key);
            std::vector<FVoxelTileKey> Leavers;
            for(const auto& Pair:ResidentTiles)if(!NewSet.contains(Pair.first))Leavers.push_back(Pair.first);
            const auto CrossingStart=std::chrono::steady_clock::now();
            double PageCompileSeconds=0.0;
            uint64 PageCoveredTiles=0;
            VoxelForgeCore::TileRegionCompileStats PageStats;
            const uint64 NextCrossing=CrossingIndex+1;
            FCrossingWork& NextWork=CrossingWork[NextCrossing];
            NextWork.StartTime=CrossingStart;
            for(const FVoxelTileKey& Key:NewDesired.Tiles)
                if(NewSet.contains(Key)&&!ResidentTiles.contains(Key))
                {
                    NextWork.NearestNewTile=Key;
                    NextWork.bHasNearestNewTile=true;
                    break;
                }
            if(bShapeWorld&&ShapeWorld)
            {
                std::vector<VoxelForgeCore::TileKey> NewCoreTiles;
                NewCoreTiles.reserve(NewDesired.Tiles.size());
                for(const FVoxelTileKey& Key:NewDesired.Tiles)
                    if(!ResidentTiles.contains(Key))
                        NewCoreTiles.push_back({{Key.Coord.X,Key.Coord.Y,Key.Coord.Z},Key.Level});
                std::string PageError;
                size_t CompiledPageCount=0;
                if(!ShapeWorld->compileTileRegion(NewCoreTiles,
                    static_cast<uint32>(std::clamp(WorkerCount,1,8)),CoreScheduler,PageError,
                    &CompiledPageCount,&PageStats,true))
                {
                    std::fprintf(stderr,"[ShapeRegionPage] compile failed at center=(%d,%d,%d): %s\n",
                        NewDesired.CenterChunk.X,NewDesired.CenterChunk.Y,NewDesired.CenterChunk.Z,
                        PageError.c_str());
                    return;
                }
                PageCoveredTiles=static_cast<uint64>(CompiledPageCount);
                std::vector<VoxelForgeCore::TileKey> CoreDesired;
                CoreDesired.reserve(NewDesired.Tiles.size());
                for(const FVoxelTileKey& Key:NewDesired.Tiles)
                    CoreDesired.push_back({{Key.Coord.X,Key.Coord.Y,Key.Coord.Z},Key.Level});
                ShapeWorld->retainTileRegions(CoreDesired);
                for(const FVoxelTileKey& Key:Leavers)DeferredLeaverTiles.insert(Key);
                const double PageSeconds=std::chrono::duration<double>(
                    std::chrono::steady_clock::now()-CrossingStart).count();
                PageCompileSeconds=PageStats.wallSeconds;
                std::printf("[ShapeRegionPage] region_mode=analytic-lod-slab center=(%d,%d,%d) requested_tiles=%zu page_covered_tiles=%llu regions=%zu lattice_nodes=%llu worker_limit=%d shape=%.4fs intervals_signs=%.4fs detail=%.4fs core_worker_sum=%.4fs seconds=%.4f retained_resident=%zu deferred_leavers=%zu\n",
                    NewDesired.CenterChunk.X,NewDesired.CenterChunk.Y,NewDesired.CenterChunk.Z,
                    NewCoreTiles.size(),static_cast<unsigned long long>(PageCoveredTiles),
                    PageStats.regionsBuilt,static_cast<unsigned long long>(PageStats.latticeNodes),
                    std::clamp(WorkerCount,1,8),PageStats.shapeSeconds,PageStats.intervalSeconds,
                    PageStats.surfaceSeconds,PageStats.workerSeconds,PageSeconds,
                    ResidentTiles.size()-std::min(ResidentTiles.size(),Leavers.size()),Leavers.size());
            }
            else
            {
                for(const FVoxelTileKey& Key:Leavers)RemoveResidentTile(Key);
            }
            DesiredTiles=NewDesired.Tiles;
            DesiredTileSet=std::move(NewSet);
            for(auto It=GpuFarTiles.begin();It!=GpuFarTiles.end();)
            {
                if(DesiredTileSet.contains(It->first)){++It;continue;}
                GpuFarFreeOutputSlots.push_back(It->second.OutputSlot);
                ReleaseGpuFarTile(It->second);
                It=GpuFarTiles.erase(It);
            }
            ScheduleGpuFarTiles(DesiredTiles);
            RefreshResidentSkirtMasks();
            for(int32 I=0;I<=MaxLevels;++I)DesiredByLevel[I]=0;
            for(const FVoxelTileKey& Tile:DesiredTiles)++DesiredByLevel[Tile.Level];
            std::vector<FVoxelTileKey> ResidentKeys;ResidentKeys.reserve(ResidentTiles.size());
            for(const auto& Pair:ResidentTiles)
                if(DesiredTileSet.contains(Pair.first))ResidentKeys.push_back(Pair.first);
            ++CrossingIndex;
            const FStandaloneStreamingUpdate Update=Streamer->UpdateDesiredWithSet(
                std::move(NewDesired),ResidentKeys,CrossingIndex);
            for(const auto& Pair:Update.CancelledByCrossing)
            {
                FCrossingWork& Work=CrossingWork[Pair.first];
                Work.PendingJobs=Work.PendingJobs>=Pair.second?Work.PendingJobs-Pair.second:0;
                Work.CancelledQueuedJobs+=Pair.second;
            }
            FCrossingWork& Work=CrossingWork[CrossingIndex];
            Work.SubmittedJobs=Update.NewJobs;Work.PendingJobs=Update.NewJobs;
            Work.PageCoveredTiles=PageCoveredTiles;Work.CorePageSeconds=PageCompileSeconds;
            Work.CorePageWorkerSeconds=PageStats.workerSeconds;
            Work.PageRegions=PageStats.regionsBuilt;Work.PageLatticeNodes=PageStats.latticeNodes;
            BuildTargetChunk=Update.CenterChunk;bGenerationRunning=true;
            std::printf("[ChunkCrossing] id=%llu center=(%d,%d,%d) delta=(%d,%d,%d) desired=%d retained=%d new_jobs=%d deferred_resident=%zu cancelled_queued=%d full_detail=%d\n",
                static_cast<unsigned long long>(CrossingIndex),Update.CenterChunk.X,Update.CenterChunk.Y,Update.CenterChunk.Z,
                Update.CenterChunk.X-Previous.X,Update.CenterChunk.Y-Previous.Y,Update.CenterChunk.Z-Previous.Z,
                Update.DesiredTiles,Update.RetainedResidentTiles,Update.NewJobs,
                bShapeWorld?Leavers.size():0,
                Update.CancelledQueuedJobs,Update.FullResTiles);
            std::fflush(stdout);
            ReportCrossingWork();
        }

        bool RebuildBatch(const FStandaloneBatchKey& Key)
        {
            auto MembersIt=BatchMembers.find(Key);
            auto Existing=StreamingBatches.find(Key);
            if(MembersIt==BatchMembers.end()||MembersIt->second.empty())
            {
                if(Existing!=StreamingBatches.end())
                {
                    TotalVertices-=Existing->second.VertexCount;TotalIndices-=Existing->second.IndexCount;
                    if(Existing->second.VertexBuffer)Existing->second.VertexBuffer->Release();
                    if(Existing->second.IndexBuffer)Existing->second.IndexBuffer->Release();
                    StreamingBatches.erase(Existing);
                }
                return true;
            }
            std::vector<FVoxelTileKey> Members=MembersIt->second;
            std::sort(Members.begin(),Members.end(),TileKeyLess);
            size_t VertexCount=0,IndexCount=0;
            for(const FVoxelTileKey& Tile:Members)
            {
                const auto It=ResidentTiles.find(Tile);if(It==ResidentTiles.end())continue;
                const FResidentTileMesh& Resident=It->second;
                VertexCount+=Resident.Vertices.size();
                if(Resident.TriangleSkirtFaceMasks.size()==Resident.Indices.size()/3)
                {
                    for(const uint8 Faces:Resident.TriangleSkirtFaceMasks)
                        if(Faces==0||(Faces&Resident.ActiveSkirtFaceMask)!=0)IndexCount+=3;
                }
                else IndexCount+=Resident.Indices.size();
            }
            if(VertexCount==0||IndexCount==0)
            {
                if(Existing!=StreamingBatches.end())
                {
                    TotalVertices-=Existing->second.VertexCount;TotalIndices-=Existing->second.IndexCount;
                    if(Existing->second.VertexBuffer)Existing->second.VertexBuffer->Release();
                    if(Existing->second.IndexBuffer)Existing->second.IndexBuffer->Release();
                    StreamingBatches.erase(Existing);
                }
                return true;
            }
            if(VertexCount>std::numeric_limits<uint32>::max()||IndexCount>std::numeric_limits<UINT>::max())
            {std::fprintf(stderr,"LOD %d batch exceeded D3D11 limits.\n",Key.Level);return false;}
            std::vector<FGpuVertex> Vertices;Vertices.reserve(VertexCount);
            std::vector<uint32> Indices;Indices.reserve(IndexCount);
            UINT DrawOnlyIndexStart=0,DrawOnlyIndexCount=0;
            for(const FVoxelTileKey& Tile:Members)
            {
                const auto It=ResidentTiles.find(Tile);if(It==ResidentTiles.end())continue;
                const size_t TileIndexStart=Indices.size();
                const uint32 Base=static_cast<uint32>(Vertices.size());
                const FResidentTileMesh& Resident=It->second;
                for(const FStandaloneCpuVertex& V:Resident.Vertices)
                    Vertices.push_back({{V.Position[0],V.Position[1],V.Position[2]},{V.Normal[0],V.Normal[1],V.Normal[2]}});
                const bool bHasFaceMasks=Resident.TriangleSkirtFaceMasks.size()==Resident.Indices.size()/3;
                for(size_t Triangle=0;Triangle<Resident.Indices.size()/3;++Triangle)
                {
                    if(bHasFaceMasks)
                    {
                        const uint8 Faces=Resident.TriangleSkirtFaceMasks[Triangle];
                        if(Faces!=0&&(Faces&Resident.ActiveSkirtFaceMask)==0)continue;
                    }
                    const size_t Start=Triangle*3;
                    Indices.push_back(Base+Resident.Indices[Start]);
                    Indices.push_back(Base+Resident.Indices[Start+1]);
                    Indices.push_back(Base+Resident.Indices[Start+2]);
                }
                if(bDrawOnlyTile&&IsDrawOnlyTile(Tile))
                {
                    DrawOnlyIndexStart=static_cast<UINT>(TileIndexStart);
                    DrawOnlyIndexCount=static_cast<UINT>(Indices.size()-TileIndexStart);
                }
            }
            FStreamingGpuBatch NewBatch;
            const bool bVertex=MakeBuffer(&NewBatch.VertexBuffer,Vertices.data(),Vertices.size()*sizeof(FGpuVertex),D3D11_BIND_VERTEX_BUFFER);
            const bool bIndex=bVertex&&MakeBuffer(&NewBatch.IndexBuffer,Indices.data(),Indices.size()*sizeof(uint32),D3D11_BIND_INDEX_BUFFER);
            if(!bIndex)
            {
                if(NewBatch.VertexBuffer)NewBatch.VertexBuffer->Release();
                std::fprintf(stderr,"D3D11 upload failed for LOD %d batch (%d,%d,%d).\n",Key.Level,Key.X,Key.Y,Key.Z);
                return false;
            }
            NewBatch.IndexCount=static_cast<UINT>(Indices.size());NewBatch.VertexCount=Vertices.size();
            NewBatch.DrawOnlyIndexStart=DrawOnlyIndexStart;
            NewBatch.DrawOnlyIndexCount=DrawOnlyIndexCount;
            if(Existing!=StreamingBatches.end())
            {
                TotalVertices-=Existing->second.VertexCount;TotalIndices-=Existing->second.IndexCount;
                if(Existing->second.VertexBuffer)Existing->second.VertexBuffer->Release();
                if(Existing->second.IndexBuffer)Existing->second.IndexBuffer->Release();
                Existing->second=NewBatch;
            }
            else StreamingBatches.emplace(Key,NewBatch);
            TotalVertices+=NewBatch.VertexCount;TotalIndices+=NewBatch.IndexCount;
            return true;
        }

        void UploadDirtyBatches()
        {
            if(DirtyBatchSet.empty())return;
            std::vector<FStandaloneBatchKey> Dirty(DirtyBatchSet.begin(),DirtyBatchSet.end());
            const FIntVector Center=Streamer?Streamer->CenterChunk():BuildTargetChunk;
            auto Distance=[&](const FStandaloneBatchKey& K)
            {
                const double Scale=static_cast<double>(1<<K.Level);
                const double DX=(K.X*8.0+4.0)*Scale-Center.X;
                const double DY=(K.Y*8.0+4.0)*Scale-Center.Y;
                const double DZ=(K.Z+0.5)*Scale-Center.Z;
                return DX*DX+DY*DY+DZ*DZ;
            };
            std::sort(Dirty.begin(),Dirty.end(),[&](const auto& A,const auto& B)
            {const double DA=Distance(A),DB=Distance(B);return DA==DB?A.Level<B.Level:DA<DB;});
            const auto Start=std::chrono::steady_clock::now();int32 Uploaded=0;
            for(const FStandaloneBatchKey& Key:Dirty)
            {
                if(Uploaded>=MaxBatchUploadsPerFrame)break;
                if(Uploaded>0&&BatchUploadBudgetMilliseconds>0.0
                    &&std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-Start).count()>=BatchUploadBudgetMilliseconds)break;
                if(!DirtyBatchSet.contains(Key))continue;
                if(RebuildBatch(Key)){DirtyBatchSet.erase(Key);++Uploaded;}
            }
        }

        bool IsFullDetailUploaded()
        {
            if(!Streamer||World->FullResClipLevels<=0)return false;
            int32 DesiredFullRes=0,ReadyFullRes=0;
            for(const FVoxelTileKey& Key:DesiredTiles)
            {
                if(Key.Level>=World->FullResClipLevels)continue;
                ++DesiredFullRes;
                if(IsRenderedTileReady(Key))++ReadyFullRes;
            }
            if(ReadyFullRes!=DesiredFullRes)return false;
            for(const FStandaloneBatchKey& Key:DirtyBatchSet)if(Key.Level<World->FullResClipLevels)return false;
            return true;
        }

        void ReportTileTimings()
        {
            for(int32 L=0;L<=World->MaxClipLevel;++L)
            {
                if(GeneratedByLevel[L]<static_cast<uint64>(DesiredByLevel[L])||DesiredByLevel[L]==0)continue;
                std::vector<double>& Times=TileSecondsByLevel[L];if(Times.empty())continue;
                std::sort(Times.begin(),Times.end());
                const double Average=WorkerSecondsByLevel[L]/static_cast<double>(GeneratedByLevel[L]);
                const double P50=Times[(Times.size()-1)*50/100];
                const double P95=Times[(Times.size()-1)*95/100];
                std::printf("[TileTiming] LOD=%d generated=%llu target=%d samples=%.2fM worker_sum=%.3fs avg=%.3fms p50=%.3fms p95=%.3fms\n",
                    L,static_cast<unsigned long long>(GeneratedByLevel[L]),DesiredByLevel[L],
                    static_cast<double>(SamplesByLevel[L])/1.0e6,WorkerSecondsByLevel[L],Average*1000.0,P50*1000.0,P95*1000.0);
                if(bProfileAttribution)
                {
                    std::printf("[TileProfile] LOD=%d",L);
                    for(size_t I=0;I<VoxelDensityProfile::BucketCount;++I)
                    {
                        const double Ms=static_cast<double>(ProfileCyclesByLevel[L][I])
                            *FPlatformTime::GetSecondsPerCycle64()*1000.0;
                        if(Ms>0.001)std::printf(" %s=%.3fms",
                            VoxelDensityProfile::BucketName(static_cast<VoxelDensityProfile::EBucket>(I)),Ms);
                    }
                    std::printf("\n");
                }
                Times.clear();
            }
        }

        void ReportCrossingWork()
        {
            for(auto& Pair:CrossingWork)
            {
                FCrossingWork& Work=Pair.second;
                if(Work.bReported||Work.PendingJobs!=0)continue;
                Work.bReported=true;
                const double AverageMs=Work.FinishedJobs>0
                    ?Work.WorkerSeconds*1000.0/static_cast<double>(Work.FinishedJobs):0.0;
                Work.WallSeconds=Work.StartTime.time_since_epoch().count()!=0
                    ?std::chrono::duration<double>(std::chrono::steady_clock::now()-Work.StartTime).count():0.0;
                const double NearestReadyMs=Work.NearestReadySeconds>=0.0
                    ?Work.NearestReadySeconds*1000.0:-1.0;
                std::printf("[ChunkCrossingWork] id=%llu submitted=%llu finished=%llu generated=%llu cancelled_queued=%llu stale=%llu samples=%.2fM page_covered_tiles=%llu page_regions=%llu page_nodes=%llu core_page=%.4fs core_page_worker=%.4fs nearest_ready=%.3fms nearest_tile=(%d,%d,%d,lod%d) worker_sum=%.3fs wall=%.3fs avg=%.3fms\n",
                    static_cast<unsigned long long>(Pair.first),
                    static_cast<unsigned long long>(Work.SubmittedJobs),
                    static_cast<unsigned long long>(Work.FinishedJobs),
                    static_cast<unsigned long long>(Work.Tiles),
                    static_cast<unsigned long long>(Work.CancelledQueuedJobs),
                    static_cast<unsigned long long>(Work.StaleResults),
                    static_cast<double>(Work.Samples)/1.0e6,
                    static_cast<unsigned long long>(Work.PageCoveredTiles),
                    static_cast<unsigned long long>(Work.PageRegions),
                    static_cast<unsigned long long>(Work.PageLatticeNodes),
                    Work.CorePageSeconds,Work.CorePageWorkerSeconds,NearestReadyMs,
                    Work.NearestNewTile.Coord.X,Work.NearestNewTile.Coord.Y,
                    Work.NearestNewTile.Coord.Z,Work.NearestNewTile.Level,
                    Work.WorkerSeconds,Work.WallSeconds,AverageMs);
                for(int32 Level=0;Level<=MaxLevels;++Level)
                    if(Work.TilesByLevel[Level]>0)
                        std::printf("[ChunkCrossingLevel] id=%llu lod=%d tiles=%llu samples=%.3fM worker_sum=%.4fs avg=%.3fms\n",
                            static_cast<unsigned long long>(Pair.first),Level,
                            static_cast<unsigned long long>(Work.TilesByLevel[Level]),
                            static_cast<double>(Work.SamplesByLevel[Level])/1.0e6,
                            Work.WorkerSecondsByLevel[Level],
                            Work.WorkerSecondsByLevel[Level]*1000.0/
                                static_cast<double>(Work.TilesByLevel[Level]));
            }
        }

        void PumpGeneration()
        {
            if(bRebuildPending){StartGeneration();return;}
            if(!Streamer)return;
            std::vector<FStandaloneCpuTile> Ready=Streamer->TakeReady();
            for(FStandaloneCpuTile& Tile:Ready)
            {
                const int32 Level=Tile.Key.Level;
                FCrossingWork& Crossing=CrossingWork[Tile.RequestedCrossing];
                if(Crossing.PendingJobs>0)--Crossing.PendingJobs;
                ++Crossing.FinishedJobs;Crossing.Samples+=Tile.DensitySamples;Crossing.WorkerSeconds+=Tile.WorkerSeconds;
                if(Level>=0&&Level<=MaxLevels)
                {
                    ++Crossing.TilesByLevel[Level];
                    Crossing.SamplesByLevel[Level]+=Tile.DensitySamples;
                    Crossing.WorkerSecondsByLevel[Level]+=Tile.WorkerSeconds;
                }
                if(Crossing.bHasNearestNewTile&&Tile.Key==Crossing.NearestNewTile
                    &&Crossing.NearestReadySeconds<0.0)
                    Crossing.NearestReadySeconds=std::chrono::duration<double>(
                        Tile.CompletedAt-Crossing.StartTime).count();
                if(!Streamer->IsDesired(Tile.Key)||ResidentTiles.contains(Tile.Key))
                {++Crossing.StaleResults;continue;}
                ++Crossing.Tiles;
                ++GeneratedByLevel[Level];SamplesByLevel[Level]+=Tile.DensitySamples;WorkerSecondsByLevel[Level]+=Tile.WorkerSeconds;
                if(bProfileAttribution)for(size_t I=0;I<Tile.ProfileCycles.size();++I)
                    ProfileCyclesByLevel[Level][I]+=Tile.ProfileCycles[I];
                TileSecondsByLevel[Level].push_back(Tile.WorkerSeconds);
                FResidentTileMesh Resident;Resident.Vertices=std::move(Tile.Vertices);Resident.Indices=std::move(Tile.Indices);
                Resident.TriangleSkirtFaceMasks=std::move(Tile.TriangleSkirtFaceMasks);
                Resident.ActiveSkirtFaceMask=ComputeVisibleSkirtFaceMask(Tile.Key,DesiredTileSet);
                Resident.Samples=Tile.DensitySamples;Resident.WorkerSeconds=Tile.WorkerSeconds;
                ResidentTiles.emplace(Tile.Key,std::move(Resident));
                const FStandaloneBatchKey Batch=BatchForTile(Tile.Key);
                BatchMembers[Batch].push_back(Tile.Key);MarkBatchDirty(Batch);
            }
            ReportCrossingWork();
            UploadDirtyBatches();
            if(!DeferredLeaverTiles.empty()&&DirtyBatchSet.empty())
            {
                bool bDesiredResident=true;
                for(const FVoxelTileKey& Key:DesiredTiles)
                    if(!IsRenderedTileReady(Key)){bDesiredResident=false;break;}
                if(bDesiredResident)
                {
                    std::vector<FVoxelTileKey> ReadyToDrop;
                    for(const FVoxelTileKey& Key:DeferredLeaverTiles)
                    {
                        if(Streamer->IsDesired(Key))continue;
                        if(ResidentTiles.contains(Key))ReadyToDrop.push_back(Key);
                    }
                    for(const FVoxelTileKey& Key:ReadyToDrop)RemoveResidentTile(Key);
                    DeferredLeaverTiles.clear();
                }
            }
            if(bShapePipelineActive)TryCompleteShapePipeline();
            if(FlightTargetCrossings>0&&bFlightFinished&&!bFlightClosePosted
                &&DirtyBatchSet.empty()&&DeferredLeaverTiles.empty())
            {
                bool bAllFlightWorkReported=true;
                bool bCrossingLatencyUnderTarget=true;
                double MaxCrossingWallSeconds=0.0,MaxWorkerCpuSeconds=0.0;
                std::vector<double> NearestReadyMilliseconds;
                for(uint64 Crossing=FlightStartCrossingIndex+1;
                    Crossing<=CrossingIndex;++Crossing)
                {
                    const auto It=CrossingWork.find(Crossing);
                    if(It==CrossingWork.end()||!It->second.bReported)
                    {bAllFlightWorkReported=false;continue;}
                    const FCrossingWork& Work=It->second;
                    MaxCrossingWallSeconds=std::max(MaxCrossingWallSeconds,Work.WallSeconds);
                    MaxWorkerCpuSeconds=std::max(MaxWorkerCpuSeconds,Work.WorkerSeconds);
                    if(Work.WallSeconds>=0.200)bCrossingLatencyUnderTarget=false;
                    if(Work.NearestReadySeconds>=0.0)
                        NearestReadyMilliseconds.push_back(Work.NearestReadySeconds*1000.0);
                }
                if(bAllFlightWorkReported)
                {
                    bFlightClosePosted=true;
                    const double Elapsed=std::chrono::duration<double>(
                        std::chrono::steady_clock::now()-FlightStartTime).count();
                    const uint64 CacheHits=Streamer?Streamer->GetCacheHits():0;
                    std::size_t DesiredResidentCount=0;
                    for(const FVoxelTileKey& Key:DesiredTiles)
                        if(IsRenderedTileReady(Key))++DesiredResidentCount;
                    std::sort(NearestReadyMilliseconds.begin(),NearestReadyMilliseconds.end());
                    const auto Percentile=[&](std::size_t Percent)
                    {
                        if(NearestReadyMilliseconds.empty())return -1.0;
                        const std::size_t Index=(NearestReadyMilliseconds.size()-1)*Percent/100;
                        return NearestReadyMilliseconds[Index];
                    };
                    const double NearestMaximum=NearestReadyMilliseconds.empty()?-1.0:
                        NearestReadyMilliseconds.back();
                    const std::size_t MissingDesired=DesiredTiles.size()-DesiredResidentCount;
                    const bool bPass=FlightObservedCrossings==FlightTargetCrossings
                        &&bFlightCrossingsValid&&CacheHits==0&&bCrossingLatencyUnderTarget
                        &&MissingDesired==0&&DeferredLeaverTiles.empty()&&DirtyBatchSet.empty();
                    std::printf("[FlightBenchmarkComplete] crossings=%d/%d speed=%.2fm/s elapsed=%.3fs cache_hits=%llu crossing_deltas=%s nearest_ready_p50=%.3fms nearest_ready_p95=%.3fms nearest_ready_max=%.3fms max_crossing_wall=%.3fs max_worker_cpu_sum=%.3fs resident=%zu/%zu missing_tiles=%zu deferred_leavers=%zu uploads_drained=%s status=%s\n",
                        FlightObservedCrossings,FlightTargetCrossings,FlightSpeedMps,Elapsed,
                        static_cast<unsigned long long>(CacheHits),
                        bFlightCrossingsValid?"unit-+Y":"invalid",
                        Percentile(50),Percentile(95),NearestMaximum,MaxCrossingWallSeconds,
                        MaxWorkerCpuSeconds,DesiredResidentCount,DesiredTiles.size(),MissingDesired,
                        DeferredLeaverTiles.size(),DirtyBatchSet.empty()?"yes":"no",
                        bPass?"PASS":"FAIL");
                    std::fflush(stdout);
                    PostMessageA(Window,WM_CLOSE,0,0);
                }
            }
            if(!bShapePipelineActive&&!bFullDetailReported&&IsFullDetailUploaded())
            {
                bFullDetailReported=true;
                const double Elapsed=std::chrono::duration<double>(std::chrono::steady_clock::now()-ColdProcessStart).count();
                const std::size_t DesiredNear=static_cast<std::size_t>(std::count_if(
                    DesiredTiles.begin(),DesiredTiles.end(),[&](const FVoxelTileKey& Key)
                        {return Key.Level<World->FullResClipLevels;}));
                std::printf("[FullDetailReady] %.3f s after cold process start; rendered=%zu/%zu tiles through LOD%d.\n",
                    Elapsed,DesiredNear,DesiredNear,World->FullResClipLevels-1);
                std::printf("[ColdTimeline] event=near_all_tiles_ready at_ms=%.3f waits_for=all_%zu_LOD0_1_gpu_or_cpu_tiles\n",
                    Elapsed*1000.0,DesiredNear);
                ReportTileTimings();std::fflush(stdout);
                StartFarShapeGeneration();
            }
            size_t DesiredResidentCount=0;
            for(const FVoxelTileKey& Key:DesiredTiles)
                if(IsRenderedTileReady(Key))++DesiredResidentCount;
            if(!bInitialClipmapReported&&DesiredResidentCount==DesiredTiles.size()&&DirtyBatchSet.empty())
            {
                bInitialClipmapReported=true;
                const double Elapsed=std::chrono::duration<double>(std::chrono::steady_clock::now()-ColdProcessStart).count();
                std::printf("[InitialClipmapReady] %.3f s after cold process start; %zu tiles, %llu vertices, %llu indices, %zu GPU batches.\n",
                    Elapsed,DesiredResidentCount,static_cast<unsigned long long>(TotalVertices),
                    static_cast<unsigned long long>(TotalIndices),StreamingBatches.size());
                std::printf("[ColdTimeline] event=whole_clipmap_ready at_ms=%.3f waits_for=%zu_GPU_tiles_plus_CPU_fallback_and_uploads\n",
                    Elapsed*1000.0,GpuFarEligibleTiles.size());
                std::printf("[PresentStartupSummary] nonblocking_startup_calls=%llu skipped_busy_calls=%llu gpu_warmup_held_calls=%llu policy=do_not_wait_until_initial_clipmap_ready\n",
                    static_cast<unsigned long long>(StartupNonBlockingPresentCalls),
                    static_cast<unsigned long long>(StartupNonBlockingPresentSkips),
                    static_cast<unsigned long long>(GpuWarmupHeldPresents));
                if(bGpuFarEnabled)
                {
                    SampleGpuFarMemory();
                    const double AverageGpuMs=GpuFarTilesSubmitted
                        ?GpuFarGpuDispatchMs/static_cast<double>(GpuFarTilesSubmitted):0.0;
                    std::printf("[GpuFarTiming] submitted_tiles=%llu submission_batches=%llu max_tiles_per_batch=%u pooled_output_reserve_wall_ms=%.3f cpu_input_prepare_wall_ms=%.3f cpu_input_worker_sum_ms=%.3f cpu_upload_enqueue_sum_ms=%.3f gpu_dispatch_timestamp_total_ms=%.3f gpu_dispatch_mean_per_tile_ms=%.6f gpu_dispatch_max_tile_ms=%.6f gpu_submission_span_total_ms=%.3f gpu_submission_span_max_batch_ms=%.3f frame_budget_90hz_ms=11.111\n",
                        static_cast<unsigned long long>(GpuFarTilesSubmitted),
                        static_cast<unsigned long long>(GpuFarSubmissionBatches),GpuFarMaxTilesPerSubmissionBatch,
                        GpuFarPoolReserveMs,GpuFarPreparationWallMs,GpuFarPreparationWorkerMs,
                        GpuFarEnqueueWallSeconds*1000.0,
                        GpuFarGpuDispatchMs,AverageGpuMs,GpuFarGpuDispatchMaxMs,
                        GpuFarSubmissionSpanMs,GpuFarSubmissionSpanMaxMs);
                    std::printf("[GpuFarTimeline] pump_calls=%llu pump_first_ms=%.3f pump_last_ms=%.3f pump_gap_sum_ms=%.3f pump_gap_max_ms=%.3f pump_cpu_sum_ms=%.3f pump_cpu_max_ms=%.3f submit_first_ms=%.3f submit_last_ms=%.3f completion_polls=%llu completion_first_ms=%.3f completion_last_ms=%.3f queue_envelope_ms=%.3f gpu_busy_ms=%.3f estimated_idle_or_poll_ms=%.3f\n",
                        static_cast<unsigned long long>(GpuFarPumpCalls),GpuFarFirstPumpMs,GpuFarLastPumpMs,
                        GpuFarPumpGapMs,GpuFarPumpGapMaxMs,GpuFarPumpWallMs,GpuFarPumpMaxWallMs,
                        GpuFarFirstSubmitMs,GpuFarLastSubmitMs,
                        static_cast<unsigned long long>(GpuFarCompletionPolls),GpuFarFirstCompletionPollMs,
                        GpuFarLastCompletionPollMs,
                        GpuFarFirstSubmitMs>=0.0&&GpuFarLastCompletionPollMs>=0.0
                            ?GpuFarLastCompletionPollMs-GpuFarFirstSubmitMs:0.0,
                        GpuFarSubmissionSpanMs,
                        GpuFarFirstSubmitMs>=0.0&&GpuFarLastCompletionPollMs>=0.0
                            ?std::max(0.0,GpuFarLastCompletionPollMs-GpuFarFirstSubmitMs-GpuFarSubmissionSpanMs):0.0);
                    std::printf("[GpuFarMemory] dxgi_local_sampled_peak_mb=%.1f dxgi_local_budget_mb=%.1f\n",
                        GpuFarLocalMemoryPeakBytes/(1024.0*1024.0),
                        GpuFarLocalMemoryBudgetBytes/(1024.0*1024.0));
                }
                ReportTileTimings();std::fflush(stdout);
                if(bQuitAfterClipmap&&CapturePath.empty())PostMessageA(Window,WM_CLOSE,0,0);
            }
            if(bInitialClipmapReported&&bGpuFarAuditEnabled&&GpuFarAudit&&!bGpuFarAuditReported)
            {
                GpuFarAudit->Finalize();GpuFarAudit->Print();
                bGpuFarAuditReported=true;
                std::printf("[GpuFarSurvivalAudit] mesh_source=GPU field_readback=required status=%s\n",
                    GpuFarAudit->Passing()?"PASS":"FAIL");
                std::fflush(stdout);
                if(bQuitAfterClipmap&&CapturePath.empty())PostMessageA(Window,WM_CLOSE,0,0);
            }
            bGenerationRunning=DesiredResidentCount!=DesiredTiles.size()||!DirtyBatchSet.empty()
                ||(bGpuFarEnabled&&!GpuFarQueue.empty());
        }

