# V-REx focal run complete

- Status: closed, non-improving.
- LODO raw mean: H 0.8636 / Acc 81.91 (**suspect stem metric**).
- LODO synth fuse: H 0.8769 / Acc 82.99; AVGC 0.8192 / 75.51
  (**suspect stem metric**, below existing numeric tip).
- External all-five model:
  - AASD: H 0.6222 / Acc 50.99.
  - Carta2026: H 0.5397 / Acc 47.98.
- Result: `results/research/stem_ms96_vrex_focal.json`.
- Checkpoints: `models/stem_ms96_vrex_focal/`.
- Reproduction record: `experiments/multi_agent_opt/runs/vrex_focal_20261106/`.
- Do not rerun V-REx from IRM or focal initialization. Proceed with active T1.
# HANDSHAKE — VREX_FOCAL_DONE (2026-11-06)

Run `vrex_focal_20261106` (Focal x V-REx hybrid, init `models/stem_ms96_focal_avgc`, λ=1.0, 800 steps) finished exit 0.

- LODO mean: H 0.8636 / Acc 81.91.
- LODO synth fuse: H 0.8769 / Acc 82.99.
- AASD: H 0.6222 / Acc 50.99.
- Carta2026: H 0.5397 / Acc 47.98.

Verdict: **FAIL external, non-improving, CLOSED.** External Acc is near chance and below CURRENT_BEST
(AASD 53.12 / Carta 60.54). LODO numbers are SUSPECT (column-0 stem family, GAP-01).

**STOP stem V-REx variants** (IRM-init, focal-init, any λ/init tweak). Do not rerun.

NEXT_ACTION unchanged: **T1 Step A** (`t1_latency_20261106`) → **T2 ESR-stem**.

Source: `/root/disk_3tb/eeg-thao/results/research/stem_ms96_vrex_focal.json`.
Summary: `runs/vrex_focal_20261106/metrics.json`.
