Experiments · E173

Once the confirmation step is switched on, does the serotonin site actually get taught?

Yes. It received 138 and 187 lessons in the two seeds; whether that helps the search was left to a control run.

In the log: serotonin, taught: E171b's protocol with the confirm rung on

recordedDate not stated in the log; it was written between the commit of 2026-09-16 19:06 and the first commit that contains it, 2026-09-19 08:35generator · fly brain0 predictions · 1 result paragraphEXPERIMENTS.md lines 11596–11617, lines 12205–12211
exp E173 diagram
What E173 did and how it came out, drawn from this record and the files it names (book/assets/diagrams/exp/E173.svg).

Pre-registration

The pre-registration, as written

E173 — serotonin, taught: E171b's protocol with the confirm rung on

E169 taught the 5-HT site 1 and 7 times because the DNa steering collapsed screen qualification, not because the gate was high: E171b's swarm produced 81 / 83 finds per seed, and confirm() runs on every screen qualifier. So the protocol is E171b + --confirm (steps 4, fb+oa+5ht, modulation, no steering), tag e173, 200 × 4 × 2, beside E178. The 5-HT site (LAL→DN, taught with 2p−1 from the confirm rung) then receives tens of lessons per seed. Control E173c: identical with FORAGER_SITES=fb,oa (no 5-HT site; DN still in the readout via lal_dna), launched after E178 frees its slot. Note that --confirm changes what counts as a find (MC-confirmed) and adds the MC cost to the spend, so E173's AUC_Q is not comparable to E171b's — only to E173c's.

Predictions.

  1. 5-HT lessons ≥ 40 per seed (confirm fires on ≥ 40 qualifiers in 800 verifications). Below 10: the confirm rung is still not firing and the protocol is wrong again.
  2. Null: E173 and E173c differ by less than 7 AUC_Q (E171b's seed sd) and fewer than 10 finds. A three-factor site taught by ~50 scalar verdicts, read out through DN's patience gain, is not expected to change a 24-walker search at this resolution.
  3. Falsification of the null: E173 above E173c by > 10 AUC_Q on both seeds — then the serotonin site carries information about which proposals survive the confirm rung, and it is the first plastic site outside the mushroom body shown to matter.
  4. The 5-HT site's cumulative |Δw| > 1 (lessons of ordinary size at LAL eligibility).

Results

EXPERIMENTS.md · line 12205

E173 result (17:06) — serotonin taught; the verdict waits for its control. E171b + --confirm (scipy, same family as E173c): AUC_Q 32.97 ± 3.98, 61.5 ± 9.5 finds, −135. Sites: 5-HT 138 / 187 lessons (prediction 1: ≥ 40 ✓), cumulative |Δw| 11.0 / 12.0 (prediction 4: > 1 ✓); fb 575 / 652, oa 799. Readout DN 5.7 / 5.0%. The AUC_Q is not comparable to E171b (the find definition changed to MC-confirmed and MC cost enters the spend); predictions 2 and 3 (null within 7 vs E173c, or > 10 above it on both seeds) are scored when E173c lands (~5 h, scipy). Per-seed values preserved from the overwritten results file.

The full record

This entry is written in 2 separate places in the log, shown here in log order.

EXPERIMENTS.md · lines 11596–11617

E173 — serotonin, taught: E171b's protocol with the confirm rung on

E169 taught the 5-HT site 1 and 7 times because the DNa steering collapsed screen qualification, not because the gate was high: E171b's swarm produced 81 / 83 finds per seed, and confirm() runs on every screen qualifier. So the protocol is E171b + --confirm (steps 4, fb+oa+5ht, modulation, no steering), tag e173, 200 × 4 × 2, beside E178. The 5-HT site (LAL→DN, taught with 2p−1 from the confirm rung) then receives tens of lessons per seed. Control E173c: identical with FORAGER_SITES=fb,oa (no 5-HT site; DN still in the readout via lal_dna), launched after E178 frees its slot. Note that --confirm changes what counts as a find (MC-confirmed) and adds the MC cost to the spend, so E173's AUC_Q is not comparable to E171b's — only to E173c's.

Predictions.

  1. 5-HT lessons ≥ 40 per seed (confirm fires on ≥ 40 qualifiers in 800 verifications). Below 10: the confirm rung is still not firing and the protocol is wrong again.
  2. Null: E173 and E173c differ by less than 7 AUC_Q (E171b's seed sd) and fewer than 10 finds. A three-factor site taught by ~50 scalar verdicts, read out through DN's patience gain, is not expected to change a 24-walker search at this resolution.
  3. Falsification of the null: E173 above E173c by > 10 AUC_Q on both seeds — then the serotonin site carries information about which proposals survive the confirm rung, and it is the first plastic site outside the mushroom body shown to matter.
  4. The 5-HT site's cumulative |Δw| > 1 (lessons of ordinary size at LAL eligibility).
EXPERIMENTS.md · lines 12205–12211

E173 result (17:06) — serotonin taught; the verdict waits for its control. E171b + --confirm (scipy, same family as E173c): AUC_Q 32.97 ± 3.98, 61.5 ± 9.5 finds, −135. Sites: 5-HT 138 / 187 lessons (prediction 1: ≥ 40 ✓), cumulative |Δw| 11.0 / 12.0 (prediction 4: > 1 ✓); fb 575 / 652, oa 799. Readout DN 5.7 / 5.0%. The AUC_Q is not comparable to E171b (the find definition changed to MC-confirmed and MC cost enters the spend); predictions 2 and 3 (null within 7 vs E173c, or > 10 above it on both seeds) are scored when E173c lands (~5 h, scipy). Per-seed values preserved from the overwritten results file.

Related entries

Built with PRISMWebsite and visualizations made using Claude