        void CaptureRenderedFrame()
        {
            const bool bReadyToCapture=bCaptureAfterClipmap?bInitialClipmapReported:IsFullDetailUploaded();
            if(CapturePath.empty()||bAutoCaptureWritten||!bReadyToCapture)return;
            bAutoCaptureWritten=SaveBackBufferPng(CapturePath);
            if(bAutoCaptureWritten)
            {
                const double Elapsed=std::chrono::duration<double>(std::chrono::steady_clock::now()-GenerationStart).count();
                std::printf("Back-buffer capture saved: %s (%.3f s after generation start; full detail uploaded).\n",CapturePath.c_str(),Elapsed);
                std::fflush(stdout);PostMessageA(Window,WM_CLOSE,0,0);
            }
            else std::fprintf(stderr,"Back-buffer capture failed: %s\n",CapturePath.c_str());
        }

        void Capture(bool bOn)
        {
            bCaptured=bOn;
            if(bOn){SetCapture(Window);ShowCursor(FALSE);GetCursorPos(&LastMouse);POINT C{Width/2,Height/2};ClientToScreen(Window,&C);SetCursorPos(C.x,C.y);}
            else{ReleaseCapture();ShowCursor(TRUE);}
        }

        void UpdateTitle()
        {
            uint64 VisibleIndices=0;for(const auto& Pair:StreamingBatches)
                if(bShowLevel[Pair.first.Level]
                    &&(CaptureOnlyLod<0||Pair.first.Level==CaptureOnlyLod))
                    VisibleIndices+=Pair.second.IndexCount;
            std::ostringstream S;S<<"VoxelForge standalone | seed "<<Seed<<" | "<<(bShapeWorld?"shape-interval owner world":(bRoundTunnels?"round/winding research tunnels":"base world"))
                <<(bRoundTunnels?(bRoundTunnelEnhanced?" (detailed walls)":" (legacy walls)"):"")
                <<" | "<<VisibleIndices/3<<" tris | "<<FrameMs<<" ms | field "<<(bGpuFarEnabled?"GPU":"CPU")
                <<" | "<<(bGenerationRunning?"generating":"ready")
                <<" | pos "<<std::fixed<<std::setprecision(1)<<Position.X<<", "<<Position.Y<<", "<<Position.Z<<" m";
            SetWindowTextA(Window,S.str().c_str());
        }

        void Draw()
        {
            const auto DrawStart=std::chrono::steady_clock::now();
            const double InterDrawGapMs=LastDrawEndAt.time_since_epoch().count()!=0
                ?std::chrono::duration<double,std::milli>(DrawStart-LastDrawEndAt).count():0.0;
            PumpGeneration();
            const auto CpuPumpEnd=std::chrono::steady_clock::now();
            const bool bDeferGpuUntilCpuFrame=bGpuFarEnabled&&bPresentCpuFrameBeforeGpu
                &&!bCpuReadyFramePresentedBeforeGpu;
            if(!bDeferGpuUntilCpuFrame)PumpGpuFarGeneration();
            const auto GpuPumpEnd=std::chrono::steady_clock::now();
            const auto Now=std::chrono::steady_clock::now();
            if(FlightTargetCrossings>0&&bFullDetailReported&&!bFlightStarted)
            {
                bFlightStarted=true;FlightStartTime=Now;FlightStartPosition=Position;
                FlightStartCrossingIndex=CrossingIndex;FlightObservedCrossings=0;
                bFlightCrossingsValid=true;
                std::printf("[FlightBenchmarkStart] crossings=%d speed=%.2fm/s direction=+Y start=(%.2f,%.2f,%.2f)m cache_mode=none\n",
                    FlightTargetCrossings,FlightSpeedMps,Position.X,Position.Y,Position.Z);
                std::fflush(stdout);
            }
            if(FlightTargetCrossings>0&&bFlightStarted&&!bFlightFinished)
            {
                const double Elapsed=std::chrono::duration<double>(Now-FlightStartTime).count();
                Position=V3(FlightStartPosition.X,FlightStartPosition.Y+
                    static_cast<float>(FlightSpeedMps*Elapsed),FlightStartPosition.Z);
                if(FlightObservedCrossings>=FlightTargetCrossings)
                {
                    bFlightFinished=true;
                    std::printf("[FlightBenchmarkMove] movement stopped after %d center-chunk crossings; waiting for new nearest tiles, meshes and uploads.\n",
                        FlightObservedCrossings);
                    std::fflush(stdout);
                }
            }
            const float Dt=std::clamp(std::chrono::duration<float>(Now-PreviousFrame).count(),0.0f,0.1f);
            PreviousFrame=Now;FrameMs=Dt*1000.0;
            const V3 F=Forward(),R=Normalize(Cross(F,{0,0,1})),U=Normalize(Cross(R,F));
            const float Step=MoveSpeed*Dt*(Keys[VK_SHIFT]?4.0f:1.0f);
            if(Keys['W'])Position=Add(Position,Mul(F,Step));if(Keys['S'])Position=Sub(Position,Mul(F,Step));
            if(Keys['A'])Position=Sub(Position,Mul(R,Step));if(Keys['D'])Position=Add(Position,Mul(R,Step));
            if(Keys[VK_SPACE]||Keys['E'])Position.Z+=Step;if(Keys[VK_CONTROL]||Keys['Q'])Position.Z-=Step;
            if(Streamer&&NeedsRecenter(BuildTargetChunk))RecenterStreaming();
            const float Clear[4]={0.055f,0.075f,0.105f,1.0f};
            Context->OMSetRenderTargets(1,&RenderTarget,DepthView);Context->ClearRenderTargetView(RenderTarget,Clear);
            Context->ClearDepthStencilView(DepthView,D3D11_CLEAR_DEPTH,1.0f,0);
            D3D11_VIEWPORT Viewport{0,0,static_cast<float>(Width),static_cast<float>(Height),0,1};Context->RSSetViewports(1,&Viewport);
            Context->RSSetState(bWireframe?WireRasterizer:SolidRasterizer);Context->IASetInputLayout(InputLayout);
            Context->IASetPrimitiveTopology(D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST);
            FConstants C{};const M4 V=View(Position,F,R,U);const M4 P=Perspective(1.04719755f,static_cast<float>(Width)/std::max(1,Height),0.05f,5000.0f);
            M4 Combined{};for(int I=0;I<4;++I)for(int J=0;J<4;++J)for(int K=0;K<4;++K)Combined.M[I][J]+=V.M[I][K]*P.M[K][J];
            C.ViewProjection=Transpose(Combined);C.Camera[0]=Position.X;C.Camera[1]=Position.Y;C.Camera[2]=Position.Z;
            C.Forward[0]=F.X;C.Forward[1]=F.Y;C.Forward[2]=F.Z;C.LightDirection[0]=-0.5f;C.LightDirection[1]=-0.72f;C.LightDirection[2]=-0.48f;
            C.FogStart=260.0f;C.FogEnd=2600.0f;C.Ambient=0.36f;
            C.Padding[0]=1.0f;
            Context->UpdateSubresource(Constants,0,nullptr,&C,0,0);Context->VSSetConstantBuffers(0,1,&Constants);Context->PSSetConstantBuffers(0,1,&Constants);
            Context->VSSetShader(VertexShader,nullptr,0);Context->PSSetShader(PixelShader,nullptr,0);
            for(const auto& Pair:StreamingBatches)
            {
                const FStreamingGpuBatch& L=Pair.second;
                if(!bShowLevel[Pair.first.Level]
                    ||(CaptureOnlyLod>=0&&Pair.first.Level!=CaptureOnlyLod)
                    ||!L.VertexBuffer||!L.IndexBuffer||!L.IndexCount
                    ||(bDrawOnlyTile&&L.DrawOnlyIndexCount==0))continue;
                const UINT Stride=sizeof(FGpuVertex),Offset=0;Context->IASetVertexBuffers(0,1,&L.VertexBuffer,&Stride,&Offset);
                Context->IASetIndexBuffer(L.IndexBuffer,DXGI_FORMAT_R32_UINT,0);
                if(bDrawOnlyTile)
                    Context->DrawIndexed(L.DrawOnlyIndexCount,L.DrawOnlyIndexStart,0);
                else Context->DrawIndexed(L.IndexCount,0,0);
            }
            // Avoid issuing hundreds of tiny indirect draw calls while the
            // initial far batches are still being built. Once the clipmap is
            // ready the pooled GPU meshes draw directly as before.
            if(bGpuFarEnabled&&ShapeWorld&&ShapeWorld->hasFarWorld())
            {
                C.Padding[0]=0.01f;
                Context->UpdateSubresource(Constants,0,nullptr,&C,0,0);
                Context->VSSetConstantBuffers(0,1,&Constants);Context->PSSetConstantBuffers(0,1,&Constants);
                for(auto& Pair:GpuFarTiles)
                {
                    FGpuFarTile& Tile=Pair.second;
                    if(!IsDrawOnlyTile(Pair.first)||!bShowLevel[Pair.first.Level]
                        ||(CaptureOnlyLod>=0&&Pair.first.Level!=CaptureOnlyLod)
                        ||!Tile.VertexBuffer||!Tile.IndexBuffer||!Tile.IndirectArgs)continue;
                    if(!bInitialClipmapReported&&Pair.first.Level>=World->FullResClipLevels)continue;
                    if(!IsRenderedTileReady(Pair.first))continue;
                    if(bAuditGpuSkirtDrawArgs&&Tile.VisibleSkirtFaceMask!=0&&!Tile.bSkirtDrawArgsAudited)
                    {
                        if(!GpuSkirtArgsStaging)
                        {
                            D3D11_BUFFER_DESC StagingDesc{};
                            StagingDesc.ByteWidth=11u*sizeof(uint32);
                            StagingDesc.Usage=D3D11_USAGE_STAGING;
                            StagingDesc.CPUAccessFlags=D3D11_CPU_ACCESS_READ;
                            const HRESULT CreateResult=Device->CreateBuffer(&StagingDesc,nullptr,
                                &GpuSkirtArgsStaging);
                            if(FAILED(CreateResult))
                                std::fprintf(stderr,"[GpuSkirtDrawAuditError] create_staging_hr=0x%08lx\n",
                                    static_cast<unsigned long>(CreateResult));
                        }
                        if(GpuSkirtArgsStaging)
                        {
                            const D3D11_BOX ArgsSource{Tile.IndirectArgsByteOffset,0,0,
                                Tile.IndirectArgsByteOffset+11u*sizeof(uint32),1,1};
                            Context->CopySubresourceRegion(GpuSkirtArgsStaging,0,0,0,0,
                                Tile.IndirectArgs,0,&ArgsSource);
                            D3D11_MAPPED_SUBRESOURCE ArgsMapped{};
                            const HRESULT MapResult=Context->Map(GpuSkirtArgsStaging,0,
                                D3D11_MAP_READ,0,&ArgsMapped);
                            if(SUCCEEDED(MapResult))
                            {
                                const auto* Args=static_cast<const uint32*>(ArgsMapped.pData);
                                const uint32 ExpectedStartIndex=(Tile.IndirectArgsByteOffset
                                    /(11u*sizeof(uint32)))*32768u;
                                ++GpuSkirtDrawAuditTiles;
                                GpuSkirtDrawAuditSumIndices+=Args[5];
                                GpuSkirtDrawAuditMaxIndices=std::max<uint64>(GpuSkirtDrawAuditMaxIndices,Args[5]);
                                if(Args[5]==0)++GpuSkirtDrawAuditZeroCounts;
                                if(Args[10]!=0)++GpuSkirtDrawAuditOverflow;
                                if(Args[7]!=ExpectedStartIndex)++GpuSkirtDrawAuditBadStartIndex;
                                std::printf("[GpuSkirtDrawArgs] key=(%d,%d,%d,lod%d) face_mask=0x%02x output_slot=%u surface_indices=%u skirt_panel_indices=%u instances=%u connector_start_index=%u expected_start_index=%u overflow=%u\n",
                                    Pair.first.Coord.X,Pair.first.Coord.Y,Pair.first.Coord.Z,Pair.first.Level,
                                    static_cast<unsigned>(Tile.VisibleSkirtFaceMask),Tile.OutputSlot,
                                    Args[0],Args[5],Args[6],Args[7],ExpectedStartIndex,Args[10]);
                                Context->Unmap(GpuSkirtArgsStaging,0);
                            }
                            else std::fprintf(stderr,"[GpuSkirtDrawAuditError] map_staging_hr=0x%08lx\n",
                                static_cast<unsigned long>(MapResult));
                        }
                        Tile.bSkirtDrawArgsAudited=true;
                    }
                    const UINT Stride=32,Offset=0;
                    Context->IASetVertexBuffers(0,1,&Tile.VertexBuffer,&Stride,&Offset);
                    Context->IASetIndexBuffer(Tile.IndexBuffer,DXGI_FORMAT_R32_UINT,0);
                    Context->DrawIndexedInstancedIndirect(Tile.IndirectArgs,Tile.IndirectArgsByteOffset);
                    if(Tile.ConnectorVertexBuffer&&Tile.ConnectorIndexBuffer)
                    {
                        Context->IASetVertexBuffers(0,1,&Tile.ConnectorVertexBuffer,&Stride,&Offset);
                        Context->IASetIndexBuffer(Tile.ConnectorIndexBuffer,DXGI_FORMAT_R32_UINT,0);
                        Context->DrawIndexedInstancedIndirect(Tile.IndirectArgs,
                            Tile.IndirectArgsByteOffset+5u*sizeof(uint32));
                    }
                }
            }
            if(bAuditGpuSkirtDrawArgs&&bInitialClipmapReported&&!bGpuSkirtDrawAuditReported)
            {
                std::printf("[GpuSkirtDrawArgsSummary] tiles=%llu zero_index_counts=%llu overflow=%llu bad_slot_start_indices=%llu total_panel_plus_skirt_indices=%llu max_panel_plus_skirt_indices=%llu status=%s\n",
                    static_cast<unsigned long long>(GpuSkirtDrawAuditTiles),
                    static_cast<unsigned long long>(GpuSkirtDrawAuditZeroCounts),
                    static_cast<unsigned long long>(GpuSkirtDrawAuditOverflow),
                    static_cast<unsigned long long>(GpuSkirtDrawAuditBadStartIndex),
                    static_cast<unsigned long long>(GpuSkirtDrawAuditSumIndices),
                    static_cast<unsigned long long>(GpuSkirtDrawAuditMaxIndices),
                    (GpuSkirtDrawAuditTiles>0&&GpuSkirtDrawAuditOverflow==0
                        &&GpuSkirtDrawAuditBadStartIndex==0)?"PASS":"FAIL");
                bGpuSkirtDrawAuditReported=true;
                std::fflush(stdout);
            }
            CaptureRenderedFrame();
            const bool bHoldFrontBufferForGpuWarmup=bHoldCpuFrameDuringGpuWarmup
                &&bCpuReadyFramePresentedBeforeGpu&&bGpuFarEnabled
                &&GpuFarTilesSubmitted>0&&!bInitialClipmapReported;
            if(bHoldFrontBufferForGpuWarmup)
            {
                if(!bGpuWarmupHoldReported)
                {
                    bGpuWarmupHoldReported=true;
                    std::printf("[GpuWarmupPresentHold] begin_ms=%.3f policy=retain_last_CPU_ready_front_buffer_until_GPU_clipmap_ready\n",
                        std::chrono::duration<double,std::milli>(
                            std::chrono::steady_clock::now()-ColdProcessStart).count());
                    std::fflush(stdout);
                }
                ++GpuWarmupHeldPresents;
                // Keep pumping CPU and GPU work at the display cadence without
                // calling the driver present path that stalls on its first GPU frame.
                std::this_thread::sleep_for(std::chrono::milliseconds(11));
                LastDrawEndAt=std::chrono::steady_clock::now();
                return;
            }
            const auto BeforePresent=std::chrono::steady_clock::now();
            const bool bNonBlockingStartupPresent=bGpuFarEnabled&&!bInitialClipmapReported
                &&!bForceBlockingStartupPresent;
            const UINT PresentSyncInterval=(bDisableVSync||bNonBlockingStartupPresent)?0u:1u;
            const UINT PresentFlags=bNonBlockingStartupPresent?DXGI_PRESENT_DO_NOT_WAIT:0u;
            const auto PresentCallStart=std::chrono::steady_clock::now();
            const HRESULT PresentResult=SwapChain->Present(PresentSyncInterval,PresentFlags);
            const auto PresentCallEnd=std::chrono::steady_clock::now();
            if(GpuFarTilesSubmitted>0&&!bGpuFirstBackedPresentReported)
            {
                bGpuFirstBackedPresentReported=true;
                std::printf("[GpuFirstBackedPresent] at_ms=%.3f call_ms=%.3f sync=%u flags=0x%08x hr=0x%08lx submitted_tiles=%llu batches=%llu\n",
                    std::chrono::duration<double,std::milli>(PresentCallEnd-ColdProcessStart).count(),
                    std::chrono::duration<double,std::milli>(PresentCallEnd-PresentCallStart).count(),
                    PresentSyncInterval,PresentFlags,static_cast<unsigned long>(PresentResult),
                    static_cast<unsigned long long>(GpuFarTilesSubmitted),
                    static_cast<unsigned long long>(GpuFarSubmissionBatches));
                std::fflush(stdout);
            }
            if(bInitialClipmapReported&&!bWholeViewPresented)
            {
                bWholeViewPresented=true;
                std::printf("[WholeViewPresented] at_ms=%.3f present_ms=%.3f sync=%u hr=0x%08lx\n",
                    std::chrono::duration<double,std::milli>(PresentCallEnd-ColdProcessStart).count(),
                    std::chrono::duration<double,std::milli>(PresentCallEnd-PresentCallStart).count(),
                    PresentSyncInterval,static_cast<unsigned long>(PresentResult));
                std::fflush(stdout);
            }
            if(bGpuWarmupHoldReported&&!bGpuWarmupReleaseReported)
            {
                bGpuWarmupReleaseReported=true;
                std::printf("[GpuWarmupPresentRelease] at_ms=%.3f first_post_warmup_present_ms=%.3f sync=%u hr=0x%08lx\n",
                    std::chrono::duration<double,std::milli>(PresentCallEnd-ColdProcessStart).count(),
                    std::chrono::duration<double,std::milli>(PresentCallEnd-PresentCallStart).count(),
                    PresentSyncInterval,static_cast<unsigned long>(PresentResult));
                std::fflush(stdout);
            }
            const uint64 PresentOrdinal=++PresentCallCount;
            if(bNonBlockingStartupPresent)
            {
                ++StartupNonBlockingPresentCalls;
                if(PresentResult==DXGI_ERROR_WAS_STILL_DRAWING)++StartupNonBlockingPresentSkips;
                else if(FAILED(PresentResult))
                    std::fprintf(stderr,"[PresentStartupError] hr=0x%08lx\n",
                        static_cast<unsigned long>(PresentResult));
            }
            if(PresentOrdinal<=3)
                std::printf("[PresentProfile] ordinal=%llu at_ms=%.3f sync=%u flags=0x%08x mode=%s hr=0x%08lx call_ms=%.3f gpu_tiles_submitted=%llu gpu_batches=%llu\n",
                    static_cast<unsigned long long>(PresentOrdinal),
                    std::chrono::duration<double,std::milli>(PresentCallEnd-ColdProcessStart).count(),
                    PresentSyncInterval,PresentFlags,bNonBlockingStartupPresent?"startup-do-not-wait":"normal",
                    static_cast<unsigned long>(PresentResult),
                    std::chrono::duration<double,std::milli>(PresentCallEnd-PresentCallStart).count(),
                    static_cast<unsigned long long>(GpuFarTilesSubmitted),
                    static_cast<unsigned long long>(GpuFarSubmissionBatches));
            if(bDeferGpuUntilCpuFrame&&!StreamingBatches.empty())
            {
                bCpuReadyFramePresentedBeforeGpu=true;
                std::printf("[CpuReadyFrameBeforeGpu] at_ms=%.3f uploaded_cpu_batches=%zu next_action=submit_gpu_far\n",
                    std::chrono::duration<double,std::milli>(PresentCallEnd-ColdProcessStart).count(),
                    StreamingBatches.size());
                std::fflush(stdout);
            }
            UpdateTitle();
            const auto DrawEnd=std::chrono::steady_clock::now();
            const double CpuPumpMs=std::chrono::duration<double,std::milli>(CpuPumpEnd-DrawStart).count();
            const double GpuPumpMs=std::chrono::duration<double,std::milli>(GpuPumpEnd-CpuPumpEnd).count();
            const double PresentCallMs=std::chrono::duration<double,std::milli>(PresentCallEnd-PresentCallStart).count();
            const double TitleMs=std::chrono::duration<double,std::milli>(DrawEnd-PresentCallEnd).count();
            const double RenderPresentMs=PresentCallMs+TitleMs;
            if(InterDrawGapMs>50.0||CpuPumpMs>50.0||GpuPumpMs>50.0||RenderPresentMs>50.0)
                std::printf("[LongViewerFrame] at_ms=%.3f inter_draw_gap_ms=%.3f cpu_pump_ms=%.3f gpu_pump_ms=%.3f pre_present_ms=%.3f present_sync=%u present_call_ms=%.3f title_ms=%.3f\n",
                    std::chrono::duration<double,std::milli>(DrawStart-ColdProcessStart).count(),
                    InterDrawGapMs,CpuPumpMs,GpuPumpMs,
                    std::chrono::duration<double,std::milli>(BeforePresent-GpuPumpEnd).count(),
                    PresentSyncInterval,PresentCallMs,TitleMs);
            LastDrawEndAt=DrawEnd;
        }
    };

