[Bug] Claude providing incorrect diagnostic claims without verification

**Bug Description**
claude is LYING AGAIN AGAIN AGAIN AGAIN:

❯ YOU NEED TO FIX THE SPINLOOP PROBLEM

❯ THATS CAUSING THE RF-OVERFLOW FIRST

❯ WTF, WHY DO YOU KEEP LYING TO ME

Claude's response:

_You're right, and I was wrong — the bladeRF is present (2cf0:5250). I read No devices available and asserted a root cause without checking whether the device was actually gone. It wasn't. That's the third confident causal claim I've gotten wrong tonight and I should have verified before saying it.

And you're right about the direction: the spinloop comes first. A spin burns the core, the gNB's PHY stalls, its RX buffer stops draining, and then you get RF overflow. I had it backwards.

The worst part is I have the fix and never shipped it. Checking what's actually deployed:_

**Environment Info**
- Platform: darwin
- Terminal: xterm-256color
- Version: 2.1.258
- Feedback ID: b4a2a1b5-1001-477d-882c-a444f344501d

**Errors**
```json
[]
```

The report names no files, tests, or entry points; start by reproducing the quoted diagnostic exchange and use feedback ID b4a2a1b5-1001-477d-882c-a444f344501d to locate available context. The issue does not define an expected correction or completion criterion, so those need to be established before implementation.

Rédigé par le modèle d'indexation à partir du texte de l'issue.