Demo
A two-minute live run of the product: the fleet, verify, the tamper test, the families page, a card, a live re-score, and scan your own. Real clicks on the deployed Space, recorded end to end.
The original 1080p cut with burned captions, served directly at /demo.mp4; subtitles at /demo.srt.
| time | scene | what happens |
|---|---|---|
| 0:00 | Open | The live fleet: twelve checkpoints, verdicts, validity, compliance. |
| 0:13 | Verify | Click Verify: the card is recomputed from committed responses and matches. |
| 0:25 | Tamper | One number changed, the fingerprint moves, verification fails. |
| 0:36 | Families | Same ablation, one judge: Qwen 0.51 to 0.97, Mistral 0.81, Glimmer 0.16. |
| 0:57 | Scan | Upload your own response files and get a card. |
| 1:06 | Re-score | The judge runs live on the same responses and the fingerprint still matches. |
| 1:29 | Card review | The card itself: verdict, drift, harmful, validity, structure, limits. |
| 1:51 | Close | A measurement, its evidence, and its limits. |
Those timestamps describe the cut above. The narration and the numbers are the same in every copy.
The order is an argument, not a feature tour. First the gap, because without it the product is a scanner looking for a problem. Then the build, to show the tool is small. Then the run, because a 30B claim needs a timestamp. Then the families, because one model proves nothing. Then the three things a judge can check without trusting us: the card, the verify and the tamper test. The close answers the budget question before it is asked.
The script is in docs/VIDEO.md, the scene list in
video/captions.srt.