2026-W40 · regular run 2026-10-04
Testnet since the 2026-09-09 reset. Test windows 22 to 27 days, 44 to 192 fills per pair. Control arm: 12 tiers, 2.5% spacing, fixed order size. Test arm: 3.25% spacing. There is no buy-and-hold comparison.
Gate result
16 experiments meet all five criteria (14+ days two-sided, 30+ fills, net > 0, drawdown < 10%, won its A/B). The control arm won all 13 new-pair experiments. DOTEUR's 8.87% drawdown is the only one near the limit.
Walk-forward flags (7-day train / 7-day test, simulated)
| Pair | Control arm | Test arm | Note |
|---|---|---|---|
| SUIEUR | robust | robust | 101 fills, robust two weeks running |
| WLDEUR | robust | robust | 192 fills, robust two weeks running |
| ICPEUR | robust | robust | |
| BCHEUR, DOTEUR, LTCEUR, POLEUR, SOLEUR, ADAEUR | robust | robust | |
| AVAXEUR, XRPEUR | robust | not robust | |
| LINKEUR, XLMEUR | overfit | overfit | Positive on old data, negative on new. Not promotable. |
Walk-forward is a backtest on 14 days of hourly candles. "Robust" means both simulated halves were positive. It is weak evidence and independent of the testnet trades.
Real money
| Bot | Since | Fills this week | Note |
|---|---|---|---|
| ADAEUR | earlier | 0 | No fills since 2026-09-21. |
| ETHEUR | earlier | 0 | No fills since 2026-09-21. |
| SHIBEUR | 2026-10-01 | 0 | Price 2.6% above the grid's top. Re-range or pause pending. |
The real-money capital cap is fully committed; no new pilot can start without new funding. Account value moved with the ETH price this week, which is not bot PnL.
Decision
Nothing promoted. A proposed cost-basis file installed on 2026-10-06 raised reported PnL on several pairs 3 to 6x; it is an accounting assumption under review, so no ranking from that re-run is used here.