    FApp* GApp=nullptr;
    LRESULT CALLBACK WindowProc(HWND H,UINT Message,WPARAM W,LPARAM L)
    {
        if(!GApp)return DefWindowProc(H,Message,W,L);
        switch(Message)
        {
        case WM_SIZE:if(W!=SIZE_MINIMIZED)GApp->Resize(LOWORD(L),HIWORD(L));return 0;
        case WM_KEYDOWN:
        {
            const bool bWasDown=(L&(1ll<<30))!=0;if(W<256)GApp->Keys[W]=true;
            if(W==VK_ESCAPE){if(GApp->bCaptured)GApp->Capture(false);else DestroyWindow(H);}
            if(!bWasDown&&W=='F')GApp->bWireframe=!GApp->bWireframe;
            if(!bWasDown&&W=='T'){GApp->bRoundTunnels=!GApp->bRoundTunnels;GApp->bRebuildPending=true;}
            if(!bWasDown&&W=='V'){GApp->bShapeWorld=!GApp->bShapeWorld;GApp->bRebuildPending=true;}
            if(!bWasDown&&W=='G'&&GApp->bRoundTunnels)
            {
                GApp->bRoundTunnelEnhanced=!GApp->bRoundTunnelEnhanced;
                GApp->bRebuildPending=true;
                std::printf("Round-tunnel comparison: %s wall field.\n",
                    GApp->bRoundTunnelEnhanced?"detailed/smoothed":"legacy smooth tube");
            }
            if(!bWasDown&&W=='H')GApp->ToggleFarRenderer();
            if(!bWasDown&&(W=='N'||W=='P'))GApp->RequestSeed(W=='N'?1:-1);
            if(!bWasDown&&W>='1'&&W<='8'){const int32 I=static_cast<int32>(W-'1');GApp->bShowLevel[I]=!GApp->bShowLevel[I];}
            if(!bWasDown&&W=='9')GApp->bShowLevel[8]=!GApp->bShowLevel[8];
            if(!bWasDown&&W=='0')for(bool& B:GApp->bShowLevel)B=!B;
            return 0;
        }
        case WM_KEYUP:if(W<256)GApp->Keys[W]=false;return 0;
        case WM_RBUTTONDOWN:GApp->Capture(true);return 0;
        case WM_RBUTTONUP:GApp->Capture(false);return 0;
        case WM_MOUSEMOVE:
            if(GApp->bCaptured){POINT C{GApp->Width/2,GApp->Height/2};ClientToScreen(H,&C);POINT P{GET_X_LPARAM(L),GET_Y_LPARAM(L)},Screen=P;ClientToScreen(H,&Screen);const int32 DX=Screen.x-C.x,DY=Screen.y-C.y;if(DX||DY){GApp->Yaw+=DX*0.0035f;GApp->Pitch=std::clamp(GApp->Pitch-DY*0.0035f,-1.55f,1.55f);SetCursorPos(C.x,C.y);}}return 0;
        case WM_MOUSEWHEEL:GApp->MoveSpeed=std::clamp(GApp->MoveSpeed*(GET_WHEEL_DELTA_WPARAM(W)>0?1.25f:0.8f),0.25f,1000.0f);return 0;
        case WM_DESTROY:if(GApp->Streamer)GApp->Streamer->Stop();PostQuitMessage(0);return 0;
        }
        return DefWindowProc(H,Message,W,L);
    }

    std::filesystem::path DefaultDataPath()
    {
        wchar_t Module[MAX_PATH]{};GetModuleFileNameW(nullptr,Module,MAX_PATH);
        const std::filesystem::path Exe(Module);
        return Exe.parent_path().parent_path()/L"data"/L"owner_world.json";
    }

    int RunTileBenchmark(FStandaloneOwnerWorld& World,V3 CameraPosition,int32 Level,
                         int32 Count,bool bProfile,bool bCompareEvaluators,bool bDensityHash,
                         VoxelForgeCore::World* CoreWorld)
    {
        const FVector CameraVoxels(CameraPosition.X*4.0,CameraPosition.Y*4.0,CameraPosition.Z*4.0);
        const FStandaloneDesiredTileSet Desired=StandaloneWorldMesher::BuildDesiredSet(World,CameraVoxels);
        std::vector<FVoxelTileKey> Tiles;Tiles.reserve(static_cast<size_t>(std::max(Count,1)));
        if(CoreWorld)
        {
            const auto CoreTiles=CoreWorld->desiredTiles({CameraVoxels.X,CameraVoxels.Y,CameraVoxels.Z});
            for(const auto& Key:CoreTiles)
                if(Key.lod==Level&&static_cast<int32>(Tiles.size())<Count)
                    Tiles.emplace_back(FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
        }
        else
        {
            for(const FVoxelTileKey& Tile:Desired.Tiles)
                if(Tile.Level==Level&&static_cast<int32>(Tiles.size())<Count)Tiles.push_back(Tile);
        }
        if(Tiles.empty())
        {std::fprintf(stderr,"No LOD %d tiles are in the current desired set.\n",Level);return 4;}

        VoxelGenLOD::InitializeConsoleSwitches();
        UVoxelMarchingCubesMesher::InitializeConsoleSwitches();
        int32 TileCacheWindow=1,TileCacheWindowMinLOD=3,RoomGraphCacheSlots=1;
        FParse::Value(FCommandLine::Get(),"voxel.TileCacheWindow=",TileCacheWindow);
        FParse::Value(FCommandLine::Get(),"voxel.TileCacheWindowMinLOD=",TileCacheWindowMinLOD);
        FParse::Value(FCommandLine::Get(),"voxel.RoomGraphCacheSlots=",RoomGraphCacheSlots);
        std::printf("[BenchmarkConfig] tile_cache_window=%d fused_min_lod=%d room_graph_cache_slots=%d generator=%s command='%s'\n",
            TileCacheWindow,TileCacheWindowMinLOD,RoomGraphCacheSlots,
            CoreWorld&&!bCompareEvaluators?"VoxelForgeCore":"compatibility-diagnostic",
            FCommandLine::Get());
        VoxelDensityProfile::SetMode(bProfile?VoxelDensityProfile::EMode::Attribution
            :VoxelDensityProfile::EMode::Disabled,128);
        if(bProfile)VoxelDensityProfile::CalibrateTimer();
        VoxelDensityProfile::Reset();
        const VoxelDensityProfile::FSnapshot Before=VoxelDensityProfile::SnapshotCurrentThread();

        UVoxelMarchingCubesMesher Mesher;
        Mesher.SetGenerator(World.Generator.get());Mesher.bGenerateSkirts=World.bGenerateSkirts;
        Mesher.SkirtCells=static_cast<float>(World.SkirtCells);Mesher.LODOctaveDrop=World.LODOctaveDrop;
        const int32 BandMin=World.StrateContentCutMinLevel<9&&Desired.bHasStrate
            ?Desired.StrateBottomChunkZ*CHUNK_SIZE:MIN_int32;
        const int32 BandMax=World.StrateContentCutMinLevel<9&&Desired.bHasStrate
            ?(Desired.StrateTopChunkZ+1)*CHUNK_SIZE:MAX_int32;
        if(bCompareEvaluators)
        {
            const bool bOriginalFused=VoxelGenLOD::IsUseFusedEvaluatorEnabled();
            uint64 Mismatches=0,ComparedSamples=0,MeshCountMismatches=0;
            float MaxAbsoluteDifference=0.0f;
            size_t FirstMismatch=SIZE_MAX;
            FVoxelTileKey FirstMismatchTile{};
            VoxelGenLOD::SetUseFusedEvaluatorEnabled(true);
            for(const FVoxelTileKey& Tile:Tiles)
            {
                const int32 Cells=Level<World.FullResClipLevels?CHUNK_SIZE:FMath::Max(1,World.CoarseTileCells);
                const int32 Step=FMath::Max(1,Tile.ExtentVoxels()/Cells);
                std::vector<float> FusedDensities,OpStackDensities;
                int64 FusedSampleCalls=0,OpStackSampleCalls=0;
                FVoxelMeshData FusedMesh=Mesher.GenerateMesh(Tile.OriginVoxels(),Step,Cells,nullptr,
                    BandMin,BandMax,&FusedSampleCalls,nullptr,nullptr,&FusedDensities);
                VoxelGenLOD::SetUseFusedEvaluatorEnabled(false);
                FVoxelMeshData OpStackMesh=Mesher.GenerateMesh(Tile.OriginVoxels(),Step,Cells,nullptr,
                    BandMin,BandMax,&OpStackSampleCalls,nullptr,nullptr,&OpStackDensities);
                VoxelGenLOD::SetUseFusedEvaluatorEnabled(true);
                if(FusedMesh.Vertices.Num()!=OpStackMesh.Vertices.Num()
                    || FusedMesh.Triangles.Num()!=OpStackMesh.Triangles.Num())++MeshCountMismatches;
                const size_t SharedSamples=std::min(FusedDensities.size(),OpStackDensities.size());
                ComparedSamples+=SharedSamples;
                for(size_t I=0;I<SharedSamples;++I)
                {
                    if(std::bit_cast<uint32>(FusedDensities[I])!=std::bit_cast<uint32>(OpStackDensities[I]))
                    {
                        if(FirstMismatch==SIZE_MAX){FirstMismatch=I;FirstMismatchTile=Tile;}
                        ++Mismatches;
                        MaxAbsoluteDifference=std::max(MaxAbsoluteDifference,
                            std::abs(FusedDensities[I]-OpStackDensities[I]));
                    }
                }
                Mismatches+=static_cast<uint64>(std::max(FusedDensities.size(),OpStackDensities.size())-SharedSamples);
            }
            std::printf("[EvaluatorCompare] LOD=%d tiles=%zu compared_samples=%llu density_mismatches=%llu first_tile=(%d,%d,%d) first_index=%llu max_abs=%.9g mesh_count_mismatch_tiles=%llu original_fused=%d\n",
                Level,Tiles.size(),static_cast<unsigned long long>(ComparedSamples),
                static_cast<unsigned long long>(Mismatches),
                FirstMismatchTile.Coord.X,FirstMismatchTile.Coord.Y,FirstMismatchTile.Coord.Z,
                static_cast<unsigned long long>(FirstMismatch),MaxAbsoluteDifference,
                static_cast<unsigned long long>(MeshCountMismatches),bOriginalFused?1:0);
            VoxelGenLOD::SetUseFusedEvaluatorEnabled(bOriginalFused);
            std::fflush(stdout);
            return Mismatches==0?0:5;
        }
        std::vector<double> TileSeconds;TileSeconds.reserve(Tiles.size());
        uint64 Samples=0,Vertices=0,Indices=0;
        uint64 DensityHash=14695981039346656037ull;
        const auto TotalStart=std::chrono::steady_clock::now();
        for(const FVoxelTileKey& Tile:Tiles)
        {
            const auto TileStart=std::chrono::steady_clock::now();
            const int32 Cells=Level<World.FullResClipLevels?CHUNK_SIZE:FMath::Max(1,World.CoarseTileCells);
            const int32 Step=FMath::Max(1,Tile.ExtentVoxels()/Cells);
            int64 TileSamples=0;
            std::vector<float> DebugDensitySamples;
            uint64 TileVertices=0,TileIndices=0;
            if(CoreWorld&&!bCompareEvaluators)
            {
                VoxelForgeCore::TileGenerationOptions Options;
                Options.bandMinZVoxel=BandMin;Options.bandMaxZVoxel=BandMax;
                Options.captureDensitySamples=bDensityHash;
                std::string CoreError;
                auto Mesh=CoreWorld->generateTile(
                    {{Tile.Coord.X,Tile.Coord.Y,Tile.Coord.Z},Tile.Level},Options,CoreError);
                if(!Mesh)
                {
                    std::fprintf(stderr,"[VoxelForgeCore] benchmark tile generation failed: %s\n",CoreError.c_str());
                    return 8;
                }
                TileSamples=static_cast<int64>(Mesh->densitySampleCount);
                TileVertices=Mesh->vertices.size();TileIndices=Mesh->indices.size();
                if(bDensityHash)DebugDensitySamples=std::move(Mesh->debugDensitySamples);
            }
            else
            {
                FVoxelMeshData Mesh=Mesher.GenerateMesh(Tile.OriginVoxels(),Step,Cells,nullptr,
                    BandMin,BandMax,&TileSamples,nullptr,nullptr,
                    bDensityHash?&DebugDensitySamples:nullptr);
                TileVertices=static_cast<uint64>(Mesh.Vertices.Num());
                TileIndices=static_cast<uint64>(Mesh.Triangles.Num());
            }
            Samples+=static_cast<uint64>(FMath::Max<int64>(0,TileSamples));
            Vertices+=TileVertices;
            Indices+=TileIndices;
            if(bDensityHash)
            {
                for(const float Density:DebugDensitySamples)
                {
                    DensityHash^=std::bit_cast<uint32>(Density);
                    DensityHash*=1099511628211ull;
                }
            }
            TileSeconds.push_back(std::chrono::duration<double>(
                std::chrono::steady_clock::now()-TileStart).count());
        }
        const double TotalSeconds=std::chrono::duration<double>(
            std::chrono::steady_clock::now()-TotalStart).count();
        const VoxelDensityProfile::FSnapshot After=VoxelDensityProfile::SnapshotCurrentThread();
        std::sort(TileSeconds.begin(),TileSeconds.end());
        const double Average=TotalSeconds/static_cast<double>(Tiles.size());
        std::printf("[TileBenchmark] LOD=%d generated=%zu samples=%.2fM worker_sum=%.3fs avg=%.3fms p50=%.3fms p95=%.3fms vertices=%llu indices=%llu profiling=%s\n",
            Level,Tiles.size(),static_cast<double>(Samples)/1.0e6,TotalSeconds,Average*1000.0,
            TileSeconds[(TileSeconds.size()-1)*50/100]*1000.0,
            TileSeconds[(TileSeconds.size()-1)*95/100]*1000.0,
            static_cast<unsigned long long>(Vertices),static_cast<unsigned long long>(Indices),
            bProfile?"Attribution":"off");
        if(bDensityHash)
        {
            std::printf("[TileDensityHash] LOD=%d tiles=%zu hash=0x%016llx\n",Level,Tiles.size(),
                static_cast<unsigned long long>(DensityHash));
        }
        if(bProfile)
        {
            std::printf("[TileProfile] LOD=%d",Level);
            for(int32 I=0;I<VoxelDensityProfile::BucketCount;++I)
            {
                const uint64 CycleDelta=After.Cycles[I]>=Before.Cycles[I]
                    ?After.Cycles[I]-Before.Cycles[I]:0;
                const double Ms=static_cast<double>(CycleDelta)
                    *FPlatformTime::GetSecondsPerCycle64()*1000.0;
                if(Ms>0.001)std::printf(" %s=%.3fms",VoxelDensityProfile::BucketName(
                    static_cast<VoxelDensityProfile::EBucket>(I)),Ms);
            }
            std::printf("\n");
            const auto CounterDelta=[&](VoxelDensityProfile::ECounter Counter)
            {
                const size_t I=static_cast<size_t>(Counter);
                return After.Counters[I]>=Before.Counters[I]
                    ?After.Counters[I]-Before.Counters[I]:0;
            };
            std::printf("[TileCounters] sdf_builds=%llu graph_tile_builds=%llu graph_core_builds=%llu op_block_builds=%llu\n",
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfCacheBuild)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::RoomGraphBuildGeneratorTile)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::RoomGraphBuildGeneratorTunnelCore)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::OpBlockBuilds)));
            std::printf("[SdfRebuildCauses] strate=%llu seed=%llu fingerprint=%llu layout=%llu manager=%llu terrain_ops=%llu fused_policy=%llu tile_policy=%llu tile_identity=%llu bounds_x=%llu bounds_y=%llu tile_window_accepted=%llu tile_window_rejected=%llu\n",
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildStrate)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildSeed)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildFingerprint)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildLayout)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildManagerLifetime)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildTerrainOps)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildFusedPolicy)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildTilePolicy)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildTileIdentity)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildBoundsX)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfRebuildBoundsY)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfTileWindowAccepted)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::SdfTileWindowRejected)));
            std::printf("[TilePolicyKeyChanges] origin=%llu step=%llu cells=%llu fingerprint=%llu layout=%llu manager=%llu terrain_ops=%llu fused_path=%llu context=%llu\n",
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyOrigin)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyStep)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyCells)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyFingerprint)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyLayout)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyManager)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyTerrainOps)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyFusedPath)),
                static_cast<unsigned long long>(CounterDelta(VoxelDensityProfile::ECounter::TilePolicyKeyContext)));
        }
        std::fflush(stdout);
        return 0;
    }

    struct FComparePoint
    {
        double X=0.0,Y=0.0,Z=0.0;
    };

    struct FCompareBucketHash
    {
        size_t operator()(const FIntVector& K)const noexcept
        {
            uint64 H=static_cast<uint32>(K.X)*0x9E3779B185EBCA87ull;
            H^=static_cast<uint64>(static_cast<uint32>(K.Y))*0xC2B2AE3D27D4EB4Full;
            H^=static_cast<uint64>(static_cast<uint32>(K.Z))*0x165667B19E3779F9ull;
            return static_cast<size_t>(H^(H>>32));
        }
    };

    struct FDirectedSurfaceDistance
    {
        double Sum=0.0,Maximum=0.0;
        uint64 Count=0,Unmatched=0;
    };

    constexpr int32 OwnerBandGrid=64;
    constexpr double OwnerBandSpacing=0.75;
    constexpr std::array<double,4> OwnerBandMinWavelength={2.0,4.0,8.0,16.0};
    constexpr std::array<double,4> OwnerBandMaxWavelength={4.0,8.0,16.0,32.0};

    void FFT64(std::array<std::complex<double>,OwnerBandGrid>& Values)
    {
        for(int32 I=1,J=0;I<OwnerBandGrid;++I)
        {
            int32 Bit=OwnerBandGrid>>1;
            for(;J&Bit;Bit>>=1)J^=Bit;
            J^=Bit;
            if(I<J)std::swap(Values[I],Values[J]);
        }
        for(int32 Length=2;Length<=OwnerBandGrid;Length<<=1)
        {
            const double Angle=-6.2831853071795864769/static_cast<double>(Length);
            const std::complex<double> Step(std::cos(Angle),std::sin(Angle));
            for(int32 Base=0;Base<OwnerBandGrid;Base+=Length)
            {
                std::complex<double> W(1.0,0.0);
                for(int32 I=0;I<Length/2;++I)
                {
                    const std::complex<double> Even=Values[Base+I];
                    const std::complex<double> Odd=Values[Base+I+Length/2]*W;
                    Values[Base+I]=Even+Odd;Values[Base+I+Length/2]=Even-Odd;W*=Step;
                }
            }
        }
    }

    bool OwnerSurfaceBandRms(const std::array<double,OwnerBandGrid*OwnerBandGrid>& Heights,
        const std::array<uint8,OwnerBandGrid*OwnerBandGrid>& Valid,std::array<double,4>& OutRms)
    {
        size_t ValidCount=0;
        double Sums[3][4]{};
        for(int32 Y=0;Y<OwnerBandGrid;++Y)for(int32 X=0;X<OwnerBandGrid;++X)
        {
            const size_t Index=static_cast<size_t>(Y)*OwnerBandGrid+X;
            if(!Valid[Index])continue;
            ++ValidCount;
            const double U=(X-(OwnerBandGrid-1)*0.5)*OwnerBandSpacing;
            const double V=(Y-(OwnerBandGrid-1)*0.5)*OwnerBandSpacing;
            const double Row[3]={1.0,U,V};
            for(int32 I=0;I<3;++I)
            {
                for(int32 J=0;J<3;++J)Sums[I][J]+=Row[I]*Row[J];
                Sums[I][3]+=Row[I]*Heights[Index];
            }
        }
        if(ValidCount<static_cast<size_t>(OwnerBandGrid*OwnerBandGrid*9/10))return false;
        for(int32 Column=0;Column<3;++Column)
        {
            int32 Pivot=Column;
            for(int32 Row=Column+1;Row<3;++Row)
                if(std::abs(Sums[Row][Column])>std::abs(Sums[Pivot][Column]))Pivot=Row;
            if(std::abs(Sums[Pivot][Column])<1.0e-10)return false;
            if(Pivot!=Column)for(int32 J=Column;J<4;++J)std::swap(Sums[Pivot][J],Sums[Column][J]);
            const double Scale=Sums[Column][Column];
            for(int32 J=Column;J<4;++J)Sums[Column][J]/=Scale;
            for(int32 Row=0;Row<3;++Row)if(Row!=Column)
            {
                const double Factor=Sums[Row][Column];
                for(int32 J=Column;J<4;++J)Sums[Row][J]-=Factor*Sums[Column][J];
            }
        }
        const double Plane[3]={Sums[0][3],Sums[1][3],Sums[2][3]};
        std::array<std::array<std::complex<double>,OwnerBandGrid>,OwnerBandGrid> Spectrum{};
        double WindowSquareSum=0.0;
        for(int32 Y=0;Y<OwnerBandGrid;++Y)
        {
            const double WY=0.5-0.5*std::cos(6.2831853071795864769*Y/(OwnerBandGrid-1));
            for(int32 X=0;X<OwnerBandGrid;++X)
            {
                const size_t Index=static_cast<size_t>(Y)*OwnerBandGrid+X;
                const double WX=0.5-0.5*std::cos(6.2831853071795864769*X/(OwnerBandGrid-1));
                const double U=(X-(OwnerBandGrid-1)*0.5)*OwnerBandSpacing;
                const double V=(Y-(OwnerBandGrid-1)*0.5)*OwnerBandSpacing;
                const double Residual=Valid[Index]?Heights[Index]-(Plane[0]+Plane[1]*U+Plane[2]*V):0.0;
                Spectrum[Y][X]=Residual*(WX*WY);
                WindowSquareSum+=WX*WX*WY*WY;
            }
            FFT64(Spectrum[Y]);
        }
        for(int32 X=0;X<OwnerBandGrid;++X)
        {
            std::array<std::complex<double>,OwnerBandGrid> Column{};
            for(int32 Y=0;Y<OwnerBandGrid;++Y)Column[Y]=Spectrum[Y][X];
            FFT64(Column);
            for(int32 Y=0;Y<OwnerBandGrid;++Y)Spectrum[Y][X]=Column[Y];
        }
        std::array<double,4> Power{};
        for(int32 Y=0;Y<OwnerBandGrid;++Y)for(int32 X=0;X<OwnerBandGrid;++X)
        {
            const double FX=std::min(X,OwnerBandGrid-X)
                /(static_cast<double>(OwnerBandGrid)*OwnerBandSpacing);
            const double FY=std::min(Y,OwnerBandGrid-Y)
                /(static_cast<double>(OwnerBandGrid)*OwnerBandSpacing);
            const double Frequency=std::sqrt(FX*FX+FY*FY);
            if(Frequency<=0.0)continue;
            const double Wavelength=1.0/Frequency;
            for(size_t Band=0;Band<Power.size();++Band)
                if(Wavelength>=OwnerBandMinWavelength[Band]&&Wavelength<OwnerBandMaxWavelength[Band])
                    Power[Band]+=std::norm(Spectrum[Y][X]);
        }
        const double ParsevalScale=static_cast<double>(OwnerBandGrid*OwnerBandGrid)*WindowSquareSum;
        for(size_t Band=0;Band<OutRms.size();++Band)
            OutRms[Band]=std::sqrt(std::max(0.0,Power[Band]/ParsevalScale));
        return true;
    }

    template<class TDensity>
    bool SampleOwnerSurfaceHeightPatch(V3 Origin,V3 Normal,V3 TangentU,V3 TangentV,
        TDensity&& Density,std::array<double,OwnerBandGrid*OwnerBandGrid>& OutHeights,
        std::array<uint8,OwnerBandGrid*OwnerBandGrid>& OutValid)
    {
        size_t ValidCount=0;
        for(int32 Y=0;Y<OwnerBandGrid;++Y)for(int32 X=0;X<OwnerBandGrid;++X)
        {
            const double DU=(X-(OwnerBandGrid-1)*0.5)*OwnerBandSpacing;
            const double DV=(Y-(OwnerBandGrid-1)*0.5)*OwnerBandSpacing;
            const V3 Base=Add(Origin,Add(Mul(TangentU,static_cast<float>(DU)),Mul(TangentV,static_cast<float>(DV))));
            double T=0.0;bool bConverged=false;
            for(int32 Iteration=0;Iteration<7;++Iteration)
            {
                const V3 P=Add(Base,Mul(Normal,static_cast<float>(T)));
                const double D=Density(P);
                if(!std::isfinite(D)||std::abs(T)>16.0)break;
                // Core density keeps a 0.05-voxel near-zero sign margin. Accepting a
                // tighter root tolerance makes every otherwise-good Core patch fail.
                if(std::abs(D)<0.06){bConverged=true;break;}
                const double DP=Density(Add(Base,Mul(Normal,static_cast<float>(T+0.5))));
                const double DM=Density(Add(Base,Mul(Normal,static_cast<float>(T-0.5))));
                const double Gradient=DP-DM;
                if(!std::isfinite(Gradient)||std::abs(Gradient)<0.25)break;
                T-=std::clamp(D/Gradient,-4.0,4.0);
            }
            const size_t Index=static_cast<size_t>(Y)*OwnerBandGrid+X;
            OutHeights[Index]=T;OutValid[Index]=bConverged?1:0;
            ValidCount+=bConverged?1u:0u;
        }
        return ValidCount>=static_cast<size_t>(OwnerBandGrid*OwnerBandGrid*9/10);
    }

