Why
Writing is the most expensive way to discover that a scene does not work.
By the time a scene exists as pages, you have made a hundred small decisions inside it — a joke you like, a line that finally cracked the character, a piece of staging that took an hour. Those decisions are sunk cost, and sunk cost is the enemy of structural judgement. The scene now argues for its own existence in a voice you cannot easily ignore, because it is your voice.
A contract moves the decision earlier, to the point where it costs four sentences instead of five pages. Four questions, answered before any prose exists:
- Objective. What does the POV character want in this scene? Not their want for the film — something they could get or fail to get before the scene ends.
- Obstacle. What stands in the way here? If the honest answer is "nothing yet", this is setup, and setup is not a scene.
- Tactic. How do they try? This is where character lives. Two people with the same objective and different tactics are two different people; if the tactic would work identically for any character, you have not written a person.
- Turn. What changes? If nothing changes, the scene has no reason to exist, and no amount of dialogue polish will give it one.
The four questions are not a form. They are a filter, and the filter's whole value is the scenes it rejects.
There is a second, quieter benefit. A scene with an approved contract is a scene you can hand to a collaborator — or to a model — with confidence, because the thing that makes it good is written down. Without a contract, the only specification is your intention, and intentions do not survive being read by anyone else.
How
- Write the contract for every scene in a sequence at once, before writing any of them. Judging ten scenes side by side surfaces redundancy that judging one at a time never will.
- Answer all four questions in one sentence each. If the objective needs a paragraph, the scene is doing two jobs and should be two scenes — or one.
- Set the entry and exit charge as integers. If they match, the scene has no turn: cut it, merge it, or find the turn you have not admitted to yet.
- Ask the deletion question directly: if this scene vanished, what would the audience fail to understand or fail to feel? "They'd miss a good joke" is not an answer. "They would not know the brother is lying" is.
- For every scene that survives, name the one thing it does that no other scene does. If two scenes have the same answer, one of them is already redundant and you have found it for free.
- Only then write pages. Treat the contract as the spec: if the draft drifts from it, either the draft is wrong or the contract was — decide which, deliberately, rather than letting the pages quietly rewrite the plan.
- Re-run the deletion question after any structural change. A scene that was load-bearing before the midpoint moved may not be load-bearing after.
Notes for AI film
This card matters more here than in live action, and the reason is arithmetic.
A scene that survives to the shot list becomes eight to fifteen generations, each with its own prompt tuning, its own drift risk, and its own cost. Cutting a scene at the contract stage costs four sentences. Cutting it after generation costs a day and a bill.
The discipline also produces something live-action writers do not get: a machine-readable statement of scene purpose that every downstream agent can read. The Cinematographer plans coverage against the turn. The Editor sets duration against the charge delta. The Continuity Supervisor knows which entities must be present. All of that is downstream of four sentences written before anyone touched a page.
The gate is deliberately hard to bypass — the script editor stays locked until the contract is approved. That friction is the product, not an accident of it.