North star

Our mission

Predict the real-world reception of a film before the world sees it — provably, auditably, and without the film ever leaving its owner’s hands.

We are building the first reception-prediction system with a published, leakage-free accuracy record. Not a report generator. Not a focus-group substitute. An instrument whose every claim can be traced to a measurement it had no way to game.

Integrity is everything

Integrity is not a value alongside the product. It is the product.

A reception prediction is worth exactly as much as the buyer’s ability to trust how it was validated — nothing more. Every architectural decision, every analysis, every claim inherits from this.

  1. The holdout wall is absolute

    Films reserved for validation are never trained on, never used to tune prompts, never peeked at and then adjusted. They answer one question honestly: does this work on films the system never learned from?

  2. Public ratings never calibrate the model

    Organic public scores are the validation showcase — proof that panel reception tracks the real world. They are never the target the system is fitted to.

  3. Every ground truth is construct-checked

    No number is trusted because it is official. Review bombing, fan stuffing, invited audiences, trailer tests, and incumbent studio scores are screened for whether they actually measure natural-audience reception.

  4. Audience facts are declared, never inferred

    Segment properties — faith-market films, client-invited audiences — are stated at intake. Effects are never discovered in the data and then circularly fit.

  5. Honest claims from honest conditions

    The prospect-facing accuracy number comes from films the model could not have memorized — because that is the prospect’s actual condition. If accuracy only held on recognized titles, we would say so.

  6. The track record is the moat

    Anyone can prompt a frontier model at footage. Nobody else has tens of thousands of calibrated human survey responses, moment-level sentiment events, and a validation methodology a skeptical research department can audit end to end. Protect the corpus. Publish the record. Never shade a number.

The thesis

Taste-matched panels measure the same thing the world will.

iScreeningRoom panels are taste-matched audiences — people selected to match who would consume the film in the wild. Panel reception and organic public reception therefore measure the same construct: how a film lands with its natural audience.

That gives a two-link chain. Each link is independently measured. Neither is ever fit to the other.

AI agent

Blind-validated on films it has never seen

Panel actuals

Taste-matched audience reception

Real-world reception

Organic public ratings, bridged by study

The chain composes into the prospect claim: we can tell you how the world will receive your unreleased film — without any step having peeked at the answer.

Who it’s for

Different tiers are different products — priced on the decisions they inform.

01

Indie proving ground

Self-serve analysis for independent filmmakers. This tier is never abandoned: it democratizes testing and feeds the validated track record the premium tiers sell on. Every film scored here grows the honest-condition record.

02

Decision support

Per-title and slate-level prediction for distributors deciding acquisitions, sales agents setting asks, streamers filling slates, and film funds evaluating investments. Acquisition-grade memos with uncertainty bands — not gut, not a one-night screening.

03

The studio play

Every human test is a leak surface. An evaluation where the footage never meets a human eyeball outside the building deletes the piracy category — and changes the economics of iteration: every cut, same day, at every stage of post, at no marginal exposure. Incumbents assert authority. We publish ours.

What success looks like

Not an oracle. A reliable instrument with an auditable record.

Success is reliable rank placement. Absolute prediction within roughly half a star on mainstream films. Segment-flagged predictions where audience utility differs. Honest uncertainty bands. A published record a skeptic can audit.

We don’t skip validation. We don’t shade numbers. We don’t peek at holdouts. We don’t crown a studio benchmark that hasn’t survived construct-checking. That is the whole point of this page.

The accuracy record is the company. Integrity is everything.