Save/reload lifecycle and stale-reference torture

✅ Verified — 0.0.29-fix1: 26 / 0 / 0

WBML discovered and safely invoked these version-bound signatures:

SaveManager.saveToCurrentPath() -> SavedMap
SaveManager.loadWorld() -> void

No path, slot or unknown argument was guessed.

Reload proof

The accepted run saved the disposable world, changed one tile Height 30 → 31, then called loadWorld(). Reload was accepted only when all of these aligned:

• MapBox.finishingUpLoading observed;
• marker value 31 disappeared;
• tiles_map/tiles_list were live with 65,536 entries;
• six stable post-load polls completed.

_load_counter changed 1 → 2 as independent observational corroboration.

Why exact Height equality is not the contract

The original 0.0.29 run really reloaded the world, but the marker test was before=1 marker=2 after=0. The old harness required after==before and falsely failed. WorldTile.Height can normalize/recompute on load. The correct proof is marker cleared + lifecycle signal + live stable collections, not byte-for-byte Height equality.

Reference result

In the accepted run the old sample classified REUSED_CURRENT. tiles_map, tiles_list, zone, chunk, region and kingdom/city/actor manager objects were the same references after reload. The old tile survived 100/100 safe reads. This reuse is an implementation observation, not a requirement for future versions.

Kingdom/City/Actor entity-object stale-vs-reuse across reload remains UNKNOWN because those registries were empty in the torture world before the load.

Evidence: /evidence/wbml-0029-result.txt and rejected-harness evidence.