Performance and mass stress baseline

✅ Functional result Verified 👁 Timings Observed

0.0.26 — performance

37 / 0 / 0 PASS. On the tested RX 550 / Unity runtime:

• 20k reflected tiles_map field reads: ~2.960 ms (~6.76M ops/s);
• 20k reflected tiles_list reads: ~2.936 ms (~6.81M ops/s);
• 20k Array.GetValue(x,y): ~1.226 ms (~16.3M ops/s);
• 5k reflected MapBox.GetTile: ~3.692 ms (~1.35M ops/s);
• 10k sample-state reads: ~11.562 ms;
• full rich 65,536-tile pass: ~113.981 ms;
• sampled list 32,768: ~26.282 ms.

The measured GC.GetTotalMemory(false) delta was 1,392,640 bytes. That is not a leak verdict. These timings are machine-specific observations. The useful design result is that a rich 65k reflection pass is large enough that it must not be placed in a per-frame wait loop.

0.0.27 — mass stress

44 / 0 / 0 PASS. Work was chunked at 4096 operations and the CPU stopwatch excluded frame-yield waits.

• 524,288 complete tile visits: typed tile, x/y, bounds, height, terrain, zone/chunk/region;
• 250,000 deterministic coordinate lookups;
• ~50,000 reflected GetTile cross-checks inside that lookup run;
• 100,000 neighbour traversals;
• 12 registry sweeps each for kingdoms/cities/actors; the final test world had 0/0/4 respectively;
• final collection refs/counts and sample state remained stable.

The functional conclusion is stronger than any ops/s number: these read paths survived high-volume use without detected map corruption.

Evidence: /evidence/wbml-0026-result.txt, /evidence/wbml-0027-result.txt.