Everything one analysis gives you.
Paste a TikTok, Reel or Short. In 3 to 5 minutes you get the score, the retention curve, the weakest moment and the fix. The first 5 minutes of video every month are free.
The score, and what to change first.
A 0 to 100 score built from eight factors: hook, emotion, curiosity gap, pattern-interrupt, story pull, concreteness, share-worthiness and usefulness. Each factor comes back with what Castle saw in your clip, at the timestamp it saw it, and the one change that would lift it.
Fixes are ordered by impact, so you always know which to do first.
Score the same video twice and you'll land within about 4 points. Smaller differences are noise, and the product says so.
| Factor | Weight | Research it draws on |
|---|---|---|
| Hook | 18% | The orienting response (Itti and Koch) |
| Emotion | 16% | Berger and Milkman, 2012 |
| Curiosity gap | 15% | Loewenstein's information-gap theory |
| Pattern-interrupt | 12% | Prediction error (Schultz) |
| Story pull | 12% | Narrative transportation (Green and Brock) |
| Concreteness | 10% | Chip and Dan Heath, Made to Stick |
| Share-worthiness | 9% | Social currency (Berger, STEPPS) |
| Useful to viewers | 8% | Practical value (Berger, STEPPS) |
The full methodology, weights included, lives on How scoring works.
Where attention dies, second by second.
Castle reads the clip frame by frame, densest on the first 3 seconds where the scroll decision happens, and predicts how many viewers are still watching at every second. You see the moment your hook peaks and every drop-off, with the reason next to it.
Your video against the one you wish you'd made.
Paste yours and a competitor's. Castle scores both, shows who hooks faster, where each loses attention, and exactly three moves the weaker video should copy.
Open CompareSimulated viewers react before real ones do.
A panel of simulated viewers grounded in 1,000,000 real demographic records. You get the average, the spread, a region split, and quotes from the strongest and weakest reactions.
Simulated, not measured, and labeled that way everywhere it appears.
When the hook is the problem, Castle rewrites it.
Thirty named hook types, each tied to the psychology it triggers. When your hook scores low, Castle matches your topic to the ones that fit and writes three to five original opening lines for your video. Nothing is copied from anyone else's.
The same clip, scored through ten markets.
Pick a target market and Castle scores through that cultural lens: global, India, the US, the UK, Brazil, Indonesia, Mexico, Japan, Germany or the UAE. The lens shapes the estimate; Castle never invents region-specific content it can't see.
Also tells you when your clip matches a public trend Castle tracks, with the reason taken from your own transcript. Most clips honestly match nothing.
In beta: from fix to live campaign.
Launch on Meta in one click
BetaPush the winning creative to Meta as a paused campaign. During beta we run the setup with you. Agency and Enterprise plans.
Generate the fixed version
BetaProduce a clip and on-brand copy so you can ship the fix the same day. Rolling out to paid plans.
Want early access? Email support@yantrava.com and we will set you up.
One analysis returns all of it.
Start with your weakest video. The report tells you what to change first.