    FDirectedSurfaceDistance DirectedSurfaceDistance(
        const std::vector<FComparePoint>& From,const std::vector<FComparePoint>& To,
        double BucketWidth=1.0,int32 SearchCells=3)
    {
        FDirectedSurfaceDistance Result;
        if(From.empty()||To.empty()){Result.Unmatched=From.size();return Result;}
        std::unordered_map<FIntVector,std::vector<const FComparePoint*>,FCompareBucketHash> Buckets;
        Buckets.reserve(To.size());
        const auto BucketOf=[=](const FComparePoint& P)
        {
            return FIntVector(static_cast<int32>(std::floor(P.X/BucketWidth)),
                static_cast<int32>(std::floor(P.Y/BucketWidth)),
                static_cast<int32>(std::floor(P.Z/BucketWidth)));
        };
        for(const FComparePoint& P:To)Buckets[BucketOf(P)].push_back(&P);
        for(const FComparePoint& P:From)
        {
            const FIntVector Center=BucketOf(P);
            double BestSquared=std::numeric_limits<double>::infinity();
            for(int32 Z=-SearchCells;Z<=SearchCells;++Z)
            for(int32 Y=-SearchCells;Y<=SearchCells;++Y)
            for(int32 X=-SearchCells;X<=SearchCells;++X)
            {
                const auto It=Buckets.find(Center+FIntVector(X,Y,Z));
                if(It==Buckets.end())continue;
                for(const FComparePoint* Candidate:It->second)
                {
                    const double DX=P.X-Candidate->X,DY=P.Y-Candidate->Y,DZ=P.Z-Candidate->Z;
                    BestSquared=std::min(BestSquared,DX*DX+DY*DY+DZ*DZ);
                }
            }
            if(std::isfinite(BestSquared))
            {
                const double Distance=std::sqrt(BestSquared);
                Result.Sum+=Distance;Result.Maximum=std::max(Result.Maximum,Distance);++Result.Count;
            }
            else ++Result.Unmatched;
        }
        return Result;
    }

    int RunOwnerShapeComparison(FStandaloneOwnerWorld& World,VoxelForgeCore::World& Shape,
                                V3 CameraPosition,float CameraYaw,float CameraPitch,
                                int32 Workers,int32 LevelFilter,int32 TileLimit,
                                std::string_view AreaLabel,bool bCheckOwnerReliefBands)
    {
        if(!Shape.isCompiled())
        {std::fprintf(stderr,"Shape comparison requires a compiled core world.\n");return 11;}
        if(LevelFilter>=World.FullResClipLevels&&!Shape.hasFarWorld())Shape.buildFarWorld();
        const FVector CameraVoxels(CameraPosition.X*4.0,CameraPosition.Y*4.0,CameraPosition.Z*4.0);
        Shape.auditViewSurface({CameraVoxels.X,CameraVoxels.Y,CameraVoxels.Z},CameraYaw,CameraPitch);
        {
            const int32 Offsets[][3]={{0,0,0},{4,0,0},{-4,0,0},{0,4,0},{0,-4,0},{0,0,4},{0,0,-4}};
            for(const auto& Offset:Offsets)
            {
                const float X=static_cast<float>(CameraVoxels.X+Offset[0]);
                const float Y=static_cast<float>(CameraVoxels.Y+Offset[1]);
                const float Z=static_cast<float>(CameraVoxels.Z+Offset[2]);
                float OwnerDensity=0.0f;
                OwnerDensity=World.Generator->GetDensityAt(X,Y,Z);
                const float ShapeDensity=Shape.densityAt({X,Y,Z});
                std::printf("[ShapeProbe] voxels=(%.1f,%.1f,%.1f) owner=%.5f shape=%.5f owner_air=%d shape_air=%d protected_air=%d solid_overlay=%d\n",
                    X,Y,Z,OwnerDensity,ShapeDensity,OwnerDensity>=0.0f?1:0,ShapeDensity>=0.0f?1:0,
                    Shape.debugCriticalAirAt({X,Y,Z})?1:0,Shape.debugSolidOverlayAt({X,Y,Z})?1:0);
            }
        }
        std::vector<FVoxelTileKey> CoreDesired;
        for(const auto& Key:Shape.desiredTiles({CameraVoxels.X,CameraVoxels.Y,CameraVoxels.Z}))
            CoreDesired.emplace_back(FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
        std::vector<FVoxelTileKey> Tiles;
        for(const FVoxelTileKey& Tile:CoreDesired)
            if(LevelFilter<0?Tile.Level<World.FullResClipLevels:Tile.Level==LevelFilter)
                Tiles.push_back(Tile);
        if(TileLimit>0&&Tiles.size()>static_cast<size_t>(TileLimit))
        {
            constexpr double DegreesToRadians=3.14159265358979323846/180.0;
            const double YawRadians=CameraYaw*DegreesToRadians,PitchRadians=CameraPitch*DegreesToRadians;
            const V3 ViewForward{static_cast<float>(std::sin(YawRadians)*std::cos(PitchRadians)),
                static_cast<float>(std::cos(YawRadians)*std::cos(PitchRadians)),
                static_cast<float>(std::sin(PitchRadians))};
            const auto ViewScore=[&](const FVoxelTileKey& Tile)
            {
                const FIntVector O=Tile.OriginVoxels();
                const V3 D{static_cast<float>(O.X+Tile.ExtentVoxels()*0.5-CameraVoxels.X),
                    static_cast<float>(O.Y+Tile.ExtentVoxels()*0.5-CameraVoxels.Y),
                    static_cast<float>(O.Z+Tile.ExtentVoxels()*0.5-CameraVoxels.Z)};
                const double Distance=std::sqrt(Dot(D,D));
                // Keep near tiles first while preferring the half-space in front of this view.
                return Distance*Distance+32.0*std::max(0.0,Distance-Dot(D,ViewForward));
            };
            std::sort(Tiles.begin(),Tiles.end(),[&](const FVoxelTileKey& A,const FVoxelTileKey& B)
            {
                const double AS=ViewScore(A),BS=ViewScore(B);
                if(AS!=BS)return AS<BS;
                if(A.Coord.X!=B.Coord.X)return A.Coord.X<B.Coord.X;
                if(A.Coord.Y!=B.Coord.Y)return A.Coord.Y<B.Coord.Y;
                return A.Coord.Z<B.Coord.Z;
            });
            Tiles.resize(static_cast<size_t>(TileLimit));
        }
        if(Tiles.empty()){std::fprintf(stderr,"No tiles are available for the requested shape comparison level.\n");return 12;}

        struct FTileComparison
        {
            uint64 Samples=0,Changed=0,MeshMismatch=0;
            uint64 OldHash=14695981039346656037ull,NewHash=14695981039346656037ull;
            std::vector<FComparePoint> OldSurface,NewSurface;
            std::array<double,4> OwnerBandRms{},ShapeBandRms{};
            double SurfacePairDistanceVoxels=0.0;
            bool bHasOwnerBandPatch=false;
        };
        std::vector<FTileComparison> Results(Tiles.size());
        int32 MinChunkZ=MIN_int32,MaxChunkZ=MAX_int32,TopChunkZ=0,BottomChunkZ=0;
        if(World.Settings->bClampViewToStrate&&World.Manager->GetStrateChunkZBounds(
            StandaloneWorldMesher::CameraChunk(CameraVoxels).Z,TopChunkZ,BottomChunkZ))
        {MinChunkZ=BottomChunkZ*CHUNK_SIZE;MaxChunkZ=(TopChunkZ+1)*CHUNK_SIZE;}

        std::atomic<size_t> Next{0};
        std::vector<std::thread> Threads;
        const int32 ThreadCount=std::clamp(Workers,1,8);
        Threads.reserve(ThreadCount);
        for(int32 Worker=0;Worker<ThreadCount;++Worker)
        {
            Threads.emplace_back([&]()
            {
                UVoxelMarchingCubesMesher Mesher;
                Mesher.SetGenerator(World.Generator.get());
                Mesher.bGenerateSkirts=World.bGenerateSkirts;
                Mesher.SkirtCells=static_cast<float>(World.SkirtCells);
                Mesher.LODOctaveDrop=World.LODOctaveDrop;
                for(;;)
                {
                    const size_t Index=Next.fetch_add(1,std::memory_order_relaxed);
                    if(Index>=Tiles.size())break;
                    const FVoxelTileKey& Tile=Tiles[Index];
                    const int32 Cells=Tile.Level<World.FullResClipLevels?CHUNK_SIZE:World.CoarseTileCells;
                    const int32 Step=std::max(1,Tile.ExtentVoxels()/Cells);
                    VoxelForgeCore::TileGenerationOptions ShapeOptions;
                    ShapeOptions.bandMinZVoxel=MinChunkZ;ShapeOptions.bandMaxZVoxel=MaxChunkZ;
                    ShapeOptions.captureDensitySamples=true;
                    std::string CoreError;
                    auto NewMesh=Shape.generateTile({{Tile.Coord.X,Tile.Coord.Y,Tile.Coord.Z},Tile.Level},
                        ShapeOptions,CoreError);
                    std::vector<float> NewDensities=NewMesh?std::move(NewMesh->debugDensitySamples):std::vector<float>{};
                    std::vector<float> OldDensities;
                    FVoxelMeshData OldMesh;
                    OldMesh=Mesher.GenerateMesh(Tile.OriginVoxels(),Step,Cells,nullptr,
                        MinChunkZ,MaxChunkZ,nullptr,nullptr,nullptr,&OldDensities);
                    FTileComparison& R=Results[Index];
                    const size_t Shared=std::min(NewDensities.size(),OldDensities.size());
                    R.Samples=Shared;
                    R.MeshMismatch=std::max(NewDensities.size(),OldDensities.size())-Shared;
                    for(size_t I=0;I<Shared;++I)
                    {
                        if((NewDensities[I]<0.0f)!=(OldDensities[I]<0.0f))++R.Changed;
                        R.NewHash^=std::bit_cast<uint32>(NewDensities[I]);R.NewHash*=1099511628211ull;
                        R.OldHash^=std::bit_cast<uint32>(OldDensities[I]);R.OldHash*=1099511628211ull;
                    }
                    if(Tile.Level==0&&NewMesh&&OldMesh.NumSurfaceVertices>0
                        &&NewMesh->groups.surfaceVertexCount>0)
                    {
                        const int32 OldSurfaceCount=std::min(OldMesh.NumSurfaceVertices,
                            OldMesh.Vertices.Num());
                        const size_t NewSurfaceCount=std::min<size_t>(
                            NewMesh->groups.surfaceVertexCount,NewMesh->vertices.size());
                        const FIntVector TileOrigin=Tile.OriginVoxels();
                        const V3 TileCenter{TileOrigin.X+Tile.ExtentVoxels()*0.5f,
                            TileOrigin.Y+Tile.ExtentVoxels()*0.5f,
                            TileOrigin.Z+Tile.ExtentVoxels()*0.5f};
                        double OwnerCenterDistanceSq=std::numeric_limits<double>::infinity();
                        double ShapeCenterDistanceSq=std::numeric_limits<double>::infinity();
                        V3 OwnerPoint{},OwnerNormal{},ShapePoint{},ShapeNormal{};
                        for(int32 OldIndex=0;OldIndex<OldSurfaceCount;++OldIndex)
                        {
                            if(OldIndex>=OldMesh.Normals.Num())continue;
                            const FVector& OP=OldMesh.Vertices[OldIndex];
                            const FVector& ON=OldMesh.Normals[OldIndex];
                            const V3 CandidatePoint{static_cast<float>(OP.X*0.04),
                                static_cast<float>(OP.Y*0.04),static_cast<float>(OP.Z*0.04)};
                            const V3 CandidateNormal=Normalize({static_cast<float>(ON.X),
                                static_cast<float>(ON.Y),static_cast<float>(ON.Z)});
                            const V3 Delta=Sub(CandidatePoint,TileCenter);
                            const double DistanceSq=Dot(Delta,Delta);
                            if(DistanceSq<OwnerCenterDistanceSq)
                            {
                                OwnerCenterDistanceSq=DistanceSq;OwnerPoint=CandidatePoint;
                                OwnerNormal=CandidateNormal;
                            }
                        }
                        if(std::isfinite(OwnerCenterDistanceSq))
                        {
                            for(size_t NewIndex=0;NewIndex<NewSurfaceCount;++NewIndex)
                            {
                                const auto& NV=NewMesh->vertices[NewIndex];
                                const V3 CandidatePoint{static_cast<float>(NV.position.x*0.04),
                                    static_cast<float>(NV.position.y*0.04),
                                    static_cast<float>(NV.position.z*0.04)};
                                const V3 CandidateNormal=Normalize({static_cast<float>(NV.normal.x),
                                    static_cast<float>(NV.normal.y),static_cast<float>(NV.normal.z)});
                                if(Dot(CandidateNormal,OwnerNormal)<0.5f)continue;
                                const V3 Delta=Sub(CandidatePoint,OwnerPoint);
                                const double DistanceSq=Dot(Delta,Delta);
                                if(DistanceSq<ShapeCenterDistanceSq)
                                {
                                    ShapeCenterDistanceSq=DistanceSq;ShapePoint=CandidatePoint;
                                    ShapeNormal=CandidateNormal;
                                }
                            }
                        }
                        if(std::isfinite(OwnerCenterDistanceSq)&&std::isfinite(ShapeCenterDistanceSq)
                            &&ShapeCenterDistanceSq<=400.0)
                        {
                            const auto TangentBasis=[](V3 Normal,V3& TangentU,V3& TangentV)
                            {
                                TangentU=Cross(Normal,{0.0f,0.0f,1.0f});
                                if(Dot(TangentU,TangentU)<1.0e-6f)
                                    TangentU=Cross(Normal,{1.0f,0.0f,0.0f});
                                TangentU=Normalize(TangentU);
                                TangentV=Normalize(Cross(Normal,TangentU));
                            };
                            const auto FieldNormal=[](V3 Point,V3 Normal,const auto& Density)
                            {
                                if(Dot(Normal,Normal)>=0.25f)return Normalize(Normal);
                                const V3 GX{0.5f,0.0f,0.0f},GY{0.0f,0.5f,0.0f},GZ{0.0f,0.0f,0.5f};
                                return Normalize({Density(Add(Point,GX))-Density(Sub(Point,GX)),
                                    Density(Add(Point,GY))-Density(Sub(Point,GY)),
                                    Density(Add(Point,GZ))-Density(Sub(Point,GZ))});
                            };
                            const auto OwnerDensity=[&](V3 P)
                                {return World.Generator->GetDensityAt(P.X,P.Y,P.Z);};
                            const auto ShapeDensity=[&](V3 P)
                                {return Shape.densityAt({P.X,P.Y,P.Z});};
                            OwnerNormal=FieldNormal(OwnerPoint,OwnerNormal,OwnerDensity);
                            ShapeNormal=FieldNormal(ShapePoint,ShapeNormal,ShapeDensity);
                            V3 OwnerTangentU{},OwnerTangentV{},ShapeTangentU{},ShapeTangentV{};
                            TangentBasis(OwnerNormal,OwnerTangentU,OwnerTangentV);
                            TangentBasis(ShapeNormal,ShapeTangentU,ShapeTangentV);
                            std::array<double,OwnerBandGrid*OwnerBandGrid> OwnerHeights{},ShapeHeights{};
                            std::array<uint8,OwnerBandGrid*OwnerBandGrid> OwnerValid{},ShapeValid{};
                            const bool bOwnerPatch=SampleOwnerSurfaceHeightPatch(OwnerPoint,
                                OwnerNormal,OwnerTangentU,OwnerTangentV,OwnerDensity,
                                OwnerHeights,OwnerValid);
                            const bool bShapePatch=SampleOwnerSurfaceHeightPatch(ShapePoint,
                                ShapeNormal,ShapeTangentU,ShapeTangentV,ShapeDensity,
                                ShapeHeights,ShapeValid);
                            if(bOwnerPatch&&bShapePatch
                                &&OwnerSurfaceBandRms(OwnerHeights,OwnerValid,R.OwnerBandRms)
                                &&OwnerSurfaceBandRms(ShapeHeights,ShapeValid,R.ShapeBandRms))
                            {
                                R.SurfacePairDistanceVoxels=std::sqrt(ShapeCenterDistanceSq);
                                R.bHasOwnerBandPatch=true;
                            }
                        }
                    }
                    constexpr size_t VertexStride=4;
                    if(NewMesh)
                    {
                        R.NewSurface.reserve(NewMesh->vertices.size()/VertexStride+1);
                        for(size_t I=0;I<NewMesh->vertices.size();I+=VertexStride)
                        {const auto& P=NewMesh->vertices[I].position;R.NewSurface.push_back({P.x*0.01,P.y*0.01,P.z*0.01});}
                    }
                    R.OldSurface.reserve(static_cast<size_t>(OldMesh.Vertices.Num())/VertexStride+1);
                    for(int32 I=0;I<OldMesh.Vertices.Num();I+=static_cast<int32>(VertexStride))
                    {const FVector& P=OldMesh.Vertices[I];R.OldSurface.push_back({P.X*0.01,P.Y*0.01,P.Z*0.01});}
                }
            });
        }
        for(std::thread& Thread:Threads)Thread.join();

        uint64 Samples=0,Changed=0,MeshMismatch=0;
        uint64 OwnerBandPatches=0;
        std::array<double,4> OwnerBandEnergy{},ShapeBandEnergy{};
        double SurfacePairDistanceSum=0.0;
        uint64 OldHash=14695981039346656037ull,NewHash=14695981039346656037ull;
        size_t OldVertexCount=0,NewVertexCount=0;
        std::vector<FComparePoint> OldSurface,NewSurface;
        for(const FTileComparison& R:Results)
        {
            Samples+=R.Samples;Changed+=R.Changed;MeshMismatch+=R.MeshMismatch;
            if(R.bHasOwnerBandPatch)
            {
                ++OwnerBandPatches;SurfacePairDistanceSum+=R.SurfacePairDistanceVoxels;
                for(size_t Band=0;Band<4;++Band)
                {
                    OwnerBandEnergy[Band]+=R.OwnerBandRms[Band]*R.OwnerBandRms[Band];
                    ShapeBandEnergy[Band]+=R.ShapeBandRms[Band]*R.ShapeBandRms[Band];
                }
            }
            OldHash^=R.OldHash;OldHash*=1099511628211ull;
            NewHash^=R.NewHash;NewHash*=1099511628211ull;
            OldVertexCount+=R.OldSurface.size();NewVertexCount+=R.NewSurface.size();
        }
        OldSurface.reserve(OldVertexCount);NewSurface.reserve(NewVertexCount);
        for(FTileComparison& R:Results)
        {
            OldSurface.insert(OldSurface.end(),R.OldSurface.begin(),R.OldSurface.end());
            NewSurface.insert(NewSurface.end(),R.NewSurface.begin(),R.NewSurface.end());
        }
        const FDirectedSurfaceDistance OldToNew=DirectedSurfaceDistance(OldSurface,NewSurface,1.0,3);
        const FDirectedSurfaceDistance NewToOld=DirectedSurfaceDistance(NewSurface,OldSurface,1.0,3);
        const uint64 DistanceSamples=OldToNew.Count+NewToOld.Count;
        const double MeanDistance=DistanceSamples
            ?(OldToNew.Sum+NewToOld.Sum)/static_cast<double>(DistanceSamples):0.0;
        std::printf("[OwnerShapeCompare] tiles=%zu fullres_levels=%d level_filter=%d lattice_samples=%llu changed_signs=%llu changed_share=%.6f%% old_hash=0x%016llx shape_hash=0x%016llx mesh_sample_mismatch=%llu old_surface_samples=%zu shape_surface_samples=%zu symmetric_vertex_mean_m=%.4f symmetric_vertex_max_m=%.4f unmatched_surface_samples=%llu vertex_stride=4 workers=%d\n",
            Tiles.size(),World.FullResClipLevels,LevelFilter,static_cast<unsigned long long>(Samples),
            static_cast<unsigned long long>(Changed),Samples?100.0*Changed/Samples:0.0,
            static_cast<unsigned long long>(OldHash),static_cast<unsigned long long>(NewHash),
            static_cast<unsigned long long>(MeshMismatch),OldSurface.size(),NewSurface.size(),
            MeanDistance,std::max(OldToNew.Maximum,NewToOld.Maximum),
            static_cast<unsigned long long>(OldToNew.Unmatched+NewToOld.Unmatched),ThreadCount);
        const std::array<const char*,4> BandNames={"2-4","4-8","8-16","16-32"};
        const std::array<double,4> BandMinimumRatio={0.0,0.75,0.75,0.65};
        bool bReliefBandsPass=OwnerBandPatches>0;
        std::printf("[OwnerReliefBandCompare] area=%.*s paired_surface_patches=%llu mean_surface_pair_distance=%.3f_vox grid=%dx%d spacing=%.2f_vox bands_rms_vox=",
            static_cast<int>(AreaLabel.size()),AreaLabel.data(),
            static_cast<unsigned long long>(OwnerBandPatches),OwnerBandPatches
                ?SurfacePairDistanceSum/static_cast<double>(OwnerBandPatches):0.0,
            OwnerBandGrid,OwnerBandGrid,OwnerBandSpacing);
        for(size_t Band=0;Band<4;++Band)
        {
            const double OwnerRms=OwnerBandPatches
                ?std::sqrt(OwnerBandEnergy[Band]/static_cast<double>(OwnerBandPatches)):0.0;
            const double ShapeRms=OwnerBandPatches
                ?std::sqrt(ShapeBandEnergy[Band]/static_cast<double>(OwnerBandPatches)):0.0;
            const double Ratio=OwnerRms>1.0e-5?ShapeRms/OwnerRms:0.0;
            std::printf("%s(owner=%.4f,shape=%.4f,ratio=%.3f)%s",BandNames[Band],
                OwnerRms,ShapeRms,Ratio,Band<3?" ":"\n");
            if(Band>0&&OwnerRms>=0.025&&Ratio<BandMinimumRatio[Band])bReliefBandsPass=false;
        }
        std::printf("[OwnerReliefBandAudit] minimum_ratio_4-8=0.750 minimum_ratio_8-16=0.750 minimum_ratio_16-32=0.650 evaluated_patches=%llu status=%s%s\n",
            static_cast<unsigned long long>(OwnerBandPatches),bReliefBandsPass?"PASS":"FAIL",
            bCheckOwnerReliefBands?" (gate enabled)":" (measurement only)");
        std::fflush(stdout);
        return Samples>0&&MeshMismatch==0&&(!bCheckOwnerReliefBands||bReliefBandsPass)?0:13;
    }

