Detect knowledge continuity errors while preserving intentional mystery, inference, bluffing, and dramatic irony.
A character knowing something surprising is not automatically an error. The audit asks whether there is a plausible authored acquisition or inference path.
Evidence Hierarchy
Prefer, in order:
- explicit information-state model
- authored scenes/events
- quest / branch state
- dialogue implication
- reasonable inference from known facts
Do not invent off-screen explanations merely to excuse a contradiction.
Audit Categories
Impossible knowledge
Actor states or acts on a fact without an established path.
Premature certainty
Actor has evidence for suspicion but speaks/acts as if knowledge is certain.
Forgotten knowledge
Actor later behaves as if a previously acquired fact is unknown.
Optional-path leakage
Content assumes a clue or revelation from an optional branch happened for everyone.
Player/avatar mismatch
Choice wording or action requires player knowledge the avatar lacks, or vice versa, without intentional separation.
Source mismatch
Actor knows the fact but attributes it to a source they could not know.
Propagation jump
One actor learns something and unrelated actors immediately know it without communication.
Temporal leak
A line references an event or revelation before it occurs.
Procedure
- Extract important factual claims from the target material.
- Identify the speaker/actor and point in chronology.
- Trace backwards for acquisition, communication, or inference.
- Classify the strongest justified state: unknown / aware / suspects / believes / knows.
- Compare justified state with authored behavior.
- Check alternate/optional paths.
- Report only material discrepancies.
Severity
- Critical — breaks mystery, major plot logic, or player choice.
- High — clear impossible knowledge affecting scene/quest behavior.
- Medium — certainty or propagation is weakly justified.
- Low — wording implies slightly more knowledge than evidence supports.
Confidence
Report separately:
- High — explicit chronology/state proves mismatch.
- Medium — likely mismatch but off-screen acquisition may be intended.
- Low — sparse material; requires author clarification.
Output
Summary
Number and distribution of findings.
Findings
For each:
Location
Actor / player
Used knowledge
Highest justified state
Missing link
Severity / confidence
Why it matters
Smallest repair category
- add acquisition beat
- lower certainty
- gate line/content
- add communication
- preserve remembered state
- clarify player/avatar separation
- mark as intentional bluff/inference
Information-flow gaps
Facts with no viable acquisition path.
Intentional asymmetries worth preserving
Do not treat these as defects.
Do Not Rewrite by Default
The audit identifies the logic problem and smallest repair category. Rewrite only when requested.
Final Quality Check
Do not flag a character for correctly inferring something merely because nobody explicitly told them. Distinguish inference, belief, and justified knowledge.