The owner was right and said so plainly: "the tunnels should be a flat D with then a
bit of noise added, as it's most basic shape -- this alone should make all tunnels not
over X steepness to be walkable." Codex had reached the same conclusion from the other
direction ("the native fix is to author the inter-strate passage floor profile during
passage construction"); it had measured the current position while he described the
destination. This is that refactor.
Walkable inter-strate passages now bake their floor profile and bounded relief during
GeneratePassages. Evaluation composes a D-shaped SmoothMax floor, with final MC
ownership after the generic writers -- the same ordering fix that made the tunnel floor
native in 2260c4d.
native compositions 7,629
legacy support fires 0
floor-profile fallbacks 0
player-fit cells 20,804 -> 20,830 (+26)
walk-reachable 10,884 -> 10,909 (+25)
climb-reachable 10,911
connectivity arrival -> departure, Connected
largest component 10,909
⭐ Support-ON and support-OFF now produce IDENTICAL export geometry (hash 07C14005,
82,273 v / 153,346 t, mesh 2.981 s vs 2.960 s). The writer that previously cost
3,359 reachable cells when disabled is now inert -- the shape provides what it used to
patch. That is the ~17.43% of banded pixels from 4f8b4a9 addressed at the source.
The legacy writer stays only as a fail-closed fallback for a malformed/non-native
passage. It is no longer load-bearing.
PUSHBACK, and it is the right restraint: only passages carrying
`bWalkableTunnelContract` -- currently the authored `Straight` style -- receive the
guaranteed D-floor. Spiral, Cascading and Worm styles do NOT silently gain a
walkability promise without a separate gradient proof. Correct: a guarantee nobody
proved is worse than no guarantee.
ROUGHNESS -- the owner's report, measured properly this time and CLOSED
He confirmed twice that changing roughness type/frequency changed generation time from
startup. The earlier sweep was a single 128^3 export at LOD0 and could not have seen
it, so this one used the game path, fixed streaming centre, identical 850-tile workload:
skip rate LOD0 70.8% LOD1 57.3% LOD2 32.6% (Cellular 31.8%) LOD3 2.5% LOD4 0%
gen p95 s FBM .472/.369/.083/.087/.145 Ridged .473/.365/.084/.084/.143
Mixed .474/.358/.084/.090/.149 Cellular .479/.360/.083/.089/.161
FBM frequency 0.025-0.2: flat across every LOD
No classifier-bound collapse by type or frequency. The startup difference was
workload/streaming behaviour, not a type-specific bound defect. No unsafe bounds change
was made to chase it.
⚠️ Worth noting for later: the skip rate collapses with coarseness -- 70.8% at LOD0 to
0% at LOD4. That is expected (a LOD4 tile spans a huge volume and almost certainly
contains something) but it means the classifier does nothing for exactly the tiles that
cover the most ground.
VERIFIED BY ME: build 0 errors; export 1.418 / 1.445 / 1.458 s (4f8b4a9 was
1.353 / 1.430 / 1.432); 153,346 triangles. Peak game RSS ~2.05 GB.
Sources consulted: Epic's UE task-systems documentation for the worker/game-path
distinction, and Lorensen-Cline for the discrete isosurface context.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012aheo2crrZmHuYtPpdY2Cb
Fr0zkaandClaude Opus 5
adb34dffa9