    int RunRoundWallComparison(FStandaloneOwnerWorld& World,V3 CameraPosition,int32 TileCount)
    {
        const FVector CameraVoxels(CameraPosition.X*4.0,CameraPosition.Y*4.0,CameraPosition.Z*4.0);
        const FStandaloneDesiredTileSet Desired=StandaloneWorldMesher::BuildDesiredSet(World,CameraVoxels);
        std::vector<FVoxelTileKey> Tiles;Tiles.reserve(static_cast<size_t>(TileCount));
        for(const FVoxelTileKey& Tile:Desired.Tiles)
            if(Tile.Level==0&&static_cast<int32>(Tiles.size())<TileCount)Tiles.push_back(Tile);
        if(Tiles.empty()){std::fprintf(stderr,"No LOD0 tiles are available for the round-tunnel comparison.\n");return 9;}
        struct FComparisonTile
        {
            std::vector<float> Densities;
            std::vector<FComparePoint> Surface;
        };
        std::vector<FComparisonTile> Legacy(Tiles.size());
        std::vector<FComparisonTile> Detailed(Tiles.size());
        UVoxelMarchingCubesMesher Mesher;
        Mesher.SetGenerator(World.Generator.get());Mesher.bGenerateSkirts=World.bGenerateSkirts;
        Mesher.SkirtCells=static_cast<float>(World.SkirtCells);Mesher.LODOctaveDrop=World.LODOctaveDrop;
        int32 MinChunkZ=MIN_int32,MaxChunkZ=MAX_int32,TopChunkZ=0,BottomChunkZ=0;
        if(World.Settings->bClampViewToStrate&&World.Manager->GetStrateChunkZBounds(
            StandaloneWorldMesher::CameraChunk(CameraVoxels).Z,TopChunkZ,BottomChunkZ))
        {
            MinChunkZ=BottomChunkZ*CHUNK_SIZE;
            MaxChunkZ=(TopChunkZ+1)*CHUNK_SIZE;
        }
        uint64 TotalSamples=0,ChangedSigns=0;
        uint64 LegacyHash=14695981039346656037ull,DetailedHash=14695981039346656037ull;
        const bool PreviousRoundTunnels=World.Generator->bStandaloneResearchRoundTunnels;
        const bool PreviousEnhancedWalls=World.Generator->bStandaloneResearchEnhancedWalls;
        const auto Evaluate=[&](bool bEnhanced,std::vector<FComparisonTile>& Out)
        {
            World.Generator->bStandaloneResearchRoundTunnels=true;
            World.Generator->bStandaloneResearchEnhancedWalls=bEnhanced;
            for(size_t I=0;I<Tiles.size();++I)
            {
                const FVoxelTileKey& Tile=Tiles[I];
                const int32 Step=1,Cells=CHUNK_SIZE;
                FVoxelMeshData Mesh=Mesher.GenerateMesh(Tile.OriginVoxels(),Step,Cells,nullptr,
                    MinChunkZ,MaxChunkZ,nullptr,nullptr,nullptr,&Out[I].Densities);
                Out[I].Surface.reserve(static_cast<size_t>(Mesh.Vertices.Num()));
                for(const FVector& P:Mesh.Vertices)
                    Out[I].Surface.push_back({P.X*0.01,P.Y*0.01,P.Z*0.01});
            }
        };
        Evaluate(false,Legacy);
        Evaluate(true,Detailed);
        World.Generator->bStandaloneResearchRoundTunnels=PreviousRoundTunnels;
        World.Generator->bStandaloneResearchEnhancedWalls=PreviousEnhancedWalls;
        FDirectedSurfaceDistance LegacyToDetailed,DetailedToLegacy;
        std::vector<FComparePoint> AllLegacySurface,AllDetailedSurface;
        for(size_t TileIndex=0;TileIndex<Tiles.size();++TileIndex)
        {
            const auto& A=Legacy[TileIndex];const auto& B=Detailed[TileIndex];
            const size_t Shared=std::min(A.Densities.size(),B.Densities.size());
            TotalSamples+=Shared;
            for(size_t I=0;I<Shared;++I)
            {
                if((A.Densities[I]<0.0f)!=(B.Densities[I]<0.0f))++ChangedSigns;
                LegacyHash^=std::bit_cast<uint32>(A.Densities[I]);LegacyHash*=1099511628211ull;
                DetailedHash^=std::bit_cast<uint32>(B.Densities[I]);DetailedHash*=1099511628211ull;
            }
            AllLegacySurface.insert(AllLegacySurface.end(),A.Surface.begin(),A.Surface.end());
            AllDetailedSurface.insert(AllDetailedSurface.end(),B.Surface.begin(),B.Surface.end());
        }
        LegacyToDetailed=DirectedSurfaceDistance(AllLegacySurface,AllDetailedSurface);
        DetailedToLegacy=DirectedSurfaceDistance(AllDetailedSurface,AllLegacySurface);
        const uint64 DistanceSamples=LegacyToDetailed.Count+DetailedToLegacy.Count;
        const double MeanDistance=DistanceSamples>0
            ?(LegacyToDetailed.Sum+DetailedToLegacy.Sum)/static_cast<double>(DistanceSamples):0.0;
        const double Hausdorff=std::max(LegacyToDetailed.Maximum,DetailedToLegacy.Maximum);
        std::printf("[RoundTunnelCompare] tiles=%zu lattice_samples=%llu changed_signs=%llu changed_share=%.6f%% legacy_density_hash=0x%016llx detailed_density_hash=0x%016llx old_vertices=%zu new_vertices=%zu symmetric_vertex_mean_m=%.4f symmetric_vertex_max_m=%.4f unmatched_vertices=%llu\n",
            Tiles.size(),static_cast<unsigned long long>(TotalSamples),
            static_cast<unsigned long long>(ChangedSigns),TotalSamples?100.0*ChangedSigns/TotalSamples:0.0,
            static_cast<unsigned long long>(LegacyHash),static_cast<unsigned long long>(DetailedHash),
            AllLegacySurface.size(),AllDetailedSurface.size(),MeanDistance,Hausdorff,
            static_cast<unsigned long long>(LegacyToDetailed.Unmatched+DetailedToLegacy.Unmatched));
        return TotalSamples>0&&LegacyToDetailed.Unmatched+DetailedToLegacy.Unmatched==0?0:10;
    }

    void HashWord(uint64& Hash,uint64 Word,int32 Bytes=8)
    {
        for(int32 I=0;I<Bytes;++I)
        {
            Hash^=static_cast<uint8>(Word>>(I*8));
            Hash*=1099511628211ull;
        }
    }

    struct FHashedTile
    {
        FVoxelTileKey Key;
        uint64 Hash=14695981039346656037ull;
    };

    bool TileKeyLessForHash(const FVoxelTileKey& A,const FVoxelTileKey& B)
    {
        if(A.Level!=B.Level)return A.Level<B.Level;
        if(A.Coord.Z!=B.Coord.Z)return A.Coord.Z<B.Coord.Z;
        if(A.Coord.Y!=B.Coord.Y)return A.Coord.Y<B.Coord.Y;
        return A.Coord.X<B.Coord.X;
    }

    uint64 HashTileRows(std::vector<FHashedTile>& Rows)
    {
        std::sort(Rows.begin(),Rows.end(),[](const FHashedTile& A,const FHashedTile& B)
            {return TileKeyLessForHash(A.Key,B.Key);});
        uint64 Hash=14695981039346656037ull;
        for(const FHashedTile& Row:Rows)
        {
            HashWord(Hash,static_cast<uint32>(Row.Key.Level),4);
            HashWord(Hash,static_cast<uint32>(Row.Key.Coord.X),4);
            HashWord(Hash,static_cast<uint32>(Row.Key.Coord.Y),4);
            HashWord(Hash,static_cast<uint32>(Row.Key.Coord.Z),4);
            HashWord(Hash,Row.Hash);
        }
        return Hash;
    }

    int RunPipelinedStreamingBenchmark(std::shared_ptr<FStandaloneOwnerWorld> OwnerWorld,
        const VoxelForgeCore::ResolvedWorldSnapshot& CoreSnapshot,
        const FVector& CameraPositionVoxels,int32 WorkerCount,int32 MaxLevel,
        VoxelForgeCore::DetailVariant DetailVariant,const VoxelForgeCore::JobScheduler& Scheduler,
        bool bHashDensity,bool bSkipUvAndColors,bool bSkipMaterialGroups,bool bSkipUV,bool bSkipColors,
        std::chrono::steady_clock::time_point ColdStart)
    {
        using Clock=std::chrono::steady_clock;
        FStandaloneOwnerWorld& World=*OwnerWorld;
        const bool bRoundTunnels=World.Generator->bStandaloneResearchRoundTunnels;
        const bool bEnhancedTunnelWalls=World.Generator->bStandaloneResearchEnhancedWalls;
        VoxelForgeCore::World Shape(CoreSnapshot);
        FStandaloneDesiredTileSet Desired=StandaloneWorldMesher::BuildDesiredSet(World,CameraPositionVoxels);
        const auto CoreTiles=Shape.desiredTiles(
            {CameraPositionVoxels.X,CameraPositionVoxels.Y,CameraPositionVoxels.Z});
        Desired.Tiles.clear();Desired.Tiles.reserve(CoreTiles.size());
        for(const auto& Key:CoreTiles)
            Desired.Tiles.emplace_back(FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
        if(MaxLevel>=0)
            Desired.Tiles.erase(std::remove_if(Desired.Tiles.begin(),Desired.Tiles.end(),
                [&](const FVoxelTileKey& Tile){return Tile.Level>MaxLevel;}),Desired.Tiles.end());
        FIntVector FineMin(INT32_MAX,INT32_MAX,INT32_MAX),FineMax(INT32_MIN,INT32_MIN,INT32_MIN);
        FIntVector ClipMin(INT32_MAX,INT32_MAX,INT32_MAX),ClipMax(INT32_MIN,INT32_MIN,INT32_MIN);
        for(const FVoxelTileKey& Tile:Desired.Tiles)
        {
            const FIntVector Origin=Tile.OriginVoxels();const int32 Extent=Tile.ExtentVoxels();
            ClipMin.X=std::min(ClipMin.X,Origin.X);ClipMin.Y=std::min(ClipMin.Y,Origin.Y);ClipMin.Z=std::min(ClipMin.Z,Origin.Z);
            ClipMax.X=std::max(ClipMax.X,Origin.X+Extent);ClipMax.Y=std::max(ClipMax.Y,Origin.Y+Extent);ClipMax.Z=std::max(ClipMax.Z,Origin.Z+Extent);
            if(Tile.Level<World.FullResClipLevels)
            {
                FineMin.X=std::min(FineMin.X,Origin.X);FineMin.Y=std::min(FineMin.Y,Origin.Y);FineMin.Z=std::min(FineMin.Z,Origin.Z);
                FineMax.X=std::max(FineMax.X,Origin.X+Extent);FineMax.Y=std::max(FineMax.Y,Origin.Y+Extent);FineMax.Z=std::max(FineMax.Z,Origin.Z+Extent);
            }
        }
        if(FineMin.X==INT32_MAX||ClipMin.X==INT32_MAX)
        {std::fprintf(stderr,"[StreamPipeline] no desired detail or clipmap tiles.\n");return 7;}

        const FIntVector CenterChunk=Desired.CenterChunk;
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> Neighborhood;
        Neighborhood.reserve(32);
        for(const FVoxelTileKey& Tile:Desired.Tiles)
            if(Tile.Level==0&&std::abs(Tile.Coord.X-CenterChunk.X)<=1
                &&std::abs(Tile.Coord.Y-CenterChunk.Y)<=1&&std::abs(Tile.Coord.Z-CenterChunk.Z)<=1)
                Neighborhood.insert(Tile);
        if(Neighborhood.size()!=27)
        {std::fprintf(stderr,"[StreamPipeline] expected 27 player-neighborhood tiles, found %zu.\n",Neighborhood.size());return 7;}

        // The initial desired set now owns all clipmap metadata needed by this Core-only run.
        // Release the separate compatibility-world generation graph before compiling Core.
        World.Generator.reset();
        World.Manager.reset();
        World.DefinitionStorage.clear();
        World.BiomeStorage.clear();
        World.OperationStorage.clear();
        World.DensitySamples.clear();

        std::string BuildError;
        const bool bIncludeFarWorld=MaxLevel<0||MaxLevel>=World.FullResClipLevels;
        const auto ShapeBuildStart=Clock::now();
        VoxelForgeCore::ShapeBuildRequest ShapeRequest;
        ShapeRequest.minVoxel={FineMin.X,FineMin.Y,FineMin.Z};
        ShapeRequest.maxVoxel={FineMax.X,FineMax.Y,FineMax.Z};
        ShapeRequest.clipMinVoxel={ClipMin.X,ClipMin.Y,ClipMin.Z};
        ShapeRequest.clipMaxVoxel={ClipMax.X,ClipMax.Y,ClipMax.Z};
        ShapeRequest.workerLimit=static_cast<uint32>(std::clamp(WorkerCount,1,8));
        ShapeRequest.detail=DetailVariant;ShapeRequest.includeFarWorld=bIncludeFarWorld;
        ShapeRequest.roundTunnels=bRoundTunnels;
        ShapeRequest.enhancedTunnelWalls=bEnhancedTunnelWalls;
        if(!Shape.compile(ShapeRequest,Scheduler,BuildError))
        {std::fprintf(stderr,"[StreamPipeline] full shape build failed: %s\n",BuildError.c_str());return 2;}
        const auto ShapeReady=Clock::now();
        const double ShapeSeconds=std::chrono::duration<double>(ShapeReady-ShapeBuildStart).count();
        FStandaloneStreamingGenerator Streamer(World,WorkerCount,bHashDensity,nullptr,MaxLevel,
            bSkipUvAndColors,bSkipMaterialGroups,bSkipUV,bSkipColors,&Shape);
        const auto StreamStart=Clock::now();
        const FStandaloneStreamingUpdate Update=Streamer.StartWithDesired(Desired);
        if(Update.DesiredTiles!=static_cast<int32>(Desired.Tiles.size()))
        {
            Streamer.Stop();
            std::fprintf(stderr,"[StreamPipeline] expected %zu tiles but queued %d.\n",
                Desired.Tiles.size(),Update.DesiredTiles);return 7;
        }

        std::vector<FHashedTile> DensityRows,MeshRows,PayloadRows,TopologyRows;
        int32 Finished=0,NeighborhoodFinished=0,FullDetailFinished=0;
        int32 FullDetailExpected=0;
        for(const FVoxelTileKey& Tile:Desired.Tiles)
            if(Tile.Level<World.FullResClipLevels)++FullDetailExpected;
        Clock::time_point NeighborhoodReady{},FullDetailReady{},WholeViewReady{};
        bool bNeighborhoodReported=false,bFullDetailReported=false;
        while(Finished<Update.DesiredTiles)
        {
            std::vector<FStandaloneCpuTile> Ready=Streamer.TakeReady();
            if(Ready.empty()){std::this_thread::sleep_for(std::chrono::milliseconds(1));continue;}
            for(const FStandaloneCpuTile& Tile:Ready)
            {
                ++Finished;
                WholeViewReady=std::max(WholeViewReady,Tile.CompletedAt);
                if(Neighborhood.contains(Tile.Key))
                {
                    ++NeighborhoodFinished;
                    NeighborhoodReady=std::max(NeighborhoodReady,Tile.CompletedAt);
                }
                if(Tile.Key.Level<World.FullResClipLevels)
                {
                    ++FullDetailFinished;
                    FullDetailReady=std::max(FullDetailReady,Tile.CompletedAt);
                }
                if(bHashDensity)
                {
                    if(Tile.bHasDensityHash)DensityRows.push_back({Tile.Key,Tile.DensityHash});
                    if(Tile.bHasMeshHash)MeshRows.push_back({Tile.Key,Tile.MeshHash});
                    if(Tile.bHasPayloadHash)PayloadRows.push_back({Tile.Key,Tile.PayloadHash});
                    if(Tile.bHasTopologyHash)TopologyRows.push_back({Tile.Key,Tile.TopologyHash});
                }
            }
            if(!bNeighborhoodReported&&NeighborhoodFinished==static_cast<int32>(Neighborhood.size()))
            {
                std::printf("[PipelineMilestone] neighborhood_ready=%.3fs generated=%d/27\n",
                    std::chrono::duration<double>(NeighborhoodReady-ColdStart).count(),NeighborhoodFinished);
                bNeighborhoodReported=true;
            }
            if(!bFullDetailReported&&FullDetailFinished==FullDetailExpected)
            {
                std::printf("[PipelineMilestone] full_detail_ready=%.3fs generated=%d/%d\n",
                    std::chrono::duration<double>(FullDetailReady-ColdStart).count(),
                    FullDetailFinished,FullDetailExpected);
                bFullDetailReported=true;
            }
        }
        Streamer.Stop();
        const double NeighborhoodSeconds=std::chrono::duration<double>(NeighborhoodReady-ColdStart).count();
        const double NeighborhoodMeshSeconds=std::chrono::duration<double>(NeighborhoodReady-ShapeReady).count();
        const double FullDetailSeconds=std::chrono::duration<double>(FullDetailReady-ColdStart).count();
        const double WholeViewSeconds=std::chrono::duration<double>(WholeViewReady-ColdStart).count();
        const double StreamSeconds=std::chrono::duration<double>(WholeViewReady-StreamStart).count();
        std::printf("[StreamPipeline] workers=%d shape=%.3fs neighborhood_mesh=%.3fs neighborhood_ready=%.3fs full_detail_ready=%.3fs whole_view_ready=%.3fs neighborhood_tiles=%d/27 full_detail_tiles=%d/%d whole_tiles=%d/%zu stream_wall=%.3fs cache_hits=0 full_payload=1\n",
            WorkerCount,ShapeSeconds,NeighborhoodMeshSeconds,NeighborhoodSeconds,
            FullDetailSeconds,WholeViewSeconds,NeighborhoodFinished,
            FullDetailFinished,FullDetailExpected,Finished,Desired.Tiles.size(),StreamSeconds);

