# cannot-say-done

Idea IB-0306. Status: DRAFT, nothing posted.

## Slides

**Cover.** My AI agent cannot say done any more. A gate refuses the word until a check passes.

**Slide 2.** Agents round up. Done meant I wrote it, not I checked it.

**Slide 3.** Every plan carries one line: what must be true when this is finished.

**Slide 4.** At wrap-up a script runs that line. If it fails, the word done is refused.

**Slide 5.** Data slide. Live: opened the live page. Tested: ran it and saw output. Fixed: the original failing case now passes. Tracking on: the hit is seen in the analytics debug view.

**Slide 6.** Checklist: write what you will call done, check it at its real source today, write what you ran and what you saw, put a date on every number, break it once and watch the check fail.

**Slide 7.** Quote. A check that has never been seen to fail is not a check.

**Slide 8.** Comment FLEET for the verify-before-done checklist plus the free Fleet Room invite. Honest limit: the gate checks what I told it to check. It does not make the plan good.

## Claims and sources

| Claim | Source | Read-date | Note |
|---|---|---|---|
| No numeric claims on these slides | n/a | 2026-10-11 | The deck uses none |
| Live / tested / fixed / tracking-on proof pairs | fleet-deliverable-01 five-checks table, pack 03 script | 2026-10-11 |  |
| Gate refuses done unless the outcome line passes | Pack 03 script | 2026-10-11 | Described, not a metric |