A character answers a question nobody asked them. A gun left in a car is back in a pocket. A promise gets treated as settled although nobody kept it. An injury vanishes for six pages and then matters again.
Each of those reads fine in the scene you are writing. They break against everything that came before, which is why a read-through finds them late, and why a continuity check has to work by putting the line that establishes a fact beside the line that breaks it.
A ledger is the cheapest way to hold that state yourself.
One row per state change
Do not summarize the screenplay scene by scene. Record only facts whose state changes, or whose persistence matters later.
Nine fields per row:
- Scene identity — a stable reference, not just the current scene number.
- Entity — character, prop, location, relationship, promise, injury, clue.
- State before — what is true entering the scene.
- Event — what changes, or appears to change.
- State after — what is true leaving the scene.
- Knows — characters who know the fact.
- Believes — characters whose belief differs from reality.
- Audience knows — whether the viewer has been shown it.
- Must persist until — the scene that can legitimately clear the state.
Knows, Believes, and Audience knows are the columns people skip. Most of the mistakes worth catching live in those three.
Knowledge and truth are different columns
Take a three-scene sequence:
Scene A. Lena hides a passport in the piano. The audience sees it. Omar does not.
Scene B. Omar tells Lena the passport was destroyed. He believes that. Lena knows it is false.
Scene C. Omar opens the piano and finds it.
A ledger that records only where the passport is has captured the least interesting fact in the sequence. Four things are true at once, and they disagree:
- the passport is in the piano;
- Lena knows the passport is in the piano;
- Omar believes the passport was destroyed;
- the audience has seen where it is.
Scene C pays off because three of those four disagree.
Which facts get a row
Keep the bar high enough that you will actually maintain the thing. Give a fact a row when at least one of these is true:
- another scene depends on it;
- two characters can hold different versions of it;
- the audience's knowledge differs from a character's;
- it persists across a meaningful time or location change;
- forgetting it would create a visible contradiction or a broken plant and payoff.
Set dressing does not need a row. A prop somebody will reach for thirty pages later does.
Use it when a scene moves
If the passport reveal moves from scene 38 to scene 21, every row whose knowledge state depended on the old position is suspect, including scenes whose text you never touched. The affected surface is larger than the edited surface, which is the case for comparing drafts by scene rather than by line.
This is also why it is worth working somewhere that can tell you what actually changed. Semantic changes compare the draft by scene and speech rather than by line, a save point gives you a fixed version to compare against, and a draft line lets you move the reveal in a parallel version while the original stays intact.
Where the software helps
Plot does not maintain this ledger for you, and you should be suspicious of any tool that claims to derive one automatically from a first draft. Belief and audience knowledge are authorial decisions, not properties a parser can read off the page — which is also why an AI rewrite can read clean and still spend a reveal early.
What Continuity Check does do is read the finished draft for the mechanical half — timeline, character knowledge, props, wardrobe, injuries, locations, names — and flag contradictions with both the establishing line and the breaking line cited. Most of its findings are review-only, with the evidence attached so you can decide what the fix is.
Put the nine columns across the top of a spreadsheet and add rows as you draft. The ledger stays yours; what Plot can tell you is which of those rows just became suspect the moment you moved the scene.