Why
A charge curve is not a summary of the film. It is a diagnostic, and it only works if it is drawn before the writing rather than after.
What it detects, better than reading can:
The plateau. Three or more consecutive beats within a point of each other. The film has stopped moving and no individual beat is at fault, which is why reading the script does not find it.
The unearned peak. A large swing with no preparation — the beat before it sits at the same level. Peaks need a floor.
The false ending. A return to baseline before the actual ending, which reads to an audience as the film finishing early.
The monotone. Every beat the same sign. A film entirely negative has no fall left to give.
The essential distinction, and the one most often collapsed: charge is valence, not intensity. How it feels, not how loud it is. A thriller can be maximally intense across its whole runtime and still swing between hope and dread — those are different axes, and a curve that plots loudness tells you nothing about whether the film moves.
The route a film is built on carries its own target curve. Comparing the board's actual curve against the route's target is the fastest way to see whether the structure is delivering the feeling that was asked for, and it is a comparison a person can make in about four seconds.
How
- Assign every beat a charge from -5 to +5 before writing any of them. Valence, not volume.
- Plot them in order and look at the shape rather than the numbers.
- Find any run of three beats within one point. That is a plateau — merge two of them or cut one.
- Check every peak has a floor before it. A high beat preceded by a high beat is half a beat.
- Compare against the route's target curve, decile by decile. Where the two disagree by more than two points, the structure is not serving the feeling that was chosen.
- Check the final decile matches the intended ending valence. A route that ends negative with a curve that lifts at the end is a film arguing with itself.
Notes for AI film
This is the one structural check that is genuinely cheap to automate and genuinely useful, because both curves are already data — the route carries a target, the board carries actuals, and comparing them needs no model call.
It is also the check most likely to be quietly wrong, because a model asked to assign charges will tend to produce a smooth and plausible curve rather than an honest one. Sanity-check the extremes against the beat summaries: if a beat reading -4 describes something mild, the number is decoration.