Sanity-check AI output before you act on it

Sanity-check this AI output before I forward it / act on it. Goal is to catch confidently-wrong claims, not to nitpick style.

The output:
```
{{paste AI output — analysis, summary, recommendation, etc.}}
```

Source material the AI was working from:
```
{{paste source data, document, or "no source — generated from prompt alone"}}
```

Run these checks:

1. **Specific claims to verify** — list every concrete claim (numbers, names, dates, citations, quoted text). For each: does it appear in the source, or is it the AI's invention?

2. **Quantitative claims** — every number, percentage, or comparison. Math right? Magnitude plausible?

3. **Logical leaps** — places where the AI moved from "X is true" to "therefore Y" without showing the work. Are any of those leaps wrong?

4. **Unsupported confidence** — sentences that sound certain but rest on assumptions the AI didn't surface.

5. **What's missing** — anything obvious the AI should have addressed but skipped (counter-arguments, caveats, alternative explanations).

Output:
- ✓ Claims that check out
- ⚠ Claims to verify before forwarding (with how to verify)
- ✗ Confirmed wrong (with evidence)
- ⊘ Things the output should address but doesn't

Be terse.