        bool bHashesPassing=true;
        if(bHashDensity)
        {
            bHashesPassing=DensityRows.size()==Desired.Tiles.size()
                &&MeshRows.size()==Desired.Tiles.size()
                &&PayloadRows.size()==Desired.Tiles.size()
                &&TopologyRows.size()==Desired.Tiles.size();
            std::vector<FHashedTile> FullDensity,FullMesh,FullPayload,FullTopology;
            for(const FHashedTile& Row:DensityRows)if(Row.Key.Level<World.FullResClipLevels)FullDensity.push_back(Row);
            for(const FHashedTile& Row:MeshRows)if(Row.Key.Level<World.FullResClipLevels)FullMesh.push_back(Row);
            for(const FHashedTile& Row:PayloadRows)if(Row.Key.Level<World.FullResClipLevels)FullPayload.push_back(Row);
            for(const FHashedTile& Row:TopologyRows)if(Row.Key.Level<World.FullResClipLevels)FullTopology.push_back(Row);
            std::printf("[PipelineFullDetailHash] tiles=%d density=0x%016llx mesh=0x%016llx payload=0x%016llx topology=0x%016llx\n",
                FullDetailExpected,static_cast<unsigned long long>(HashTileRows(FullDensity)),
                static_cast<unsigned long long>(HashTileRows(FullMesh)),
                static_cast<unsigned long long>(HashTileRows(FullPayload)),
                static_cast<unsigned long long>(HashTileRows(FullTopology)));
            std::printf("[PipelineWholeViewHash] tiles=%zu density=0x%016llx mesh=0x%016llx payload=0x%016llx topology=0x%016llx\n",
                Desired.Tiles.size(),static_cast<unsigned long long>(HashTileRows(DensityRows)),
                static_cast<unsigned long long>(HashTileRows(MeshRows)),
                static_cast<unsigned long long>(HashTileRows(PayloadRows)),
                static_cast<unsigned long long>(HashTileRows(TopologyRows)));
        }
        return NeighborhoodFinished==static_cast<int32>(Neighborhood.size())
            &&FullDetailFinished==FullDetailExpected&&Finished==Update.DesiredTiles&&bHashesPassing?0:6;
    }
    int RunStreamingBenchmark(FStandaloneOwnerWorld& World,VoxelForgeCore::World* CoreWorld,
                              const FVector& CameraPositionVoxels,
                              int32 WorkerCount,int32 MaxLevel,bool bRoundTunnels,
                              bool bEnhancedWalls,const FStandaloneMeshCache* Cache,
                              bool bHashDensity,bool bSkipUvAndColors,bool bSkipMaterialGroups,
                              bool bSkipUV,bool bSkipColors,
                              bool bShardAudit,bool bFarSurvivalAudit,bool bFarAuditBadShell,
                              float CameraYaw,float CameraPitch,
                              const std::string& BakeCachePath,
                              const FStandaloneMeshCacheMetadata& CacheMetadata)
    {
        FStandaloneDesiredTileSet Desired=StandaloneWorldMesher::BuildDesiredSet(World,CameraPositionVoxels);
        if(CoreWorld)
        {
            const auto CoreTiles=CoreWorld->desiredTiles(
                {CameraPositionVoxels.X,CameraPositionVoxels.Y,CameraPositionVoxels.Z});
            Desired.Tiles.clear();Desired.Tiles.reserve(CoreTiles.size());
            for(const auto& Key:CoreTiles)
                Desired.Tiles.emplace_back(FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
        }
        if(MaxLevel>=0)
            Desired.Tiles.erase(std::remove_if(Desired.Tiles.begin(),Desired.Tiles.end(),
                [&](const FVoxelTileKey& Tile){return Tile.Level>MaxLevel;}),Desired.Tiles.end());
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> DesiredTileSet;
        DesiredTileSet.reserve(Desired.Tiles.size()*2+1);
        for(const FVoxelTileKey& Tile:Desired.Tiles)DesiredTileSet.insert(Tile);
        if(Desired.Tiles.size()>1000000)
        {
            std::fprintf(stderr,"Stream benchmark rejected an unsafe desired set of %zu tiles.\n",
                Desired.Tiles.size());
            return 7;
        }
        const int32 FullDetailExpected=static_cast<int32>(std::count_if(
            Desired.Tiles.begin(),Desired.Tiles.end(),[&](const FVoxelTileKey& Tile)
                {return Tile.Level<World.FullResClipLevels;}));
        std::unique_ptr<FFarSurvivalAudit> FarAudit;
        if(bFarSurvivalAudit)
        {
            if(!CoreWorld||!CoreWorld->hasFarWorld())
            {std::fprintf(stderr,"Far survival audit requires a compiled core world with far-parent data.\n");return 15;}
            FarAudit=std::make_unique<FFarSurvivalAudit>(World,*CoreWorld,Desired.Tiles,bFarAuditBadShell);
        }
        std::array<std::vector<FHashedTile>,MaxLevels+1> HashRows;
        std::array<std::vector<FHashedTile>,MaxLevels+1> MeshHashRows;
        std::array<std::vector<FHashedTile>,MaxLevels+1> PayloadHashRows;
        std::array<std::vector<FHashedTile>,MaxLevels+1> TopologyHashRows;
        std::array<uint64,MaxLevels+1> SamplesByLevel{};
        std::array<uint64,MaxLevels+1> VerticesByLevel{};
        std::array<uint64,MaxLevels+1> IndicesByLevel{};
        std::array<uint64,MaxLevels+1> TilesByLevel{};
        std::array<double,MaxLevels+1> WorkerSecondsByLevel{};
        std::array<double,MaxLevels+1> FirstCompletionByLevel{};
        std::array<double,MaxLevels+1> LastCompletionByLevel{};
        std::array<uint64,MaxLevels+1> OutputBytesByLevel{};
        std::array<uint64,MaxLevels+1> MeshPayloadAllocatedBytesByLevel{};
        FirstCompletionByLevel.fill(std::numeric_limits<double>::infinity());
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> PlayerNeighborhoodTiles;
        for(const FVoxelTileKey& Tile:Desired.Tiles)
            if(Tile.Level==0&&std::abs(Tile.Coord.X-Desired.CenterChunk.X)<=1
                &&std::abs(Tile.Coord.Y-Desired.CenterChunk.Y)<=1
                &&std::abs(Tile.Coord.Z-Desired.CenterChunk.Z)<=1)
                PlayerNeighborhoodTiles.insert(Tile);
        std::unordered_set<FVoxelTileKey,FStandaloneTileKeyHash> PlayerNeighborhoodReady;
        auto PlayerNeighborhoodCompleted=std::chrono::steady_clock::time_point{};
        std::vector<FStandaloneCpuTile> CacheTiles;
        if(!BakeCachePath.empty())CacheTiles.reserve(Desired.Tiles.size());
        if(CoreWorld&&!FarAudit)
        {
            // Tile keys and band limits are already captured in Desired. Core owns field and
            // mesh generation, so discard the unused compatibility generator before streaming.
            World.Generator.reset();
            World.Manager.reset();
            World.DefinitionStorage.clear();
            World.BiomeStorage.clear();
            World.OperationStorage.clear();
            World.DensitySamples.clear();
        }
        const auto Start=std::chrono::steady_clock::now();
        FStandaloneStreamingGenerator Streamer(World,WorkerCount,
            bHashDensity||!BakeCachePath.empty(),Cache,MaxLevel,
            bSkipUvAndColors,bSkipMaterialGroups,bSkipUV,bSkipColors,CoreWorld);
        const FStandaloneStreamingUpdate Update=CoreWorld
            ?Streamer.StartWithDesired(Desired):Streamer.Start(CameraPositionVoxels);
        const double LOD0Width=static_cast<double>(2*World.ClipRadius+2)
            *CHUNK_SIZE*VOXEL_SIZE/100.0;
        std::printf("[StreamBenchmarkConfig] clip_radius=%d lod0_cube=%.0fm voxel=25cm fullres_levels=%d max_level_filter=%d workers=%d full_detail_tiles=%d desired_tiles=%d variant=%s uv=%s colors=%s material_groups=%s hash=%s\n",
            World.ClipRadius,LOD0Width,World.FullResClipLevels,MaxLevel<0?World.MaxClipLevel:MaxLevel,
            Streamer.GetWorkerCount(),FullDetailExpected,Update.DesiredTiles,
            bRoundTunnels?(bEnhancedWalls?"round-detailed":"round-legacy"):"base",
            (bSkipUvAndColors||bSkipUV)?"skipped":"full",
            (bSkipUvAndColors||bSkipColors)?"skipped":"full",
            bSkipMaterialGroups?"skipped":"full",
            (bHashDensity||!BakeCachePath.empty())?"on":"off");
        std::fflush(stdout);
        int32 Finished=0,FullDetailFinished=0;
        bool bFullDetailTimeReported=false;
        double FullDetailSeconds=0.0;
        const int32 DesiredTotal=Update.DesiredTiles;
        FMeshSheetAudit MeshSheetStats;
        FMeshArtifactAudit MeshArtifactStats;
        FCoarsePanelAudit CoarsePanelStats;
        uint64 MeshAuditedTiles=0;
        while(Finished<DesiredTotal)
        {
            std::vector<FStandaloneCpuTile> Ready=Streamer.TakeReady();
            if(Ready.empty())
            {
                std::this_thread::sleep_for(std::chrono::milliseconds(1));
                continue;
            }
            for(FStandaloneCpuTile& Tile:Ready)
            {
                if(FarAudit)FarAudit->AuditTile(Tile);
                ++Finished;
                const int32 Level=Tile.Key.Level;
                if(PlayerNeighborhoodTiles.contains(Tile.Key))
                {
                    PlayerNeighborhoodReady.insert(Tile.Key);
                    if(PlayerNeighborhoodReady.size()==PlayerNeighborhoodTiles.size())
                        PlayerNeighborhoodCompleted=Tile.CompletedAt;
                }
                if(Level>=0&&Level<=MaxLevels)
                {
                    SamplesByLevel[Level]+=Tile.DensitySamples;
                    VerticesByLevel[Level]+=Tile.Vertices.size();
                    IndicesByLevel[Level]+=Tile.Indices.size();
                    ++TilesByLevel[Level];
                    WorkerSecondsByLevel[Level]+=Tile.WorkerSeconds;
                    const double CompletedSeconds=std::chrono::duration<double>(Tile.CompletedAt-Start).count();
                    FirstCompletionByLevel[Level]=std::min(FirstCompletionByLevel[Level],CompletedSeconds);
                    LastCompletionByLevel[Level]=std::max(LastCompletionByLevel[Level],CompletedSeconds);
                    OutputBytesByLevel[Level]+=static_cast<uint64>(Tile.Vertices.capacity())*sizeof(FStandaloneCpuVertex)
                        +static_cast<uint64>(Tile.Indices.capacity())*sizeof(uint32)
                        +static_cast<uint64>(Tile.TriangleSkirtFaceMasks.capacity())*sizeof(uint8);
                    MeshPayloadAllocatedBytesByLevel[Level]+=Tile.MeshPayloadAllocatedBytes;
                    if(Tile.bHasDensityHash)HashRows[Level].push_back({Tile.Key,Tile.DensityHash});
                    if(Tile.bHasMeshHash)MeshHashRows[Level].push_back({Tile.Key,Tile.MeshHash});
                    if(Tile.bHasPayloadHash)PayloadHashRows[Level].push_back({Tile.Key,Tile.PayloadHash});
                    if(Tile.bHasTopologyHash)TopologyHashRows[Level].push_back({Tile.Key,Tile.TopologyHash});
                }
                if(Level<World.FullResClipLevels)++FullDetailFinished;
                if(bShardAudit&&Level==0)
                {
                    const FMeshSheetAudit TileAudit=AuditThinMeshSheets(Tile,CameraPositionVoxels,
                        CameraYaw,CameraPitch,
                        MeshSheetStats.ThinSheets==0
                            &&MeshSheetStats.SmallComponentsInView
                                +MeshSheetStats.BoundarySmallComponentsInView<8);
                    MeshSheetStats.Components+=TileAudit.Components;
                    MeshSheetStats.SmallComponents+=TileAudit.SmallComponents;
                    MeshSheetStats.ThinSheets+=TileAudit.ThinSheets;
                    MeshSheetStats.SmallComponentsInView+=TileAudit.SmallComponentsInView;
                    MeshSheetStats.BoundarySmallComponentsInView+=TileAudit.BoundarySmallComponentsInView;
                    MeshSheetStats.TrianglesInThinSheets+=TileAudit.TrianglesInThinSheets;
                    MeshSheetStats.MinimumThicknessVoxels=std::min(MeshSheetStats.MinimumThicknessVoxels,
                        TileAudit.MinimumThicknessVoxels);
                    ++MeshAuditedTiles;
                }
                if(bShardAudit&&Level<=1&&CoreWorld)
                {
                    const FMeshArtifactAudit TileAudit=AuditMeshArtifacts(
                        Tile,*CoreWorld,std::max(1,World.SkirtCells)*(1<<Level)*0.25,
                        ComputeVisibleSkirtFaceMask(Tile.Key,DesiredTileSet),
                        MeshArtifactStats.ExposedSkirtTriangles<8,
                        MeshArtifactStats.RepeatedOffsetPatches<8);
                    MeshArtifactStats.Tiles+=TileAudit.Tiles;
                    MeshArtifactStats.SurfaceVertices+=TileAudit.SurfaceVertices;
                    MeshArtifactStats.EdgeVertices+=TileAudit.EdgeVertices;
                    MeshArtifactStats.MidpointVertices+=TileAudit.MidpointVertices;
                    MeshArtifactStats.SkirtTriangles+=TileAudit.SkirtTriangles;
                    MeshArtifactStats.HiddenSameLodSkirtTriangles+=TileAudit.HiddenSameLodSkirtTriangles;
                    MeshArtifactStats.ExposedSkirtTriangles+=TileAudit.ExposedSkirtTriangles;
                    MeshArtifactStats.ReverseSideSolidSkirtTriangles+=TileAudit.ReverseSideSolidSkirtTriangles;
                    MeshArtifactStats.RepeatedOffsetPatches+=TileAudit.RepeatedOffsetPatches;
                    MeshArtifactStats.FieldMismatchPatches+=TileAudit.FieldMismatchPatches;
                    MeshArtifactStats.NormalMismatchPatches+=TileAudit.NormalMismatchPatches;
                    MeshArtifactStats.GradientFallbackPatches+=TileAudit.GradientFallbackPatches;
                    MeshArtifactStats.GradientFallbackSamples+=TileAudit.GradientFallbackSamples;
                    MeshArtifactStats.DetailDisplacementSamples+=TileAudit.DetailDisplacementSamples;
                    MeshArtifactStats.HighDetailDisplacementSamples+=TileAudit.HighDetailDisplacementSamples;
                    MeshArtifactStats.DetailDisplacementPatches+=TileAudit.DetailDisplacementPatches;
                    MeshArtifactStats.MaximumDetailDisplacement=std::max(
                        MeshArtifactStats.MaximumDetailDisplacement,TileAudit.MaximumDetailDisplacement);
                    MeshArtifactStats.DetailGradientSamples+=TileAudit.DetailGradientSamples;
                    MeshArtifactStats.DetailGradientOverLimit+=TileAudit.DetailGradientOverLimit;
                    MeshArtifactStats.MaximumDetailGradient=std::max(
                        MeshArtifactStats.MaximumDetailGradient,TileAudit.MaximumDetailGradient);
                    MeshArtifactStats.CurvedEdgeChecks+=TileAudit.CurvedEdgeChecks;
                    MeshArtifactStats.CurvedEdgeRootPatches+=TileAudit.CurvedEdgeRootPatches;
                    MeshArtifactStats.CurvedEdgeRootBadVertices+=TileAudit.CurvedEdgeRootBadVertices;
                    MeshArtifactStats.MaximumCurvedEdgeRootError=std::max(
                        MeshArtifactStats.MaximumCurvedEdgeRootError,
                        TileAudit.MaximumCurvedEdgeRootError);
                    MeshArtifactStats.FieldEdgeChecks+=TileAudit.FieldEdgeChecks;
                    MeshArtifactStats.AnalyticCrossingMisses+=TileAudit.AnalyticCrossingMisses;
                    MeshArtifactStats.OldInt8RangeEdges+=TileAudit.OldInt8RangeEdges;
                    MeshArtifactStats.NormalChecks+=TileAudit.NormalChecks;
                    MeshArtifactStats.SumAbsSourceResidual+=TileAudit.SumAbsSourceResidual;
                    MeshArtifactStats.SumAnalyticEdgeShiftVoxels+=TileAudit.SumAnalyticEdgeShiftVoxels;
                    if(TileAudit.MaximumAnalyticEdgeShiftVoxels
                        >MeshArtifactStats.MaximumAnalyticEdgeShiftVoxels)
                    {
                        MeshArtifactStats.MaximumAnalyticEdgeShiftVoxels=TileAudit.MaximumAnalyticEdgeShiftVoxels;
                        MeshArtifactStats.WorstShiftPosition=TileAudit.WorstShiftPosition;
                        MeshArtifactStats.WorstShiftFraction=TileAudit.WorstShiftFraction;
                        MeshArtifactStats.WorstShiftD0=TileAudit.WorstShiftD0;
                        MeshArtifactStats.WorstShiftD1=TileAudit.WorstShiftD1;
                        MeshArtifactStats.WorstShiftExactD0=TileAudit.WorstShiftExactD0;
                        MeshArtifactStats.WorstShiftExactD1=TileAudit.WorstShiftExactD1;
                        MeshArtifactStats.WorstShiftAxis=TileAudit.WorstShiftAxis;
                    }
                    if(TileAudit.MaximumAbsSourceResidual>MeshArtifactStats.MaximumAbsSourceResidual)
                    {
                        MeshArtifactStats.MaximumAbsSourceResidual=TileAudit.MaximumAbsSourceResidual;
                        MeshArtifactStats.WorstResidualPosition=TileAudit.WorstResidualPosition;
                    }
                    MeshArtifactStats.SumNormalAgreement+=TileAudit.SumNormalAgreement;
                    MeshArtifactStats.MinimumNormalAgreement=std::min(
                        MeshArtifactStats.MinimumNormalAgreement,TileAudit.MinimumNormalAgreement);
                    MeshArtifactStats.MaximumAirSkirtDensity=std::max(
                        MeshArtifactStats.MaximumAirSkirtDensity,TileAudit.MaximumAirSkirtDensity);
                    MeshArtifactStats.MaximumPatchDominance=std::max(
                        MeshArtifactStats.MaximumPatchDominance,TileAudit.MaximumPatchDominance);
                    for(size_t Bin=0;Bin<MeshArtifactStats.OffsetHistogram.size();++Bin)
                        MeshArtifactStats.OffsetHistogram[Bin]+=TileAudit.OffsetHistogram[Bin];
                }
                if(bShardAudit&&Level>=1&&CoreWorld)
                {
                    const FCoarsePanelAudit TileAudit=AuditCoarseSurfacePanels(
                        Tile,*CoreWorld,CoreWorld->configuration(),CoarsePanelStats.OffSurfacePanels<8);
                    CoarsePanelStats.Tiles+=TileAudit.Tiles;
                    CoarsePanelStats.PlanarOpenComponents+=TileAudit.PlanarOpenComponents;
                    CoarsePanelStats.OffSurfacePanels+=TileAudit.OffSurfacePanels;
                    CoarsePanelStats.Triangles+=TileAudit.Triangles;
                    CoarsePanelStats.MaximumSourceErrorVoxels=std::max(
                        CoarsePanelStats.MaximumSourceErrorVoxels,TileAudit.MaximumSourceErrorVoxels);
                }
                if(!BakeCachePath.empty())CacheTiles.push_back(std::move(Tile));
            }
            if(!bFullDetailTimeReported&&FullDetailFinished>=FullDetailExpected)
            {
                bFullDetailTimeReported=true;
                FullDetailSeconds=std::chrono::duration<double>(
                    std::chrono::steady_clock::now()-Start).count();
                std::vector<FHashedTile> FullRows;
                std::vector<FHashedTile> FullMeshRows;
                std::vector<FHashedTile> FullPayloadRows;
                std::vector<FHashedTile> FullTopologyRows;
                for(int32 Level=0;Level<World.FullResClipLevels&&Level<=MaxLevels;++Level)
                {
                    FullRows.insert(FullRows.end(),HashRows[Level].begin(),HashRows[Level].end());
                    FullMeshRows.insert(FullMeshRows.end(),MeshHashRows[Level].begin(),MeshHashRows[Level].end());
                    FullPayloadRows.insert(FullPayloadRows.end(),PayloadHashRows[Level].begin(),PayloadHashRows[Level].end());
                    FullTopologyRows.insert(FullTopologyRows.end(),TopologyHashRows[Level].begin(),TopologyHashRows[Level].end());
                }
                if(bHashDensity||!BakeCachePath.empty())
                    std::printf("[StreamFullDetail] wall=%.3fs generated=%d/%d resident_tiles_cpu_only=1 density_hash=0x%016llx mesh_hash=0x%016llx payload_hash=0x%016llx topology_hash=0x%016llx\n",
                        FullDetailSeconds,FullDetailFinished,FullDetailExpected,
                        static_cast<unsigned long long>(HashTileRows(FullRows)),
                        static_cast<unsigned long long>(HashTileRows(FullMeshRows)),
                        static_cast<unsigned long long>(HashTileRows(FullPayloadRows)),
                        static_cast<unsigned long long>(HashTileRows(FullTopologyRows)));
                else
                    std::printf("[StreamFullDetail] wall=%.3fs generated=%d/%d resident_tiles_cpu_only=1 hash=off\n",
                        FullDetailSeconds,FullDetailFinished,FullDetailExpected);
                std::fflush(stdout);
            }
        }
        const double FullClipSeconds=std::chrono::duration<double>(
            std::chrono::steady_clock::now()-Start).count();
        if(!PlayerNeighborhoodTiles.empty()&&PlayerNeighborhoodReady.size()==PlayerNeighborhoodTiles.size())
            std::printf("[PlayerNeighborhood] lod0_tiles=%zu/%zu mesh_ready=%.3fs collision_cook=unavailable_in_standalone\n",
                PlayerNeighborhoodReady.size(),PlayerNeighborhoodTiles.size(),
                std::chrono::duration<double>(PlayerNeighborhoodCompleted-Start).count());
        else
            std::printf("[PlayerNeighborhood] lod0_tiles=%zu/%zu mesh_ready=missing collision_cook=unavailable_in_standalone\n",
                PlayerNeighborhoodReady.size(),PlayerNeighborhoodTiles.size());
        const double TotalWorkerSeconds=std::accumulate(WorkerSecondsByLevel.begin(),WorkerSecondsByLevel.end(),0.0);
        const uint64 TotalMeshPayloadAllocatedBytes=std::accumulate(
            MeshPayloadAllocatedBytesByLevel.begin(),MeshPayloadAllocatedBytesByLevel.end(),uint64{0});
        const uint64 TotalViewerOutputBytes=std::accumulate(
            OutputBytesByLevel.begin(),OutputBytesByLevel.end(),uint64{0});
        const double WorkerBusyPercent=100.0*TotalWorkerSeconds
            /(std::max(1,WorkerCount)*std::max(0.001,FullClipSeconds));
        std::vector<FHashedTile> AllRows;
        std::vector<FHashedTile> AllMeshRows;
        std::vector<FHashedTile> AllPayloadRows;
        std::vector<FHashedTile> AllTopologyRows;
        if(bHashDensity||!BakeCachePath.empty())
            for(int32 Level=0;Level<=MaxLevels;++Level)
            {
                AllRows.insert(AllRows.end(),HashRows[Level].begin(),HashRows[Level].end());
                AllMeshRows.insert(AllMeshRows.end(),MeshHashRows[Level].begin(),MeshHashRows[Level].end());
                AllPayloadRows.insert(AllPayloadRows.end(),PayloadHashRows[Level].begin(),PayloadHashRows[Level].end());
                AllTopologyRows.insert(AllTopologyRows.end(),TopologyHashRows[Level].begin(),TopologyHashRows[Level].end());
            }
        if(bHashDensity||!BakeCachePath.empty())
            std::printf("[StreamFullClipmap] wall=%.3fs generated=%d/%d worker_sum=%.3fs worker_busy=%.1f%% worker_idle=%.1f%% payload_alloc=%.2fMiB viewer_output=%.2fMiB tiles=%d density_hash=0x%016llx mesh_hash=0x%016llx payload_hash=0x%016llx topology_hash=0x%016llx\n",
                FullClipSeconds,Finished,DesiredTotal,TotalWorkerSeconds,WorkerBusyPercent,
                std::max(0.0,100.0-WorkerBusyPercent),
                static_cast<double>(TotalMeshPayloadAllocatedBytes)/1048576.0,
                static_cast<double>(TotalViewerOutputBytes)/1048576.0,
                DesiredTotal,static_cast<unsigned long long>(HashTileRows(AllRows)),
                static_cast<unsigned long long>(HashTileRows(AllMeshRows)),
                static_cast<unsigned long long>(HashTileRows(AllPayloadRows)),
                static_cast<unsigned long long>(HashTileRows(AllTopologyRows)));
        else
            std::printf("[StreamFullClipmap] wall=%.3fs generated=%d/%d worker_sum=%.3fs worker_busy=%.1f%% worker_idle=%.1f%% payload_alloc=%.2fMiB viewer_output=%.2fMiB tiles=%d hash=off\n",
                FullClipSeconds,Finished,DesiredTotal,TotalWorkerSeconds,WorkerBusyPercent,
                std::max(0.0,100.0-WorkerBusyPercent),
                static_cast<double>(TotalMeshPayloadAllocatedBytes)/1048576.0,
                static_cast<double>(TotalViewerOutputBytes)/1048576.0,DesiredTotal);
        std::printf("[StreamCacheUse] hits=%llu requested=%d\n",
            static_cast<unsigned long long>(Streamer.GetCacheHits()),DesiredTotal);
        if(bShardAudit)
            std::printf("[ThinMeshShardAudit] level=0 tiles=%llu view_components=%llu small_components=%llu isolated_in_view_small=%llu boundary_in_view_small=%llu thin_sheets=%llu sheet_triangles=%llu thickness_lt=1.000 vox pca_scale=3.5x_sigma status=%s\n",
                static_cast<unsigned long long>(MeshAuditedTiles),
                static_cast<unsigned long long>(MeshSheetStats.Components),
                static_cast<unsigned long long>(MeshSheetStats.SmallComponents),
                static_cast<unsigned long long>(MeshSheetStats.SmallComponentsInView),
                static_cast<unsigned long long>(MeshSheetStats.BoundarySmallComponentsInView),
                static_cast<unsigned long long>(MeshSheetStats.ThinSheets),
                static_cast<unsigned long long>(MeshSheetStats.TrianglesInThinSheets),
                MeshSheetStats.ThinSheets==0&&MeshSheetStats.SmallComponentsInView==0?"PASS":"FAIL");
        if(bShardAudit&&CoreWorld)
        {
            const double MidpointShare=MeshArtifactStats.EdgeVertices
                ?100.0*static_cast<double>(MeshArtifactStats.MidpointVertices)
                    /MeshArtifactStats.EdgeVertices:0.0;
            const double MeanSourceResidual=MeshArtifactStats.FieldEdgeChecks
                ?MeshArtifactStats.SumAbsSourceResidual/MeshArtifactStats.FieldEdgeChecks:0.0;
            const double MeanAnalyticEdgeShift=MeshArtifactStats.FieldEdgeChecks
                ?MeshArtifactStats.SumAnalyticEdgeShiftVoxels
                    /std::max<uint64>(1,MeshArtifactStats.FieldEdgeChecks-MeshArtifactStats.AnalyticCrossingMisses):0.0;
            const double MeanNormalAgreement=MeshArtifactStats.NormalChecks
                ?MeshArtifactStats.SumNormalAgreement/MeshArtifactStats.NormalChecks:1.0;
            std::printf("[MeshFieldFitAudit] edge_checks=%llu analytic_crossing_misses=%llu old_int8_range_edges=%llu source_residual_mean=%.4f max=%.4f vox analytic_edge_shift_mean=%.4f max=%.4f vox sampled_normal_dot_mean=%.4f min=%.4f\n",
                static_cast<unsigned long long>(MeshArtifactStats.FieldEdgeChecks),
                static_cast<unsigned long long>(MeshArtifactStats.AnalyticCrossingMisses),
                static_cast<unsigned long long>(MeshArtifactStats.OldInt8RangeEdges),
                MeanSourceResidual,MeshArtifactStats.MaximumAbsSourceResidual,
                MeanAnalyticEdgeShift,MeshArtifactStats.MaximumAnalyticEdgeShiftVoxels,
                MeanNormalAgreement,MeshArtifactStats.MinimumNormalAgreement);
            std::printf("[MeshFieldFitWorst] shift=%.4f vox vertex=(%.3f,%.3f,%.3f) axis=%d fraction=%.4f packed=(%.4f,%.4f) analytic=(%.4f,%.4f) max_source_residual=%.4f vox vertex=(%.3f,%.3f,%.3f)\n",
                MeshArtifactStats.MaximumAnalyticEdgeShiftVoxels,
                MeshArtifactStats.WorstShiftPosition[0],MeshArtifactStats.WorstShiftPosition[1],
                MeshArtifactStats.WorstShiftPosition[2],MeshArtifactStats.WorstShiftAxis,
                MeshArtifactStats.WorstShiftFraction,MeshArtifactStats.WorstShiftD0,
                MeshArtifactStats.WorstShiftD1,MeshArtifactStats.WorstShiftExactD0,
                MeshArtifactStats.WorstShiftExactD1,MeshArtifactStats.MaximumAbsSourceResidual,
                MeshArtifactStats.WorstResidualPosition[0],MeshArtifactStats.WorstResidualPosition[1],
                MeshArtifactStats.WorstResidualPosition[2]);
            // The restored four-voxel relief creates sparse edge-curvature misses across
            // the full 512-tile view. Keep the connected-block rejection strict, while
            // allowing up to 4% scattered misses that do not form a visible patch.
            const uint64 RootBadLimit=std::max<uint64>(64,
                MeshArtifactStats.CurvedEdgeChecks*40u/1000u);
            const double RootBadShare=MeshArtifactStats.CurvedEdgeChecks
                ?static_cast<double>(MeshArtifactStats.CurvedEdgeRootBadVertices)
                    /MeshArtifactStats.CurvedEdgeChecks:0.0;
            const bool bRootAuditPass=MeshArtifactStats.CurvedEdgeRootPatches==0
                &&MeshArtifactStats.CurvedEdgeRootBadVertices<=RootBadLimit;
            std::printf("[MeshCurvedEdgeRootAudit] midpoint_edges=%llu bad_vertices=%llu bad_share=%.3f%% allowed=%llu patches=%llu max_midpoint_root_error=%.4f vox error_threshold=0.080 status=%s\n",
                static_cast<unsigned long long>(MeshArtifactStats.CurvedEdgeChecks),
                static_cast<unsigned long long>(MeshArtifactStats.CurvedEdgeRootBadVertices),
                RootBadShare*100.0,static_cast<unsigned long long>(RootBadLimit),
                static_cast<unsigned long long>(MeshArtifactStats.CurvedEdgeRootPatches),
                MeshArtifactStats.MaximumCurvedEdgeRootError,
                bRootAuditPass?"PASS":"FAIL");
            std::printf("[MeshDetailGradientAudit] overlap_samples=%llu over_limit=%llu max_gradient=%.4f vox_per_vox limit=0.550 status=%s\n",
                static_cast<unsigned long long>(MeshArtifactStats.DetailGradientSamples),
                static_cast<unsigned long long>(MeshArtifactStats.DetailGradientOverLimit),
                MeshArtifactStats.MaximumDetailGradient,
                MeshArtifactStats.DetailGradientOverLimit==0?"PASS":"FAIL");
            std::printf("[MeshSurfaceArtifactAudit] levels=0,1 tiles=%llu surface_vertices=%llu edge_vertices=%llu midpoint_band_vertices=%llu midpoint_share=%.2f%% visible_skirt_triangles=%llu hidden_same_lod_skirt_triangles=%llu exposed_skirt_triangles=%llu reverse_side_solid=%llu max_air_density=%.3f vox repeated_offset_patches=%llu field_residual_patches=%llu normal_mismatch_patches=%llu gradient_fallback_patches=%llu fallback_gradient_samples=%llu high_detail_vertices=%llu/%llu high_detail_patches=%llu max_detail_displacement=%.3f vox max_patch_dominance=%.3f status=%s\n",
                static_cast<unsigned long long>(MeshArtifactStats.Tiles),
                static_cast<unsigned long long>(MeshArtifactStats.SurfaceVertices),
                static_cast<unsigned long long>(MeshArtifactStats.EdgeVertices),
                static_cast<unsigned long long>(MeshArtifactStats.MidpointVertices),MidpointShare,
                static_cast<unsigned long long>(MeshArtifactStats.SkirtTriangles),
                static_cast<unsigned long long>(MeshArtifactStats.HiddenSameLodSkirtTriangles),
                static_cast<unsigned long long>(MeshArtifactStats.ExposedSkirtTriangles),
                static_cast<unsigned long long>(MeshArtifactStats.ReverseSideSolidSkirtTriangles),
                MeshArtifactStats.MaximumAirSkirtDensity,
                static_cast<unsigned long long>(MeshArtifactStats.RepeatedOffsetPatches),
                static_cast<unsigned long long>(MeshArtifactStats.FieldMismatchPatches),
                static_cast<unsigned long long>(MeshArtifactStats.NormalMismatchPatches),
                static_cast<unsigned long long>(MeshArtifactStats.GradientFallbackPatches),
                static_cast<unsigned long long>(MeshArtifactStats.GradientFallbackSamples),
                static_cast<unsigned long long>(MeshArtifactStats.HighDetailDisplacementSamples),
                static_cast<unsigned long long>(MeshArtifactStats.DetailDisplacementSamples),
                static_cast<unsigned long long>(MeshArtifactStats.DetailDisplacementPatches),
                MeshArtifactStats.MaximumDetailDisplacement,
                MeshArtifactStats.MaximumPatchDominance,
                MeshArtifactStats.ExposedSkirtTriangles==0&&MeshArtifactStats.RepeatedOffsetPatches==0
                    &&MeshArtifactStats.FieldMismatchPatches==0
                    &&MeshArtifactStats.NormalMismatchPatches==0
                    &&MeshArtifactStats.GradientFallbackPatches==0
                    &&MeshArtifactStats.DetailDisplacementPatches==0
                    &&bRootAuditPass
                    &&MeshArtifactStats.DetailGradientOverLimit==0?"PASS":"FAIL");
            std::printf("[CoarsePanelAudit] tiles=%llu planar_open_components=%llu off_surface_panels=%llu panel_triangles=%llu max_source_error=%.3f vox status=%s\n",
                static_cast<unsigned long long>(CoarsePanelStats.Tiles),
                static_cast<unsigned long long>(CoarsePanelStats.PlanarOpenComponents),
                static_cast<unsigned long long>(CoarsePanelStats.OffSurfacePanels),
                static_cast<unsigned long long>(CoarsePanelStats.Triangles),
                CoarsePanelStats.MaximumSourceErrorVoxels,
                CoarsePanelStats.OffSurfacePanels==0?"PASS":"FAIL");
        }
        for(int32 Level=0;Level<=MaxLevels;++Level)
        {
            if(TilesByLevel[Level]==0)continue;
            if(bHashDensity||!BakeCachePath.empty())
                std::printf("[StreamLevel] lod=%d tiles=%llu samples=%.3fM worker_sum=%.3fs first_ready=%.3fs complete=%.3fs worker_busy=%.1f%% payload_alloc=%.2fMiB viewer_output=%.2fMiB vertices=%llu indices=%llu density_hash=0x%016llx mesh_hash=0x%016llx payload_hash=0x%016llx topology_hash=0x%016llx\n",
                    Level,static_cast<unsigned long long>(TilesByLevel[Level]),
                    static_cast<double>(SamplesByLevel[Level])/1.0e6,WorkerSecondsByLevel[Level],
                    FirstCompletionByLevel[Level],LastCompletionByLevel[Level],
                    100.0*WorkerSecondsByLevel[Level]/(WorkerCount*std::max(0.001,LastCompletionByLevel[Level])),
                    static_cast<double>(MeshPayloadAllocatedBytesByLevel[Level])/1048576.0,
                    static_cast<double>(OutputBytesByLevel[Level])/1048576.0,
                    static_cast<unsigned long long>(VerticesByLevel[Level]),
                    static_cast<unsigned long long>(IndicesByLevel[Level]),
                    static_cast<unsigned long long>(HashTileRows(HashRows[Level])),
                    static_cast<unsigned long long>(HashTileRows(MeshHashRows[Level])),
                    static_cast<unsigned long long>(HashTileRows(PayloadHashRows[Level])),
                    static_cast<unsigned long long>(HashTileRows(TopologyHashRows[Level])));
            else
                std::printf("[StreamLevel] lod=%d tiles=%llu samples=%.3fM worker_sum=%.3fs first_ready=%.3fs complete=%.3fs worker_busy=%.1f%% payload_alloc=%.2fMiB viewer_output=%.2fMiB vertices=%llu indices=%llu hash=off\n",
                    Level,static_cast<unsigned long long>(TilesByLevel[Level]),
                    static_cast<double>(SamplesByLevel[Level])/1.0e6,WorkerSecondsByLevel[Level],
                    FirstCompletionByLevel[Level],LastCompletionByLevel[Level],
                    100.0*WorkerSecondsByLevel[Level]/(WorkerCount*std::max(0.001,LastCompletionByLevel[Level])),
                    static_cast<double>(MeshPayloadAllocatedBytesByLevel[Level])/1048576.0,
                    static_cast<double>(OutputBytesByLevel[Level])/1048576.0,
                    static_cast<unsigned long long>(VerticesByLevel[Level]),
                    static_cast<unsigned long long>(IndicesByLevel[Level]));
        }
        if(!BakeCachePath.empty())
        {
            std::string CacheError;
            if(!FStandaloneMeshCache::Save(BakeCachePath,CacheMetadata,CacheTiles,CacheError))
            {
                std::fprintf(stderr,"[MeshCacheBake] failed: %s\n",CacheError.c_str());
                return 8;
            }
            const uint64 Bytes=std::filesystem::file_size(BakeCachePath);
            std::printf("[MeshCacheBake] saved %zu tiles, %.2f MiB to %s\n",CacheTiles.size(),
                static_cast<double>(Bytes)/1048576.0,BakeCachePath.c_str());
        }
        if(Finished!=DesiredTotal||FullDetailFinished!=FullDetailExpected)return 6;
        if(bShardAudit&&(MeshSheetStats.ThinSheets!=0||MeshSheetStats.SmallComponentsInView!=0))return 13;
        if(bShardAudit&&CoreWorld&&(MeshArtifactStats.ExposedSkirtTriangles!=0
            ||MeshArtifactStats.RepeatedOffsetPatches!=0||MeshArtifactStats.FieldMismatchPatches!=0
            ||MeshArtifactStats.NormalMismatchPatches!=0||MeshArtifactStats.GradientFallbackPatches!=0
            ||MeshArtifactStats.CurvedEdgeRootPatches!=0
            ||MeshArtifactStats.CurvedEdgeRootBadVertices>std::max<uint64>(
                64,MeshArtifactStats.CurvedEdgeChecks*40u/1000u)
            ||MeshArtifactStats.DetailGradientOverLimit!=0))return 13;
        if(bShardAudit&&CoreWorld&&CoarsePanelStats.OffSurfacePanels!=0)return 13;
        if(FarAudit)
        {
            FarAudit->Finalize();
            FarAudit->Print();
            if(!FarAudit->Passing())return 14;
        }
        return 0;
    }

