WBML failed assumptions — negative knowledge

❌ Failed assumptions ✅ Preserved as evidence

These failures belong to the harness assumptions, not automatically to WorldBox.

1. Worldgen requires a new WorldTile reference — false. Same-size worldgen repeatedly reused tile/array references.
2. Old tile must disappear from the map — false. Held tile references remained current after repeated generation on the tested size.
3. Exact WorldTile.Height restoration proves reload — false. Height can normalize/recompute during load.
4. Total AppDomain assembly count must stay constant — false. Reflection lazily loaded three framework assemblies without a mod conflict.
5. A world-owned coroutine survives world replacement — false. The runner must live outside the replaced world lifecycle.
6. on_world_loaded is authoritative for generateNewMap — false. finishMakingWorld was the useful explicit signal in this path.
7. A full 65k scan is acceptable in every wait frame — operationally false. The original harness drove FPS to roughly 6.
8. Destroyed Unity wrapper == ordinary CLR null — false model. Fake-null and delayed destruction require Unity-aware checks.

Why we publish this

Negative knowledge prevents future modders and AI assistants from rebuilding the exact bugs that consumed the research time in the first place.

Sanitized rejected-run excerpts