Experiments · E199
Does teaching one brain region, once the signal can reach it, improve the search?
Withdrawn. The run was declared void: its lessons were coin flips produced by a shared random-number fault.
In the log: FB arm (2026-09-19 06:13)
book/assets/diagrams/exp/E199.svg).Results
EXPERIMENTS.md · line 12660E199, FB arm (2026-09-19 06:13). fb,oa,5ht at steps 5, --confirm: AUC_Q 81.38 ± 11.21, 149.5 ± 11.5 finds, best −102. Not comparable to E195b (confirm changes the find definition and adds MC cost to spend); the no-FB arm at the same depth decides, prediction on record: null within one seed-sd.
E199 control lost (2026-09-19 09:30). The oa,5ht arm's seed 0 finished at 08:30 (readout saved) and the process was gone by 09:29 with no traceback and no result line; seed 1 never finished. Cause not established (the night chain's parent was stopped at 08:11, the child ran on; later kills matched other patterns). Per-seed scores are printed only at the end, so seed 0 is unrecoverable. Re-run of both seeds queued behind E205 (runs/e199_control_rerun_chain.sh); the FB arm's 81.38 ± 11.21 / 149.5 ± 11.5 stands and waits for it. Lesson for stage_b: per-seed results should be written as they land (memory: long runs checkpoint or report failure) — added to the follow-ups.
E199 control — not lost after all (note 2026-09-24 00:12). The oa,5ht arm finished both seeds at 2026-09-19 10:46: runs/odt_map/stage_b_e199_oa5ht.json holds AUC_Q 88.94 and 80.44 (mean 84.69), distinct finds 156 and 145 (mean 150.5), best -102.5 meV/atom. It never reached this log. It changes nothing: E199 is void on the RNG coupling (below), which affects both arms alike.
The full record
This entry is written in 3 separate places in the log, shown here in log order.
EXPERIMENTS.md · line 12660
E199, FB arm (2026-09-19 06:13). fb,oa,5ht at steps 5, --confirm: AUC_Q 81.38 ± 11.21, 149.5 ± 11.5 finds, best −102. Not comparable to E195b (confirm changes the find definition and adds MC cost to spend); the no-FB arm at the same depth decides, prediction on record: null within one seed-sd.
EXPERIMENTS.md · line 12697
E199 control lost (2026-09-19 09:30). The oa,5ht arm's seed 0 finished at 08:30 (readout saved) and the process was gone by 09:29 with no traceback and no result line; seed 1 never finished. Cause not established (the night chain's parent was stopped at 08:11, the child ran on; later kills matched other patterns). Per-seed scores are printed only at the end, so seed 0 is unrecoverable. Re-run of both seeds queued behind E205 (runs/e199_control_rerun_chain.sh); the FB arm's 81.38 ± 11.21 / 149.5 ± 11.5 stands and waits for it. Lesson for stage_b: per-seed results should be written as they land (memory: long runs checkpoint or report failure) — added to the follow-ups.
EXPERIMENTS.md · line 12699
E199 control — not lost after all (note 2026-09-24 00:12). The oa,5ht arm finished both seeds at 2026-09-19 10:46: runs/odt_map/stage_b_e199_oa5ht.json holds AUC_Q 88.94 and 80.44 (mean 84.69), distinct finds 156 and 145 (mean 150.5), best -102.5 meV/atom. It never reached this log. It changes nothing: E199 is void on the RNG coupling (below), which affects both arms alike.
