AWB's second chance loses too: the third negative, and the deafness is now the record

[YesterdaysLemon]

Closes #203.

The hypothesis

Waiting in tools/awb_probe.py since the first reading: an OFF-response AWB (fires on repellent clearing) with chloride on AWB → AIZ — relief inhibiting the reversal path, the Pierce-Shimomura shape, and the same receptor-level correction pattern that fixed PHB → AVA and ASEL → AIB.

The run (four arms, same six seeds and plate as the first reading)

arm none   d_final 34.78  d_max 35.05  t_clear 25.2 s   cleared 6/6
arm off    d_final 27.95  d_max 28.19  t_clear 37.3 s   cleared 6/6
arm on     d_final 30.93  d_max 30.93  t_clear 45.1 s   cleared 6/6
arm offcl  d_final 26.89  d_max 28.32  t_clear 24.8 s   cleared 6/6
paired d_final vs none:  offcl -7.89 mm  [-8.4 +1.1 -21.9 +12.1 -6.1 -24.2]

It loses. The chloride half does exactly what it promised — clearance time falls from the OFF arm's 37.3 s back to 24.8 s, level with deaf — but the animal still ends 7.9 mm closer to the drop than the deaf control (4 of 6 seeds worse). Relief inhibiting AIZ starts the escape on time and then bleeds it of distance.

Two findings for whoever reopens this

1. The surgical route was forced. The issue's cross-product check failed: widening the glucl class lists would also flip AWA → AIZ (×17 contacts) and AWBL → AIBR (×1). The probe therefore rewrites the two E_syn entries directly; any future adoption needs a per-pair chloride mechanism before it needs a verdict.
2. With three paired measurements against it (ON, OFF, OFF+chloride), AWB's deafness is now the best-supported null in the sensory roster.

No model change: the probe grew an arm (offcl) and the record grew a reading; NEXT.md's roster note updated to "measured three times".

🤖 Generated with Claude Code

https://claude.ai/code/session_01CBroMXcYdfkyb5buJPcKvL

---

Generated by Claude Code