predicted
actual
EP 12 — "The Claw Always Wins"
Retaincast analyzes a cut, predicts where viewers drop and why, publishes everywhere, then pulls the real curve from YouTube — and scores its own prediction against it.
Free tier included · no credit card · one invite email, nothing else.
Every pass through the loop makes the next prediction sharper, because the misses are measured, not forgotten.
ffmpeg reads the hook window frame by frame, the cut rhythm, dead air and loudness; Claude judges it against a retention rubric. You get a predicted curve with a reason for every drop, and the five timestamped edits that matter most — while changes are still cheap.
retaincast analyze ep1.mp4 --kind short
One upload goes to YouTube, TikTok and Instagram Reels from your own dashboard — scheduled or on demand, no tab-juggling.
Every night Retaincast pulls views, watch time and the full audience-retention curve for each published video. No CSV exports, ever.
Predicted and actual curves are compared point by point. The error becomes a channel prior — "your audience drops at logo stings; predictions ran 5 points optimistic" — that corrects every future analysis.
retaincast calibrate
Not vibes — measured signals, judged against what makes people stay.
Dense frame sampling across the hook window: does the opening image, text and sound make a promise worth staying for?
Shot lengths, silence spans and loudness — the mechanical signatures of pace. One command exports the dead-air cuts to your editor.
Where the video re-earns attention, and where it coasts long enough to lose it.
The same analyst reads short-form scripts, screenplays, TV pilots and books — fixing engagement at the cheapest possible stage.
Every other tool reads your audience retention after you publish — after the video has already spent its first-48-hours push. Retaincast works before, then logs every prediction and scores it against the curve YouTube reports: in the open, per video, in points.
The longer you use it, the harder it is to replace: the calibration record is yours, built from your audience.
| Checkpoint | Predicted | Actual |
|---|---|---|
| @ 5% | 78% | 74% |
| @ 50% | 46% | 43% |
| @ 90% | 31% | 33% |
| Mean error | 4.2 pts |
Join from the waitlist and your price is locked for life.
Prove it on your own videos
The full loop, every upload
Series, teams and long scripts
For reference: vidIQ runs $19–49/mo and TubeBuddy $12–27/mo for after-the-fact analytics. Retaincast is the only one working before you publish — and grading itself after.
Retaincast measures what drives retention — the hook window frame by frame, cut rhythm, dead air, loudness, the transcript — and judges it against a retention rubric to produce a predicted curve with a reason for every drop. Then it checks itself: once the video is live, the real YouTube retention graph is compared point by point. Typical error after a few calibrated videos: under 10 points per checkpoint, and falling.
They analyze what already happened. By the time the retention graph appears in YouTube Studio, the algorithm has judged your video. Retaincast moves the same analysis to before you publish, while the fix costs one edit instead of one video.
Not a score — a work order: the five timestamped edits that matter most, dead-air cuts exported straight to Resolve, Premiere or Final Cut, and burned-in captions for muted feed viewers.
Yes. The same engine reads short-form scripts, video scripts, screenplays, TV pilots and manuscripts — predicting where readers or viewers quit, before anything is shot or typeset.
Free tier included · no credit card · one invite email, nothing else.