auditoria-global-consistencia-entidades

Extract event triples (subject, action, object, location, narrative time) from each chapter and load them into a temporal entity graph. Then run boolean rules checking location continuity, item possession, and vital status to flag contradictions like dead characters acting.

Vector similarity search treats contradictory facts as semantically similar, so it misses binary conflicts such as a character being dead versus alive. A temporal graph with imperative boolean rules evaluates physical state transitions deterministically.

It checks three rule families: location continuity (no teleportation without a travel edge), inventory possession (an item must be owned before use), and vital status (no active verbs from characters recorded as dead at an earlier narrative time).

Yes. Actions inside memories, dreams, or flashbacks can trigger false vital-status violations. Mitigate this by modeling scene-type and narrative-time metadata so flashback events are excluded from linear continuity checks.

Avoid running it during free-form brainstorming drafts where chapter chronology is undefined, and avoid real-time checks while typing. The global graph scan is computationally expensive and should run asynchronously or on chapter completion.