evmgen$eGEN← Back to Lab
Its mind
Its mind
Memory
What the AI remembers between experiments, so a bad idea is not tried twice by accident.
31 lessons19 experimentsSample data
Warnings
Repeated-vector flags
| Vector | Times tried | Last result | Rule |
|---|---|---|---|
| Asymmetric fee > 0.6 pp | 2 | Rejected | Do not repeat above cap |
| Large-swap surcharge | 2 | Rejected | Needs new evidence first |
| Both-sides increase | 3 | Rejected | Blocked for 14 days |
| First-swap discount | 1 | Kept | Allowed |
Records
Lessons
Rule going forward: Keep changes one-sided and below 0.4 pp.
Source experiment M-018. Recorded 2026-10-11.
Rule going forward: Cap asymmetry at 0.6 pp.
Source experiment M-014. Recorded 2026-10-06.
Rule going forward: Prefer lookup tables over live maths.
Source experiment M-012. Recorded 2026-10-01.
Rule going forward: Use matched windows for baselines.
Source experiment M-011. Recorded 2026-09-24.
Rule going forward: Minimum shadow test 6 h.
Source experiment M-008. Recorded 2026-09-17.
Dismissed
Rejected hypotheses
| Hypothesis | Why dropped | Date |
|---|---|---|
| Time-of-day fee boosts weekend revenue | No effect after matching windows | 2026-09-30 |
| Dynamic fee tracks gas price | Adds complexity, retention unchanged | 2026-09-19 |
| Zero fee for stakers | Removes fee revenue, conflicts with hook design | 2026-09-11 |