Questions.

A model — Claude, from Anthropic. We say so plainly because the way it's used is the whole point.

The model is handed one scene at a time and has to answer before it is shown the next one. It genuinely cannot react with knowledge of your ending, because it doesn't have it yet. That constraint is what makes the reaction worth something.

Because it will read the whole thing first and then tell you about scene three with full knowledge of scene forty. Every setup looks deliberate in hindsight. Nothing can surprise it. That's exactly the blind spot that makes ordinary notes unhelpful.

Cold Read holds the pages back, keeps each reader in a consistent point of view across the entire script, and anchors every reaction to a scene number so you can see where attention actually turned.

Then you've still learned something, which is where a reader's attention went. A reaction isn't an instruction. The useful signal is usually not one reader's opinion but the same reaction repeating across several very different readers.

Running the same pages past a sympathetic reader and a hostile one, and finding they both went flat at scene twelve, is worth more than either of them alone.

You advance it scene by scene, so it's up to you — a feature-length script is usually a few minutes of clicking through. Each scene's reaction comes back in a few seconds.

Yes, mid-read, on a paid read. You can ask what they think is going to happen, whether they still care about someone, or why a scene landed the way it did — and they answer knowing only what they've been shown.

They're sent to Anthropic's API to produce the read, and stored so you can pick a session back up later. They are not used to train models, not published, and not shown to other users.

The full detail of what's stored and for how long is on the privacy page.

No. Your script stays yours. Uploading it grants only the narrow permission needed to run the read you asked for and store the session for you. See the terms.

Yes. Ask from the address on your account and they'll be removed. email

Splitting works off standard sluglines — lines beginning INT. or EXT. You see the scene count before you spend anything, so if the number looks obviously wrong you'll know before it costs you.

Unusual formatting can land a scene or two off. It's been tested against real screenplay text extracted from PDFs and comes within a few scenes of a human split.

Not yet — paste only. Copying out of a PDF viewer or Final Draft works fine. PDF upload is the next thing being built.

There's an upper bound to stop enormous pastes, but a normal feature screenplay is comfortably inside it. If your paste is rejected as too long, it's far bigger than a feature.

One complete read of one script by one reader, first scene to last, regardless of length. A 30-page short and a 120-page feature both cost a single credit. Credits never expire.

Two things. Without any account, you can read the first three scenes of a script with six of the readers — no card, nothing stored. With a free account, you get one complete read of a whole script on the house, and all twenty-four readers unlock.

So you can find out whether this tells you anything useful before money is mentioned.

Neither. There's no subscription and no recurring charge. You buy a pack of credits once, and they sit there until you use them.

A feature-length script costs a real amount in model usage to read — the bulk of what you pay is that. Packs get cheaper per read at volume because the overhead doesn't scale with them.

Payment is handled entirely by Stripe. Card details never reach Cold Read's servers. Credits are added to your balance only when Stripe confirms a completed payment — never by your browser returning from a checkout page.

If a read fails, breaks part way through, or returns something obviously broken, email and the credit goes back on your balance. That's the common case and it's not an argument.

For unused credits, ask within 30 days of buying and you'll get a refund for the ones you haven't spent. Credits already spent on completed reads aren't refundable, because the cost of producing them has already been incurred — but if a completed read was genuinely useless to you, say so and we'll sort it out.

email

Because a password on a tool like this is a liability with no benefit. You enter your email, we send a link, clicking it signs you in for 30 days. Nothing to remember, nothing to leak.

Check spam first. Links expire after 15 minutes, so if it's been sitting a while just request another. If nothing arrives at all, email and it'll get sorted manually.

Yes — email from the address on the account and it goes, along with the scripts and reads attached to it. Unused credits are refundable at that point under the policy above.