Experiments · E195c
With a third run each, does the fly still only tie the best simple search?
No. The fly found 299 distinct alloys against 238, well beyond the run-to-run spread, at the same rate.
In the log: queued (2026-09-19 01:12)
book/assets/diagrams/exp/E195c.svg).Results
EXPERIMENTS.md · line 12611E195c queued (2026-09-19 01:12). Third seed (offset 2) for elitist and fly on E195b's settings. Prediction: the count tie stands (|fly − elitist| within one three-seed sd) and elitist's AUC_Q lead stands (≥ 10). Chain runs/e195c_chain.sh, after E200.
E195c — third seeds, combined with E201b. Three seeds each, pure reward:
| arm | AUC_Q | distinct finds |
|---|---|---|
| fly | 147.1 ± 17.7 (135.4 / 138.4 / 167.5) | 299.3 ± 31.0 (300 / 268 / 330) |
| elitist | 144.8 ± 11.0 | 238.3 ± 13.6 |
| fly − elitist: +61 finds against a seed-sd of 31 (two sd); AUC_Q +2.3, a dead tie. | ||
| My prediction that elitist's rate lead (≥ 10) would stand is falsified — at three seeds | ||
| there is no rate lead. The fly, unmodified, on the reward as it should always have been | ||
| paid, finds a quarter more distinct qualifying alloys than keep-the-best-and-mutate at the | ||
| same rate. This is the generator line's result and it is positive. |
The full record
This entry is written in 2 separate places in the log, shown here in log order.
EXPERIMENTS.md · line 12611
E195c queued (2026-09-19 01:12). Third seed (offset 2) for elitist and fly on E195b's settings. Prediction: the count tie stands (|fly − elitist| within one three-seed sd) and elitist's AUC_Q lead stands (≥ 10). Chain runs/e195c_chain.sh, after E200.
EXPERIMENTS.md · lines 12967–12976
E195c — third seeds, combined with E201b. Three seeds each, pure reward:
| arm | AUC_Q | distinct finds |
|---|---|---|
| fly | 147.1 ± 17.7 (135.4 / 138.4 / 167.5) | 299.3 ± 31.0 (300 / 268 / 330) |
| elitist | 144.8 ± 11.0 | 238.3 ± 13.6 |
| fly − elitist: +61 finds against a seed-sd of 31 (two sd); AUC_Q +2.3, a dead tie. | ||
| My prediction that elitist's rate lead (≥ 10) would stand is falsified — at three seeds | ||
| there is no rate lead. The fly, unmodified, on the reward as it should always have been | ||
| paid, finds a quarter more distinct qualifying alloys than keep-the-best-and-mutate at the | ||
| same rate. This is the generator line's result and it is positive. |
