How it works

One script in. Four movements of coverage out, a printed PDF, and a link you can hand to anyone. Here is every step, including the parts most tools leave out.

The script is deleted the moment the report is assembled. Not after a retention window, not archived first — the same request that writes the report blanks the text. The PDF and the share link live only as long as the report does.

01What we can and cannot see

We cannot read your script, and neither can anyone who steals our database. Your upload is encrypted before it is stored, the LLM reads it, and the same request that saves your report destroys the script. An hour later the report and the PDF go too. There is no log, no file, no admin screen, and nothing left to leak.

One honest limit: the machine that does the work can read your script while it is working on it. That machine is ours, in our building — the LLM, the PDF, the portraits, all of it. No provider sees your pages, because no provider is involved.

02The seven axes

Each scored out of ten, and each has to name a scene.

• The Protagonist — who we follow, and whether they are worth following.
• Story Engine — what drives it, and whether the pressure holds.
• The Voices — who sounds like a person and who sounds like a script.
• Freshness — what is new here, honestly.
• Pull — whether it holds you, and why.
• Point — what it is actually about.
• The Whole — the aggregate.

03What it will not do

Every tool has a floor. Here is ours, stated up front.

• It will not always be right. We run a verification pass that reads the report back against the script and flags anywhere it contradicts the source. It catches a good deal. It is not infallible, and a confident wrong sentence is the failure mode you should expect before any other.
• It will not replace a reader who knows your market. It has read your screenplay. It has not read the room.
• It will not invent a verdict for you. The score is an opinion with reasons attached, not a measurement. Ask two readers and you will get two numbers.
• It is not an authority. It is a tool that read your pages and wrote down what it noticed. That is all it is. Nothing here is a ruling, a grade, or a permission slip — the verdict is a reader's opinion, and the only verdict that matters comes from a person who hires, buys, or loves the thing you made.
• It is not your editor, and it is not your taste. Disagree with it. A report that never makes you argue is a report you did not need. If it tells you something you already believed, that is not confirmation — that is the cheapest thing this kind of software does.
• It will not make you a better writer. Only writing, and time, and reading people better than you, do that. This is one reader's notes, on one draft, once. Use it that way and it is useful; use it as a mirror for how you feel about your work and it will tell you what you want to hear.

04How often to use this

Not often. We would rather you did not.

A draft is not finished because a tool said so. It is finished because you decided it was, and that decision needs time you cannot outsource — the sitting-with-it, the reading it aloud, the part where nothing happens for a week and then one thing clicks. Feedback you receive faster than you can act on it stops being feedback and becomes noise you feel obliged to answer.

So we suggest running a script here on major revisions only — a draft you have genuinely reworked, not each pass through the pages. If you find yourself uploading nightly, that is usually a sign the tool has replaced your judgment rather than supported it. That is the failure mode we would most like to avoid, because it is the one that quietly makes writers worse.

We are not going to rate-limit you or hide the button. The restraint has to be yours, and we would rather say so plainly than pretend a design choice is a policy.

Two related things we feel strongly about. This is not a tool for writing with. It will not draft a scene, and it is not built to help you generate one. It reads what you wrote. And do not run AI writing through it. A report on a script a model produced is a model evaluating a model — a closed loop that tells you nothing about your own work and a fair amount about how quickly a tool can flatter one. If your pages came out of a chat window, this will not help you, and we would rather tell you that than take the money.

We think these models are oversold, that they flatten writing more often than they sharpen it, and that the best use of one is a small, occasional, honest second opinion — not a chair you sit in. Please do not make ours a habit.

05You can change the prompts

The prompts are the product. They are yours to edit.

Every instruction sent to the LLM is readable and editable from your account — the Data Processing pass, the notes pass, the house style, even the closing note. Change one and the next run uses it. If you drive tools from an agent, there is an MCP server: an agent can read what the pipeline sends, diff it against the default, and retune it for you without a human opening a settings screen.

Most coverage products are a black box with a price. Ours is a box you can open.

Questions about any of this? The prompts are the honest answer — you can read every word the pipeline sends.