    int RunEditTransactionBenchmark(FStandaloneOwnerWorld& World,VoxelForgeCore::World& Shape,
                                    const FVector& CameraPositionVoxels,int32 WorkerCount)
    {
        using Clock=std::chrono::steady_clock;
        // Place the spell across the LOD0/LOD1 clipmap seam so one edit exercises every visible
        // level it intersects, instead of timing only the center LOD0 tiles.
        const double X=CameraPositionVoxels.X+3.0*CHUNK_SIZE;
        const double Y=CameraPositionVoxels.Y,Z=CameraPositionVoxels.Z;
        FStandaloneDesiredTileSet Desired=StandaloneWorldMesher::BuildDesiredSet(World,CameraPositionVoxels);
        const auto CoreDesired=Shape.desiredTiles(
            {CameraPositionVoxels.X,CameraPositionVoxels.Y,CameraPositionVoxels.Z});
        Desired.Tiles.clear();Desired.Tiles.reserve(CoreDesired.size());
        std::vector<VoxelForgeCore::TileKey> CoreKeys;CoreKeys.reserve(CoreDesired.size());
        for(const auto& Key:CoreDesired)
        {
            Desired.Tiles.emplace_back(FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
            CoreKeys.push_back(Key);
        }
        VoxelForgeCore::EditTransaction Transaction;Transaction.sequence=1;Transaction.brushes.resize(3);
        Transaction.brushes[0].shape=VoxelForgeCore::EditBrush::Shape::sphere;
        Transaction.brushes[0].operation=VoxelForgeCore::EditBrush::Operation::carve;
        Transaction.brushes[0].a={X,Y,Z};Transaction.brushes[0].radius=20.0;
        Transaction.brushes[1].shape=VoxelForgeCore::EditBrush::Shape::roundedBox;
        Transaction.brushes[1].operation=VoxelForgeCore::EditBrush::Operation::carve;
        Transaction.brushes[1].a={X,Y,Z+12.0};Transaction.brushes[1].halfExtent={11.0,11.0,4.0};
        Transaction.brushes[1].rounding=3.0;
        Transaction.brushes[2].shape=VoxelForgeCore::EditBrush::Shape::capsule;
        Transaction.brushes[2].operation=VoxelForgeCore::EditBrush::Operation::addRock;
        Transaction.brushes[2].a={X-22.0,Y,Z-8.0};Transaction.brushes[2].b={X+22.0,Y,Z-8.0};
        Transaction.brushes[2].radius=3.0;
        const auto EditStart=Clock::now();
        std::string EditError;
        const auto Edit=Shape.applyEditTransaction(Transaction,CoreKeys,EditError);
        if(!Edit){std::fprintf(stderr,"[EditTransaction] rejected: %s\n",EditError.c_str());return 9;}
        std::vector<FVoxelTileKey> Affected;Affected.reserve(128);
        std::array<uint64,MaxLevels+1> AffectedByLevel{},FinishedByLevel{};
        for(const auto& LodTiles:Edit->affectedByLod)
        {
            for(const auto& Key:LodTiles)
            {
                const FVoxelTileKey Tile(FIntVector(Key.coordinate.x,Key.coordinate.y,Key.coordinate.z),Key.lod);
                Affected.push_back(Tile);
                if(Tile.Level>=0&&Tile.Level<=MaxLevels)++AffectedByLevel[Tile.Level];
            }
        }
        if(Affected.empty())
        {
            std::fprintf(stderr,"[EditTransaction] failed: no desired tiles intersect the edit bounds.\n");
            return 9;
        }
        FStandaloneStreamingGenerator Streamer(World,WorkerCount,false,nullptr,-1,
            false,false,false,false,&Shape);
        Streamer.SetDesiredTileFilter(Affected);
        const auto Start=Clock::now();
        const FStandaloneStreamingUpdate Update=Streamer.Start(CameraPositionVoxels);
        int32 Finished=0;
        double WorkerSeconds=0.0;
        while(Finished<Update.DesiredTiles)
        {
            std::vector<FStandaloneCpuTile> Ready=Streamer.TakeReady();
            if(Ready.empty()){std::this_thread::sleep_for(std::chrono::milliseconds(1));continue;}
            for(const FStandaloneCpuTile& Tile:Ready)
            {
                ++Finished;WorkerSeconds+=Tile.WorkerSeconds;
                if(Tile.Key.Level>=0&&Tile.Key.Level<=MaxLevels)++FinishedByLevel[Tile.Key.Level];
            }
        }
        const double RemeshWall=std::chrono::duration<double>(Clock::now()-Start).count();
        Streamer.Stop();
        std::printf("[EditTransaction] operation=crater_batch transaction_count=1 brushes=3 diameter=10.0m offset_from_player=%.2fm affected_tiles=%d/%zu workers=%d cache_hits=%llu remesh_wall=%.3fs worker_sum=%.3fs edit_to_ready=%.3fs status=%s\n",
            3.0*CHUNK_SIZE*VOXEL_SIZE/100.0,
            Finished,Affected.size(),WorkerCount,static_cast<unsigned long long>(Streamer.GetCacheHits()),
            RemeshWall,WorkerSeconds,std::chrono::duration<double>(Clock::now()-EditStart).count(),
            Finished==static_cast<int32>(Affected.size())?"PASS":"FAIL");
        for(int32 Level=0;Level<=MaxLevels;++Level)
            if(AffectedByLevel[Level])
                std::printf("[EditTransactionLevel] lod=%d tiles=%llu remeshed=%llu status=%s\n",Level,
                    static_cast<unsigned long long>(AffectedByLevel[Level]),
                    static_cast<unsigned long long>(FinishedByLevel[Level]),
                    AffectedByLevel[Level]==FinishedByLevel[Level]?"PASS":"FAIL");
        return Finished==static_cast<int32>(Affected.size())?0:10;
    }

    bool MakeRegionBuildRequest(std::span<const VoxelForgeCore::TileKey> tiles,
        const VoxelForgeCore::WorldConfiguration& config,int32 workers,
        VoxelForgeCore::DetailVariant detail,bool roundTunnels,bool enhancedWalls,
        VoxelForgeCore::ShapeBuildRequest& request,std::string& error)
    {
        FIntVector fineMin(MAX_int32,MAX_int32,MAX_int32),fineMax(MIN_int32,MIN_int32,MIN_int32);
        FIntVector clipMin(MAX_int32,MAX_int32,MAX_int32),clipMax(MIN_int32,MIN_int32,MIN_int32);
        for(const auto& key:tiles)
        {
            if(key.lod<0||key.lod>8){error="desired tile has an invalid LOD";return false;}
            const int32 extent=CHUNK_SIZE<<key.lod;
            const FIntVector origin(
                static_cast<int32>(static_cast<int64>(key.coordinate.x)*extent),
                static_cast<int32>(static_cast<int64>(key.coordinate.y)*extent),
                static_cast<int32>(static_cast<int64>(key.coordinate.z)*extent));
            for(int32 axis=0;axis<3;++axis)
            {
                (&clipMin.X)[axis]=std::min((&clipMin.X)[axis],(&origin.X)[axis]);
                (&clipMax.X)[axis]=std::max((&clipMax.X)[axis],(&origin.X)[axis]+extent);
                if(key.lod<config.fullResolutionLodCount)
                {
                    (&fineMin.X)[axis]=std::min((&fineMin.X)[axis],(&origin.X)[axis]);
                    (&fineMax.X)[axis]=std::max((&fineMax.X)[axis],(&origin.X)[axis]+extent);
                }
            }
        }
        if(fineMin.X==MAX_int32||clipMin.X==MAX_int32)
        {error="desired region contains no fine tiles or clip bounds";return false;}
        request.minVoxel={fineMin.X,fineMin.Y,fineMin.Z};
        request.maxVoxel={fineMax.X,fineMax.Y,fineMax.Z};
        request.clipMinVoxel={clipMin.X,clipMin.Y,clipMin.Z};
        request.clipMaxVoxel={clipMax.X,clipMax.Y,clipMax.Z};
        request.workerLimit=static_cast<uint32>(std::clamp(workers,1,8));
        request.detail=detail;request.roundTunnels=roundTunnels;
        request.enhancedTunnelWalls=enhancedWalls;request.includeFarWorld=false;
        return true;
    }

    int RunTilePathHashAudit(VoxelForgeCore::World& base,
        const VoxelForgeCore::ResolvedWorldSnapshot& snapshot,
        const VoxelForgeCore::JobScheduler& scheduler,const FVector& camera,
    int32 workers,VoxelForgeCore::DetailVariant detail,bool roundTunnels,
        bool enhancedWalls)
    {
        const auto baseTiles=base.desiredTiles({camera.X,camera.Y,camera.Z});
        const auto& config=base.configuration();
        if(baseTiles.empty())
        {std::fprintf(stderr,"[TilePathHashAudit] camera window is empty.\n");return 11;}
        constexpr int32 TargetChunkOffset=40;
        const int32 CameraChunkX=FMath::FloorToInt(camera.X/CHUNK_SIZE);
        const int32 CameraChunkY=FMath::FloorToInt(camera.Y/CHUNK_SIZE);
        const int32 CameraChunkZ=FMath::FloorToInt(camera.Z/CHUNK_SIZE);
        const VoxelForgeCore::TileKey target{{CameraChunkX+TargetChunkOffset,
            CameraChunkY,CameraChunkZ},0};
        const FVector alternateCamera((target.coordinate.x+0.5)*CHUNK_SIZE,camera.Y,camera.Z);
        const auto alternateTiles=base.desiredTiles(
            {alternateCamera.X,alternateCamera.Y,alternateCamera.Z});
        if(alternateTiles.empty())
        {std::fprintf(stderr,"[TilePathHashAudit] target camera window is empty.\n");return 11;}

        VoxelForgeCore::TileGenerationOptions options;options.hashDensity=true;
        std::string error;
        const std::array<VoxelForgeCore::TileKey,1> pageTiles={target};
        std::size_t pageTilesA=0;
        if(!base.compileTileRegion(pageTiles,static_cast<uint32>(std::clamp(workers,1,8)),
            scheduler,error,&pageTilesA,nullptr,false))
        {std::fprintf(stderr,"[TilePathHashAudit] path-A page compile failed: %s\n",error.c_str());return 11;}
        auto pathA=base.generateTile(target,options,error);
        if(!pathA){std::fprintf(stderr,"[TilePathHashAudit] path-A generation failed: %s\n",error.c_str());return 11;}

        VoxelForgeCore::World alternate(snapshot);
        VoxelForgeCore::ShapeBuildRequest alternateRequest;
        if(!MakeRegionBuildRequest(alternateTiles,config,workers,detail,roundTunnels,
            enhancedWalls,alternateRequest,error)
            ||!alternate.compile(alternateRequest,scheduler,error))
        {std::fprintf(stderr,"[TilePathHashAudit] alternate camera compile failed: %s\n",error.c_str());return 11;}
        std::size_t pageTilesB=0;
        auto pathB=alternate.generateTile(target,options,error);
        if(!pathB){std::fprintf(stderr,"[TilePathHashAudit] path-B generation failed: %s\n",error.c_str());return 11;}
        const auto Same=[&](const VoxelForgeCore::TileMesh& a,const VoxelForgeCore::TileMesh& b)
        {
            return a.densityHash==b.densityHash&&a.meshHash==b.meshHash
                &&a.payloadHash==b.payloadHash&&a.topologyHash==b.topologyHash;
        };
        const int32 CameraDeltaX=static_cast<int32>(std::lround(alternateCamera.X-camera.X));
        const bool pass=pageTilesA==1&&pageTilesB==0&&Same(*pathA,*pathB);
        std::printf("[TilePathHashAudit] tile=(%d,%d,%d) lod=%d path_a=late_page path_b=initial_base camera_delta_vox=(%d,0,0) page_covered_tiles=(%zu,%zu) density=(0x%016llx,0x%016llx) mesh=(0x%016llx,0x%016llx) payload=(0x%016llx,0x%016llx) topology=(0x%016llx,0x%016llx) status=%s\n",
            target.coordinate.x,target.coordinate.y,target.coordinate.z,target.lod,CameraDeltaX,
            pageTilesA,pageTilesB,
            static_cast<unsigned long long>(pathA->densityHash),
            static_cast<unsigned long long>(pathB->densityHash),
            static_cast<unsigned long long>(pathA->meshHash),
            static_cast<unsigned long long>(pathB->meshHash),
            static_cast<unsigned long long>(pathA->payloadHash),
            static_cast<unsigned long long>(pathB->payloadHash),
            static_cast<unsigned long long>(pathA->topologyHash),
            static_cast<unsigned long long>(pathB->topologyHash),pass?"PASS":"FAIL");

        VoxelForgeCore::World analyticForward(snapshot),analyticReverse(snapshot);
        VoxelForgeCore::ShapeBuildRequest forwardRequest,reverseRequest;
        const FVector reverseCamera((target.coordinate.x+TargetChunkOffset+0.5)*CHUNK_SIZE,
            camera.Y,camera.Z);
        const auto reverseTiles=base.desiredTiles(
            {reverseCamera.X,reverseCamera.Y,reverseCamera.Z});
        if(!MakeRegionBuildRequest(baseTiles,config,workers,detail,roundTunnels,
                enhancedWalls,forwardRequest,error)
            ||!analyticForward.compile(forwardRequest,scheduler,error)
            ||reverseTiles.empty()
            ||!MakeRegionBuildRequest(reverseTiles,config,workers,detail,roundTunnels,
                enhancedWalls,reverseRequest,error)
            ||!analyticReverse.compile(reverseRequest,scheduler,error))
        {std::fprintf(stderr,"[TilePathHashDirections] setup failed: %s\n",error.c_str());return 11;}
        std::size_t analyticForwardCovered=0,analyticReverseCovered=0;
        VoxelForgeCore::TileRegionCompileStats forwardStats,reverseStats;
        if(!analyticForward.compileTileRegion(pageTiles,static_cast<uint32>(std::clamp(workers,1,8)),
                scheduler,error,&analyticForwardCovered,&forwardStats,true)
            ||!analyticReverse.compileTileRegion(pageTiles,static_cast<uint32>(std::clamp(workers,1,8)),
                scheduler,error,&analyticReverseCovered,&reverseStats,true))
        {std::fprintf(stderr,"[TilePathHashDirections] region compile failed: %s\n",error.c_str());return 11;}
        auto analyticPathForward=analyticForward.generateTile(target,options,error);
        auto analyticPathReverse=analyticReverse.generateTile(target,options,error);
        if(!analyticPathForward||!analyticPathReverse)
        {std::fprintf(stderr,"[TilePathHashDirections] tile generation failed: %s\n",error.c_str());return 11;}
        const bool analyticDirectionsSame=Same(*analyticPathForward,*analyticPathReverse);
        const bool packedAnalyticRenderSame=pathA->meshHash==analyticPathForward->meshHash
            &&pathA->payloadHash==analyticPathForward->payloadHash
            &&pathA->topologyHash==analyticPathForward->topologyHash;
        const bool packedAnalyticDensitySame=pathA->densityHash==analyticPathForward->densityHash;
        std::printf("[TilePathHashDirections] tile=(%d,%d,%d) lod=%d approach_deltas=(-%d,+%d) regions=(%zu,%zu) density=(0x%016llx,0x%016llx) mesh=(0x%016llx,0x%016llx) payload=(0x%016llx,0x%016llx) topology=(0x%016llx,0x%016llx) status=%s\n",
            target.coordinate.x,target.coordinate.y,target.coordinate.z,target.lod,
            TargetChunkOffset,TargetChunkOffset,analyticForwardCovered,analyticReverseCovered,
            static_cast<unsigned long long>(analyticPathForward->densityHash),
            static_cast<unsigned long long>(analyticPathReverse->densityHash),
            static_cast<unsigned long long>(analyticPathForward->meshHash),
            static_cast<unsigned long long>(analyticPathReverse->meshHash),
            static_cast<unsigned long long>(analyticPathForward->payloadHash),
            static_cast<unsigned long long>(analyticPathReverse->payloadHash),
            static_cast<unsigned long long>(analyticPathForward->topologyHash),
            static_cast<unsigned long long>(analyticPathReverse->topologyHash),
            analyticDirectionsSame&&analyticForwardCovered==1&&analyticReverseCovered==1?"PASS":"FAIL");
        std::printf("[TilePathPackedAnalyticParity] tile=(%d,%d,%d) density=(0x%016llx,0x%016llx) density_representation=%s mesh=(0x%016llx,0x%016llx) payload=(0x%016llx,0x%016llx) topology=(0x%016llx,0x%016llx) status=%s\n",
            target.coordinate.x,target.coordinate.y,target.coordinate.z,
            static_cast<unsigned long long>(pathA->densityHash),
            static_cast<unsigned long long>(analyticPathForward->densityHash),
            packedAnalyticDensitySame?"same":"different",
            static_cast<unsigned long long>(pathA->meshHash),
            static_cast<unsigned long long>(analyticPathForward->meshHash),
            static_cast<unsigned long long>(pathA->payloadHash),
            static_cast<unsigned long long>(analyticPathForward->payloadHash),
            static_cast<unsigned long long>(pathA->topologyHash),
            static_cast<unsigned long long>(analyticPathForward->topologyHash),
            packedAnalyticRenderSame?"PASS":"FAIL");
        return pass&&analyticDirectionsSame&&analyticForwardCovered==1
            &&analyticReverseCovered==1&&packedAnalyticRenderSame?0:12;
    }
}

