Corpses do not survive save/continue (deferred known issue, ex BUG-018)

[aaronstatic]

Summary

Killed characters (enemies, recruits, players) do not survive save → quit → Continue. The corpse and its lootable inventory are simply absent after continuing. This was a new capability attempted during the vanilla-persistence migration (EPF never persisted corpses either, so it is not a regression), but after two fix attempts it still fails the real quit-and-continue play-test, so we're deferring it past 1.4.0 and documenting it here for visibility. Contributions welcome.

Repro: kill an armed enemy, do NOT loot the body, save, quit to menu, Continue → body is gone.

What has been tried (internal tracker: BUG-018)

Attempt 1 — scripted persistence config rule (Phase 4 of the persistence migration). A script-defined PersistenceConfigRule (OVT_DeadCharacterPersistenceConfigRule) bound in Configs/Systems/Persistence/Overthrow.conf (Priority 36000, SelfSpawn 1, Character collection, vanilla's six character serializers), plus a death-time re-match via ReloadConfig(victim).

Result: instrumentation proved the engine never calls a script-defined rule's IsMatch — zero calls across world load, initial tracking of every character, and 300 forced ReloadConfig() re-matches. Every rule vanilla binds in a .conf is a native generated class; scripted rules are effectively dead code.

Attempt 2 — dynamic config flip via the API vanilla actually exposes. On the character-killed event, OVT_PersistenceTracking.MarkForSelfSpawn(victim) does GetConfig() → m_bSelfSpawn = true → SetConfig(), keeping the matched config's collection/serializers and flipping only that bit. Applied on the death frame. An automated Init-tier test (OVT_TEST_Init_Persistence_DeadCharacterConfigSelfSpawns) kills a spawned recruit and asserts the matched config self-spawns — it was authored red against the live defect and went green with the fix.

Result: the test is still green, but the 2026-08-04 single-player play-test shows corpses still do not survive an actual quit → Continue (tested enemies, recruits and player corpses).

Where the mystery now sits

The config demonstrably has SelfSpawn set after death (asserted in-session), yet nothing comes back through a real restart. So the failure is somewhere downstream that our test harness cannot reach — the true quit-and-continue path restarts the world, which the autotest harness cannot follow (documented limitation). Untested hypotheses from the tracker:

1. Record lifetime / SelfDelete semantics — if the engine drops a tracked instance's record when the body despawns or cleans up before/at save time, there is nothing to restore.
2. Restore-side behavior — a self-spawned character record may not rebuild a dead character (life state may not be part of what the six vanilla character serializers capture/apply, or the entity spawns alive and is immediately cleaned up).
3. Save-time filtering — dead characters may be excluded from the save pass entirely regardless of their config's SelfSpawn flag.

Pointers for anyone picking this up

• Internal tracker doc: docs/bugs/BUG-018.md (full history, instrumentation numbers)
• Death-time config flip: OVT_PersistenceTracking.MarkForSelfSpawn (called from OVT_OverthrowGameMode's character-killed subscription)
• In-session coverage: OVT_TEST_Init_Persistence_DeadCharacterConfigSelfSpawns (Init suite; green — the gap is the restart path, which only manual play-testing exercises)
• Vanilla persistence API notes: docs/features/core/persistence/vanilla-api-reference.md

Any fix should keep the existing Init test green and add whatever in-session assertions are possible for the newly-discovered failure layer, proven able to fail first (project testing rule: no retries, every case demonstrated red before shipping).