Binance USDT-M L1 (same venue for signal/quote/fill proxy). Fill proxy = bid touched by next-bar mid. No queue, no adverse selection -> a ceiling. Taker fee 1.0 bps/side; maker 0. Kill bar: maker net > +1 bps/trade at h=36.
| signal | h | n | fill rate | mid bps | taker bps | maker bps | maker t |
|---|
| qimb | 12 | 1663 | 50% | +3.60 | +1.58 | -6.34 | -3.84 |
| qimb | 36 | 1661 | 50% | +5.90 | +3.87 | -3.67 | -1.12 |
| mpdev_bps | 12 | 1430 | 50% | +1.90 | -0.14 | -8.80 | -3.98 |
| mpdev_bps | 36 | 1430 | 50% | +3.24 | +1.21 | -7.97 | -2.12 |
| signal | h | n | fill rate | mid bps | taker bps | maker bps | maker t |
|---|
| qimb | 12 | 1640 | 50% | +3.65 | +1.61 | -7.66 | -4.12 |
| qimb | 36 | 1636 | 50% | +3.09 | +1.05 | -7.19 | -2.07 |
| mpdev_bps | 12 | 1482 | 50% | +2.24 | +0.19 | -8.32 | -3.63 |
| mpdev_bps | 36 | 1478 | 50% | +1.57 | -0.48 | -8.97 | -2.33 |
- The maker leg is negative despite being an optimistic ceiling (~ -4 to -9 bps/trade, t -1 to -4): a passive bid fills 50% of the time, and precisely when it fills the mid has ticked down — classic adverse selection. For these momentum-style signals, passivity destroys the edge; aggression is better. No maker path is justified by this.
- The taker leg is small-positive on this 3-month window (+1 to +4 bps) for qimb — this contradicts the full-history REJECT (
crypto_l1_edge_test.md) and is most likely regime/sample noise, not a reason to act. The maker conclusion does not depend on it.
- Result: REJECT — fast microstructure signals are not rescued by maker entry; only paid L2/queue data could test a smarter passive policy (join/queue/peg), and even then adverse selection caps it.