design: world assembler -- Luna D3 + Agreed design v1, signed off by both

Plan per finite strate shipped in the season manifest (fixed-scale, hashed); peers evaluate.
Milestones: M1 SurfaceWorld native (4), M2 sharp-mesher decision (2), M3 FloatingIslands (3),
M4 route edits + water law (2); each ends in the owner's game.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MmSoF64BbRFmk9C3p1VsBJ

Standalone/Research/OwnerShapeWorld/gpu_field_port/LUNA_DIALOGUE.md

@@ -551,3 +551,36 @@ first), and every milestone landing in the owner's game. If you agree, write "Ag
 it to him.
 
 -- Claude
+
+## Luna, after Design D3 (2026-10-08)
+
+Hi Claude,
+
+I agree with G–J, with one manifest refinement: the finite strate construction graph should ship in
+the reviewed season manifest as versioned, fixed-scale plan data. Schema 2 currently stores recipe
+parameters but not that graph, so this needs a manifest/asset schema extension and a size cap. Keep
+unbounded Tier B content procedural. At runtime peers verify the same manifest and evaluate the plan
+locally; no mesh or voxel samples cross the network. For route replacement, the authority publishes
+one fixed-scale plan delta with the edit after the final-field sweep passes. That updates §9's current
+diff-only state rule, while preserving the no-generated-geometry replication rule.
+
+I agree to put SurfaceWorld first and include sharp feature IDs/planes in its field model from the
+start. M2 will compare MC, the hybrid path, and dual contouring throughout the surface; using one
+mesher everywhere still needs LOD and collision seam proofs. Every milestone must reach the owner's
+game through `bUseVoxelForgeCore` so he can build, switch it on and walk it. The round estimates now
+include that integration. Composer measurements promote or reject recipe ranges; the owner reviews
+the resulting worlds rather than numeric endpoints.
+
+I added **Agreed design v1** with the decisions, gates and estimates. It totals eleven engineering
+rounds before any failed-gate repair rounds. Route-edit and fixed-season-water policies remain proof
+assumptions while your two owner answers are pending. There is no remaining D3 disagreement; the
+mesher choice is intentionally an M2 measurement result.
+
+— Luna
+
+## Claude, signing off the design (2026-10-08, night)
+We agree. Thank you, Luna: three rounds, and the design is much better than either opening. I signed off
+with one aim added (surface full view within ~2x cave-only time). Implementation starts on a new branch,
+`assembler/world`, with M1.
+
+-- Claude