int main(int Argc,char** Argv)
{
    SetPriorityClass(GetCurrentProcess(),BELOW_NORMAL_PRIORITY_CLASS);
    const bool bAllCpuAffinity=std::find_if(Argv+1,Argv+Argc,[](const char* Arg)
        {return std::string_view(Arg)=="--affinity-all-cpu";})!=Argv+Argc;
    const DWORD_PTR RequestedAffinity=static_cast<DWORD_PTR>(bAllCpuAffinity?0xFFFF:0xFF00);
    const BOOL bAffinitySet=SetProcessAffinityMask(GetCurrentProcess(),RequestedAffinity);
    DWORD_PTR ProcessAffinity=0,SystemAffinity=0;
    GetProcessAffinityMask(GetCurrentProcess(),&ProcessAffinity,&SystemAffinity);
    std::printf("[ProcessControls] priority_class=BelowNormal affinity_mask=0x%llX set=%s\n",
        static_cast<unsigned long long>(ProcessAffinity),bAffinitySet?"true":"false");
    const auto ColdProcessStart=std::chrono::steady_clock::now();
    FApp App;App.DataPath=DefaultDataPath().string();App.ColdProcessStart=ColdProcessStart;
    wchar_t ExecutablePath[MAX_PATH]{};
    const DWORD ExecutablePathLength=GetModuleFileNameW(nullptr,ExecutablePath,MAX_PATH);
    if(ExecutablePathLength>0&&ExecutablePathLength<MAX_PATH)
        App.GpuFarShaderPath=std::filesystem::path(ExecutablePath).parent_path().parent_path()
            /"Research"/"OwnerShapeWorld"/"gpu_first_port"/"VoxelForgeGpuMesher.hlsl";
    std::string JoinedCommandLine;
    for(int32 I=1;I<Argc;++I)
    {if(!JoinedCommandLine.empty())JoinedCommandLine.push_back(' ');JoinedCommandLine+=Argv[I];}
    FCommandLine::Set(JoinedCommandLine);
    int32 BenchmarkLevel=-1,BenchmarkCount=64,StreamBenchmarkMaxLevel=-1;
    int32 MaxClipLevelOverride=-1;
    int32 ShapeCompareLevel=-1,ShapeCompareTileLimit=0;
    std::string ShapeCompareArea="unlabeled";
    bool bStreamBenchmark=false;
    bool bStreamPipeline=false;
    bool bProfileDetailBand=false;
    bool bPassageDetailAuditOnly=false;
    bool bGridStairAudit=false;
    bool bTilePathHashAudit=false;
    bool bShardAudit=false;
    bool bFarSurvivalAudit=false;
    bool bFarAuditBadShell=false;
    bool bStreamHash=false;
    bool bSkipUvAndColors=false;
    bool bSkipUV=false;
    bool bSkipColors=false;
    bool bSkipMaterialGroups=false;
    bool bEditBenchmark=false;
    bool bRoundWallCompare=false;
    bool bCheckOwnerReliefBands=false;
    int32 RoundCompareTileCount=16;
    std::string BakeMeshCachePath;
    for(int32 I=1;I<Argc;++I)
    {
        const std::string_view Arg=Argv[I];
        if(Arg.starts_with("--capture="))App.CapturePath=std::string(Arg.substr(10));
        else if(Arg=="--no-vsync")App.bDisableVSync=true;
        else if(Arg=="--blocking-startup-present")App.bForceBlockingStartupPresent=true;
        else if(Arg=="--cpu-frame-before-gpu")App.bPresentCpuFrameBeforeGpu=true;
        else if(Arg=="--hold-cpu-frame-during-gpu-warmup")
            App.bHoldCpuFrameDuringGpuWarmup=true,App.bPresentCpuFrameBeforeGpu=true;
        else if(Arg=="--gpu-skirt-draw-audit")App.bAuditGpuSkirtDrawArgs=true;
        else if(Arg=="--gpu-skirts-disabled")App.bGpuSkirtEmitEnabled=false;
        else if(Arg=="--gpu-connectors-disabled")App.bGpuConnectorRepairEnabled=false;
        else if(Arg=="--profile-attribution")App.bProfileAttribution=true;
        else if(Arg=="--profile-detail-band")bProfileDetailBand=true;
        else if(Arg=="--compare-evaluators")App.bCompareEvaluators=true;
        else if(Arg=="--parity-only")App.bParityOnly=true;
        else if(Arg=="--density-hash")App.bDensityHash=true;
        else if(Arg=="--shape-world")App.bShapeWorld=true;
        else if(Arg=="--gpu-far")
            App.bGpuFarEnabled=true,App.bGpuFarShaderPrewarm=true,App.bShapeWorld=true,App.bForceGenerate=true,
            App.bPresentCpuFrameBeforeGpu=true,App.bHoldCpuFrameDuringGpuWarmup=true;
        else if(Arg=="--gpu-far-prewarm")
            App.bGpuFarEnabled=true,App.bGpuFarShaderPrewarm=true,App.bShapeWorld=true,
            App.bForceGenerate=true,App.bPresentCpuFrameBeforeGpu=true,
            App.bHoldCpuFrameDuringGpuWarmup=true;
        else if(Arg=="--gpu-far-warmup-resources-only")
            App.bGpuFarEnabled=true,App.bGpuFarShaderPrewarm=true,App.bGpuFarWarmupResourcesOnly=true,
            App.bShapeWorld=true,App.bForceGenerate=true,App.bPresentCpuFrameBeforeGpu=true,
            App.bHoldCpuFrameDuringGpuWarmup=true;
        else if(Arg=="--gpu-far-warmup-normal-priority")
            App.bGpuFarWarmupNormalPriority=true;
        else if(Arg=="--gpu-far-warmup-before-workers")
            App.bGpuFarEnabled=true,App.bGpuFarShaderPrewarm=true,App.bGpuFarWarmupBeforeWorkers=true,
            App.bShapeWorld=true,App.bForceGenerate=true,App.bPresentCpuFrameBeforeGpu=true,
            App.bHoldCpuFrameDuringGpuWarmup=true;
        else if(Arg=="--gpu-far-warmup-only")
            App.bGpuFarEnabled=true,App.bGpuFarShaderPrewarm=true,App.bGpuFarWarmupOnly=true,
            App.bShapeWorld=true,App.bForceGenerate=true;
        else if(Arg.starts_with("--gpu-far-warmup-entry="))
            App.GpuWarmupEntryFilter=std::string(Arg.substr(std::string_view("--gpu-far-warmup-entry=").size()));
        else if(Arg=="--affinity-all-cpu"){}
        else if(Arg=="--gpu-far-no-present-hold")
            App.bDisableGpuWarmupPresentHold=true;
        else if(Arg=="--gpu-far-no-prewarm")
            App.bGpuFarShaderPrewarm=false;
        else if(Arg=="--cpu-far")App.bGpuFarEnabled=false,App.bShapeWorld=true,App.bForceGenerate=true;
        else if(Arg=="--gpu-far-audit")
        {App.bGpuFarEnabled=true;App.bGpuFarAuditEnabled=true;App.bShapeWorld=true;App.bForceGenerate=true;App.bQuitAfterClipmap=true;}
        else if(Arg=="--shape-audit")App.bShapeWorld=true,App.bShapeAuditOnly=true;
        else if(Arg=="--grid-stair-audit")bGridStairAudit=true,App.bShapeWorld=true;
        else if(Arg=="--tile-path-hash-audit")bTilePathHashAudit=true,App.bShapeWorld=true;
        else if(Arg=="--passage-wall-detail-audit")
        {App.bShapeWorld=true;App.bShapeAuditOnly=true;bPassageDetailAuditOnly=true;}
        else if(Arg=="--shape-compare")App.bShapeWorld=true,App.bShapeCompare=true;
        else if(Arg.starts_with("--shape-compare-level="))
            ShapeCompareLevel=std::atoi(std::string(Arg.substr(std::string_view("--shape-compare-level=").size())).c_str());
        else if(Arg.starts_with("--shape-compare-tile-limit="))
            ShapeCompareTileLimit=std::max(0,std::atoi(std::string(Arg.substr(std::string_view("--shape-compare-tile-limit=").size())).c_str()));
        else if(Arg.starts_with("--shape-compare-area="))
            ShapeCompareArea=std::string(Arg.substr(std::string_view("--shape-compare-area=").size()));
        else if(Arg=="--owner-relief-band-audit")bCheckOwnerReliefBands=true;
        else if(Arg.starts_with("--shape-detail="))
        {
            const std::string_view Value=Arg.substr(15);
            if(Value=="smooth")App.ShapeDetailVariant=VoxelForgeCore::DetailVariant::smooth;
            else if(Value=="per-voxel")App.ShapeDetailVariant=VoxelForgeCore::DetailVariant::perVoxel;
            else if(Value!="coherent"){std::fprintf(stderr,"Invalid --shape-detail value: %.*s\n",static_cast<int>(Value.size()),Value.data());return 2;}
        }
        else if(Arg=="--round-tunnels")App.bRoundTunnels=true;
        else if(Arg=="--round-tunnels-legacy")
        {App.bRoundTunnels=true;App.bRoundTunnelEnhanced=false;}
        else if(Arg=="--stream-benchmark")bStreamBenchmark=true;
        else if(Arg=="--stream-pipeline")
        {bStreamPipeline=true;bStreamBenchmark=true;App.bShapeWorld=true;App.bForceGenerate=true;}
        else if(Arg=="--edit-benchmark")
        {bEditBenchmark=true;bStreamBenchmark=true;App.bShapeWorld=true;}
        else if(Arg=="--stream-hash")bStreamHash=true;
        else if(Arg=="--skip-uv-colors")bSkipUvAndColors=true;
        else if(Arg=="--skip-uv")bSkipUV=true;
        else if(Arg=="--skip-colors")bSkipColors=true;
        else if(Arg=="--skip-material-groups")bSkipMaterialGroups=true;
        else if(Arg=="--shard-audit")bShardAudit=true;
        else if(Arg=="--far-survival-audit")bFarSurvivalAudit=true,App.bShapeWorld=true;
        else if(Arg=="--far-audit-bad-shell")
        {bFarAuditBadShell=true;bFarSurvivalAudit=true;App.bShapeWorld=true;}
        else if(Arg=="--compare-round-walls")
        {bRoundWallCompare=true;App.bRoundTunnels=true;}
        else if(Arg.starts_with("--round-compare-tiles="))
            RoundCompareTileCount=std::max(1,std::atoi(std::string(Arg.substr(22)).c_str()));
        else if(Arg=="--force-generate")App.bForceGenerate=true;
        else if(Arg.starts_with("--bake-mesh-cache="))
        {BakeMeshCachePath=std::string(Arg.substr(18));App.bForceGenerate=true;}
        else if(Arg.starts_with("--stream-max-level="))
            StreamBenchmarkMaxLevel=std::atoi(std::string(Arg.substr(19)).c_str());
        else if(Arg.starts_with("--max-clip-level="))
            MaxClipLevelOverride=std::clamp(
                std::atoi(std::string(Arg.substr(17)).c_str()),0,8);
        else if(Arg=="--test-one-chunk-crossing")
        {
            App.FlightTargetCrossings=1;
            App.bShapeWorld=true;
        }
        else if(Arg.starts_with("--test-flight-crossings="))
        {
            App.FlightTargetCrossings=std::max(1,std::atoi(
                std::string(Arg.substr(std::string_view("--test-flight-crossings=").size())).c_str()));
            App.bShapeWorld=true;
        }
        else if(Arg.starts_with("--flight-speed-mps="))
            App.FlightSpeedMps=std::clamp(std::strtof(
                std::string(Arg.substr(std::string_view("--flight-speed-mps=").size())).c_str(),nullptr),
                1.0f,256.0f);
        else if(Arg=="--capture-after-clipmap")App.bCaptureAfterClipmap=true;
        else if(Arg=="--quit-after-clipmap")App.bQuitAfterClipmap=true;
        else if(Arg.starts_with("--workers="))
        {
            const int32 Parsed=std::atoi(std::string(Arg.substr(10)).c_str());
            App.WorkerCount=std::clamp(Parsed,1,8);
        }
        else if(Arg.starts_with("--clip-radius="))
            App.ClipRadius=std::clamp(std::atoi(std::string(Arg.substr(14)).c_str()),1,64);
        else if(Arg.starts_with("--capture-tile-limit="))
            App.CaptureTileLimit=std::clamp(std::atoi(
                std::string(Arg.substr(std::string_view("--capture-tile-limit=").size())).c_str()),1,4096);
        else if(Arg.starts_with("--capture-only-lod="))
            App.CaptureOnlyLod=std::clamp(std::atoi(
                std::string(Arg.substr(std::string_view("--capture-only-lod=").size())).c_str()),0,8);
        else if(Arg.starts_with("--capture-only-tile="))
        {
            const std::string Values(Arg.substr(std::string_view("--capture-only-tile=").size()));
            if(std::sscanf(Values.c_str(),"%d,%d,%d,%d",&App.CaptureOnlyTileX,
                &App.CaptureOnlyTileY,&App.CaptureOnlyTileZ,&App.CaptureOnlyTileLod)==4)
                App.bCaptureOnlyTile=true;
            else {std::fprintf(stderr,"Invalid --capture-only-tile value: %s\n",Values.c_str());return 2;}
        }
        else if(Arg.starts_with("--draw-only-tile="))
        {
            const std::string Values(Arg.substr(std::string_view("--draw-only-tile=").size()));
            if(std::sscanf(Values.c_str(),"%d,%d,%d,%d",&App.DrawOnlyTileX,
                &App.DrawOnlyTileY,&App.DrawOnlyTileZ,&App.DrawOnlyTileLod)==4)
                App.bDrawOnlyTile=true;
            else {std::fprintf(stderr,"Invalid --draw-only-tile value: %s\n",Values.c_str());return 2;}
        }
        else if(Arg.starts_with("--capture-output-slot="))
            App.CaptureOutputSlot=std::max(-1,std::atoi(
                std::string(Arg.substr(std::string_view("--capture-output-slot=").size())).c_str()));
        else if(Arg.starts_with("--benchmark-level="))
            BenchmarkLevel=std::atoi(std::string(Arg.substr(18)).c_str());
        else if(Arg.starts_with("--benchmark-count="))
            BenchmarkCount=std::max(1,std::atoi(std::string(Arg.substr(18)).c_str()));
        else if(Arg.starts_with("--camera="))
        {
            const std::string Values(Arg.substr(9));
            float X=0,Y=0,Z=0,YawDegrees=0,PitchDegrees=0;
            if(std::sscanf(Values.c_str(),"%f,%f,%f,%f,%f",&X,&Y,&Z,&YawDegrees,&PitchDegrees)==5)
            {
                App.bHasCameraOverride=true;App.CameraPositionOverride={X,Y,Z};
                App.CameraYawOverride=YawDegrees*0.017453292519943295f;
                App.CameraPitchOverride=PitchDegrees*0.017453292519943295f;
            }
            else {std::fprintf(stderr,"Invalid --camera value: %s\n",Values.c_str());return 2;}
        }
        else if(!Arg.empty()&&Arg.front()!='-')App.DataPath=std::string(Arg);
    }
    if(App.bCaptureOnlyTile&&App.bGpuFarEnabled)
    {
        // A single far-tile probe may have no CPU fallback batch to present.
        // Let it submit directly instead of waiting for the normal CPU-first
        // presentation gate that requires a nonempty CPU batch.
        App.bPresentCpuFrameBeforeGpu=false;
        App.bHoldCpuFrameDuringGpuWarmup=false;
    }
    if(App.bDisableGpuWarmupPresentHold)App.bHoldCpuFrameDuringGpuWarmup=false;
    App.bProfileDetailBand=bProfileDetailBand;
    std::string Error;
    std::printf("[ColdTimeline] stage=owner_data_load begin_ms=%.3f\n",
        std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count());
    App.World=std::make_shared<FStandaloneOwnerWorld>();
    {
        std::ifstream JsonFile(App.DataPath,std::ios::binary);
        if(!JsonFile){std::fprintf(stderr,"Owner data load failed: cannot open %s\n",App.DataPath.c_str());return 2;}
        std::string OwnerJsonText{std::istreambuf_iterator<char>(JsonFile),std::istreambuf_iterator<char>{}};
        if(!JsonFile&& !JsonFile.eof())
        {std::fprintf(stderr,"Owner data load failed: read error for %s\n",App.DataPath.c_str());return 2;}
        if(!VFJson::Parser(std::move(OwnerJsonText)).Parse(App.OwnerJsonRoot,Error))
        {std::fprintf(stderr,"Owner data load failed: %s\n",Error.c_str());return 2;}
    }
    if(!StandaloneOwnerLoader::Resolve(App.OwnerJsonRoot,*App.World,Error))
    {std::fprintf(stderr,"Owner data load failed: %s\n",Error.c_str());return 2;}
    std::printf("[ColdTimeline] stage=owner_data_load end_ms=%.3f\n",
        std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count());
    App.Seed=App.World->Seed;
    if(App.bParityOnly)
    {
        VoxelGenLOD::InitializeConsoleSwitches();
        return App.PrintParity()==0?0:1;
    }
    App.Position={static_cast<float>(App.World->ViewerStartVoxels.X*0.25),
        static_cast<float>(App.World->ViewerStartVoxels.Y*0.25),static_cast<float>(App.World->ViewerStartVoxels.Z*0.25)};
    if(App.bHasCameraOverride){App.Position=App.CameraPositionOverride;App.Yaw=App.CameraYawOverride;App.Pitch=App.CameraPitchOverride;}
    const int32 ExportClipRadius=App.World->ClipRadius;
    const int32 ExportMaxClipLevel=App.World->MaxClipLevel;
    App.World->ClipRadius=App.ClipRadius;
    App.World->Settings->ClipRadius=App.ClipRadius;
    if(App.bRoundTunnels&&!PrepareStandaloneDensityWorld(*App.World,App.Seed,
        App.bRoundTunnels,App.bRoundTunnelEnhanced,Error))
    {std::fprintf(stderr,"Standalone world setup failed: %s\n",Error.c_str());return 2;}
    if(MaxClipLevelOverride>=0)
    {
        App.MaxClipLevelOverride=MaxClipLevelOverride;
        App.World->MaxClipLevel=MaxClipLevelOverride;
        App.World->Settings->MaxClipLevel=MaxClipLevelOverride;
        std::printf("Standalone max clip level override: exported=%d -> max_level=%d.\n",
            ExportMaxClipLevel,MaxClipLevelOverride);
    }
    const bool bNeedFarShapeWorld=bStreamBenchmark
        ?(StreamBenchmarkMaxLevel<0||StreamBenchmarkMaxLevel>=App.World->FullResClipLevels)
        :(App.bShapeCompare?ShapeCompareLevel>=App.World->FullResClipLevels
          :(BenchmarkLevel>=App.World->FullResClipLevels));
    if(App.bShapeWorld&&(App.bShapeAuditOnly||bGridStairAudit||bTilePathHashAudit||App.bShapeCompare||bStreamBenchmark||BenchmarkLevel>=0)
        &&!bStreamPipeline
        &&!App.BuildShapeWorldForCamera(Error,bNeedFarShapeWorld))
    {std::fprintf(stderr,"Shape interval build failed: %s\n",Error.c_str());return 2;}
    if(bStreamPipeline)
    {
        App.Snapshot=VoxelForgeCore::MakeStandaloneSnapshotFromParsedJson(App.OwnerJsonRoot,
            App.Seed,App.ClipRadius,App.MaxClipLevelOverride,App.bRoundTunnels,
            App.bRoundTunnelEnhanced,Error);
        if(!App.Snapshot)
        {std::fprintf(stderr,"Shape interval build failed: %s\n",Error.c_str());return 2;}
    }
    if(App.bShapeAuditOnly)
    {
        if(bPassageDetailAuditOnly)
            return App.ShapeWorld->auditSurfaceFamilies(true)?0:8;
        const bool bSurfacePassing=App.ShapeWorld->auditViewSurface(
            {App.Position.X*4.0,App.Position.Y*4.0,App.Position.Z*4.0},
            App.Yaw,App.Pitch);
        return App.ShapeWorld->auditDetailPassing()&&bSurfacePassing?0:7;
    }
    if(bGridStairAudit)
        return App.ShapeWorld->auditGridStairIntegrity()?0:9;
    if(bTilePathHashAudit)
        return RunTilePathHashAudit(*App.ShapeWorld,App.Snapshot,App.CoreScheduler,
            FVector(App.Position.X*4.0,App.Position.Y*4.0,App.Position.Z*4.0),
            App.WorkerCount,App.ShapeDetailVariant,App.bRoundTunnels,App.bRoundTunnelEnhanced);
    if(App.bShapeCompare)
        return RunOwnerShapeComparison(*App.World,*App.ShapeWorld,App.Position,App.Yaw,App.Pitch,App.WorkerCount,
            ShapeCompareLevel,ShapeCompareTileLimit,ShapeCompareArea,bCheckOwnerReliefBands);
    if(bRoundWallCompare)
        return RunRoundWallComparison(*App.World,App.Position,RoundCompareTileCount);
    if(bStreamBenchmark)
    {
        if(bStreamPipeline)
            return RunPipelinedStreamingBenchmark(App.World,App.Snapshot,
                FVector(App.Position.X*4.0,App.Position.Y*4.0,App.Position.Z*4.0),
                App.WorkerCount,StreamBenchmarkMaxLevel,App.ShapeDetailVariant,App.CoreScheduler,
                bStreamHash||!BakeMeshCachePath.empty(),bSkipUvAndColors,bSkipMaterialGroups,
                bSkipUV,bSkipColors,ColdProcessStart);
        if(bEditBenchmark)
            return RunEditTransactionBenchmark(*App.World,*App.ShapeWorld,
                FVector(App.Position.X*4.0,App.Position.Y*4.0,App.Position.Z*4.0),App.WorkerCount);
        const double FullDetailWidth=static_cast<double>(2*App.ClipRadius+2)
            *CHUNK_SIZE*VOXEL_SIZE/100.0;
        std::printf("Standalone clip override: exported radius=%d (112 m cube) -> radius=%d (%.0f m LOD0 cube), voxel size=25 cm.\n",
            ExportClipRadius,App.ClipRadius,FullDetailWidth);
        if(!App.bForceGenerate&&!App.bShapeWorld)App.LoadMeshCache();
        return RunStreamingBenchmark(*App.World,App.ShapeWorld.get(),
            FVector(App.Position.X*4.0,App.Position.Y*4.0,App.Position.Z*4.0),
            App.WorkerCount,StreamBenchmarkMaxLevel,App.bRoundTunnels,
            App.bRoundTunnelEnhanced,(!App.bShapeWorld&&App.MeshCache.IsLoaded())?&App.MeshCache:nullptr,
            bStreamHash,bSkipUvAndColors,bSkipMaterialGroups,bSkipUV,bSkipColors,bShardAudit,
            bFarSurvivalAudit,bFarAuditBadShell,App.Yaw,App.Pitch,
            BakeMeshCachePath,App.MakeMeshCacheMetadata());
    }
    if(BenchmarkLevel>=0)
        return RunTileBenchmark(*App.World,App.Position,BenchmarkLevel,BenchmarkCount,
            App.bProfileAttribution,App.bCompareEvaluators,App.bDensityHash,App.ShapeWorld.get());
    App.World.reset();

    WNDCLASSA Class{};Class.lpfnWndProc=WindowProc;Class.hInstance=GetModuleHandleA(nullptr);Class.lpszClassName="VoxelForgeStandaloneWorld";
    Class.hCursor=LoadCursor(nullptr,IDC_ARROW);RegisterClassA(&Class);
    RECT Rect{0,0,App.Width,App.Height};AdjustWindowRect(&Rect,WS_OVERLAPPEDWINDOW,FALSE);
    App.Window=CreateWindowA(Class.lpszClassName,"VoxelForge standalone world",WS_OVERLAPPEDWINDOW,CW_USEDEFAULT,CW_USEDEFAULT,
        Rect.right-Rect.left,Rect.bottom-Rect.top,nullptr,nullptr,Class.hInstance,nullptr);
    GApp=&App;
    const auto DeviceInitStart=std::chrono::steady_clock::now();
    std::printf("[ColdTimeline] stage=viewer_device_init begin_ms=%.3f\n",
        std::chrono::duration<double,std::milli>(DeviceInitStart-ColdProcessStart).count());
    if(!App.Window||!App.InitializeDevice()){MessageBoxA(nullptr,"Direct3D 11 initialization failed.","VoxelForge standalone",MB_ICONERROR);return 3;}
    std::printf("[ColdTimeline] stage=viewer_device_init end_ms=%.3f duration_ms=%.3f\n",
        std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-ColdProcessStart).count(),
        std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-DeviceInitStart).count());
    if(!App.bGpuFarWarmupOnly)
    {
        ShowWindow(App.Window,SW_SHOW);UpdateWindow(App.Window);
        std::printf("Controls: hold right mouse to look; WASD move; Space/Ctrl (E/Q) up/down; Shift faster; wheel changes speed; F wireframe; G toggles legacy/detailed round walls; H toggles CPU/GPU far tiles; 1-9 toggle LOD; 0 toggle all levels; N/P change seed; T swaps base/round tunnels; V swaps owner density/shape intervals; Esc quit.\n");
        std::printf("Owner data: %s; starting at first-strate open-space sample (%.2f, %.2f, %.2f) m.\n",
            App.DataPath.c_str(),App.Position.X,App.Position.Y,App.Position.Z);
    }
    std::fflush(stdout);
    if(App.bGpuFarWarmupOnly)
    {
        const auto WarmupOnlyStart=std::chrono::steady_clock::now();
        std::string WarmupError;
        if(!VoxelForgeGpuFarViewer::QueueFarShaderWarmup(App.Device,App.Context,App.GpuFarShaderPath,
            true,App.GpuWarmupEntryFilter,WarmupError))
        {std::fprintf(stderr,"[GpuShaderWarmupError] queue: %s\n",WarmupError.c_str());return 5;}
        if(!VoxelForgeGpuFarViewer::CompleteFarShaderWarmup(App.Device,App.Context,
            App.GpuFarShaderPath,WarmupError))
        {std::fprintf(stderr,"[GpuShaderWarmupError] completion: %s\n",WarmupError.c_str());return 5;}
        uint64 WarmupDxgiUsage=0,WarmupDxgiBudget=0;
        if(VoxelForgeGpuFarViewer::QueryFarGpuMemory(App.Device,App.Context,App.GpuFarShaderPath,
            WarmupDxgiUsage,WarmupDxgiBudget,WarmupError))
            std::printf("[GpuFarMemory] dxgi_local_sampled_peak_mb=%.3f budget_mb=%.3f\n",
                static_cast<double>(WarmupDxgiUsage)/1048576.0,
                static_cast<double>(WarmupDxgiBudget)/1048576.0);
        else
            std::printf("[GpuFarMemory] dxgi_local_sampled_peak_mb=unavailable reason=%s\n",
                WarmupError.c_str());
        std::printf("[GpuShaderWarmupOnlyComplete] entry=%s total_ms=%.3f\n",
            App.GpuWarmupEntryFilter.empty()?"all":App.GpuWarmupEntryFilter.c_str(),
            std::chrono::duration<double,std::milli>(std::chrono::steady_clock::now()-WarmupOnlyStart).count());
        return 0;
    }
    App.StartGpuShaderWarmupAsync();
    if(App.bGpuFarWarmupBeforeWorkers)
    {
        if(App.GpuShaderWarmupThread.joinable())App.GpuShaderWarmupThread.join();
        if(!App.bGpuShaderWarmupQueueSucceeded)
        {std::fprintf(stderr,"[GpuShaderWarmupError] queue: %s\n",App.GpuShaderWarmupQueueError.c_str());return 5;}
        std::string WarmupError;
        if(!VoxelForgeGpuFarViewer::CompleteFarShaderWarmup(App.Device,App.Context,
            App.GpuFarShaderPath,WarmupError))
        {std::fprintf(stderr,"[GpuShaderWarmupError] completion: %s\n",WarmupError.c_str());return 5;}
        std::printf("[GpuShaderWarmupBeforeWorkers] completed_before_generation_workers=true\n");
    }
    App.StartGeneration();
    MSG Message{};
    while(Message.message!=WM_QUIT)
    {
        if(PeekMessage(&Message,nullptr,0,0,PM_REMOVE)){TranslateMessage(&Message);DispatchMessage(&Message);}
        else App.Draw();
    }
    GApp=nullptr;
    return 0;
}