# STORYBOARD and SCRIPT: Proof of Done, 3 minutes

NOTE: This is a storyboard and narration script only. NO VIDEO HAS BEEN RECORDED OR PUBLISHED.

## STORYBOARD

| Time | Visual | On screen text |
|------|--------|----------------|
| 0:00-0:20 | Chat window: an agent writes "Done, deployed!" | "Done on paper?" |
| 0:20-0:50 | Owner opens the URL: a 404 page. | "Report said live. Reality: 404." |
| 0:50-1:20 | The five signals card, one line highlighted at a time. | "5 signals" |
| 1:20-2:10 | Terminal: `proof_of_done.py --url ... --expect-status 200`. First run prints FAIL with the evidence lines. | "Observed, not claimed" |
| 2:10-2:30 | Fix applied; re-run prints LIVE_VERIFIED with timestamp. | "LIVE_VERIFIED + UTC time" |
| 2:30-2:50 | Three labels: LIVE, CODE_ONLY, BLOCKED. | "Say which one it is" |
| 2:50-3:00 | Text card. | "Advice to an agent, not an enforcement gate." |

## SCRIPT

[0:00] An agent tells you the work is done and deployed. Is it?

[0:20] You open the link and get a 404. The report was about the code. Your customer cares about the URL.

[0:50] Here are five signals that a report is done only on paper: no URL; no fetch after the change; the URL answers with the wrong thing; a file nobody compared byte for byte; and evidence from the author's own machine.

[1:20] The verifier is a small Python script with no dependencies. You give it the exact URL and what you expect: a status, a redirect target, a phrase, or a checksum. It fetches the URL, shows every redirect hop, insists on a trusted certificate, and prints what it saw. This run fails, and the evidence says why.

[2:10] After the fix, the same command prints LIVE_VERIFIED with a UTC timestamp. That line is the report, not the agent's memory.

[2:30] Every outcome gets one of three labels: live, code only, or blocked. If you could not check, say blocked.

[2:50] One honest limit. The skill file is advice to an agent. It is not a gate, and it does not force anyone to run the check. Thanks for watching.