What does taker order flow look like in the hour before LB6 enters?
FRAME
- Subject: LB6 Config B, 6 BTC engines (3 long-only, 3 short-only), simulated tradelist; taker flow from Binance Vision BTCUSDT USDT-M perpetual 1-minute klines (
quote_volume,taker_buy_quote_volume). - Why it matters: LB6 decides from options data only; order flow is a separate source. This maps flow at LB6 entries before any bet that flow sorts good entries from bad.
- Scope: trades entered before 2026-02-01 (201 trades); holdout untouched (
docs/data-policy.md). Descriptive, no claim. Method locked indocs/bets/2026-11-18-lb6-flow-baseline.md, commite0e3d5a, written before computing. - Terms: taker-buy share = taker-buy quote volume / quote volume over the 60 whole minutes before entry (323 of 337 entries are on the hour, so this is the last full hour the engines could see). With flow = long while share > 0.5, or short while share < 0.5; against flow = the opposite. Within engine = against-minus-with difference inside each engine, averaged weighted by trades. CI = 2,000 bootstrap resamples within engine, seed 0.
MEASURES
| measure | value |
|---|---|
| hourly taker-buy share, all hours: 10th / median / 90th percentile | 0.425 / 0.495 / 0.566 |
| median share before long entries | 0.490 |
| median share before short entries | 0.496 |
| trades entered against flow | 103 of 201 (51%) |
| mean net %, with flow | +0.85 |
| mean net %, against flow | +0.67 |
| against minus with, pooled (95% CI) | -0.17 pp (-0.69 to +0.41) |
| against minus with, within engine (95% CI) | -0.21 pp (-0.75 to +0.41) |
| engines where against-flow trades are weaker | 3 of 6 (one of them by -0.02 pp) |
| against minus with, within engine, without deepput_strike_C14_G09 | +0.07 pp |
Per engine:
Post-hoc (run after the measures above were seen): without the 0.5 cut, flow aligned with the trade's side ((share - 0.5) x +1 for long, x -1 for short) ranked against net % inside each engine: rank correlation +0.03 overall; per engine FOCUSED_TOP_SHARPE -0.27, deepdw_iv_M04 -0.25, deepput_strike_C14_G09 +0.27, dw_tail_C04 +0.19, iv_deepcall_0804_SL8 +0.00, netdw_skew_C01 +0.05.
| engine_id | side | trades | n_with | n_against | median_share | against_pct | with_mean | against_mean | diff |
|---|---|---|---|---|---|---|---|---|---|
| deepput_strike_C14_G09 | long | 34 | 11 | 23 | 0.486 | 68 | +1.70 | +0.14 | -1.56 |
| netdw_skew_C01 | short | 60 | 26 | 34 | 0.508 | 57 | +0.75 | +0.59 | -0.16 |
| deepdw_iv_M04 | long | 25 | 13 | 12 | 0.503 | 48 | +0.41 | +0.39 | -0.02 |
| dw_tail_C04 | short | 26 | 21 | 5 | 0.478 | 19 | +0.62 | +0.72 | +0.10 |
| iv_deepcall_0804_SL8 | long | 34 | 16 | 18 | 0.494 | 53 | +0.84 | +1.13 | +0.29 |
| FOCUSED_TOP_SHARPE | short | 22 | 11 | 11 | 0.498 | 50 | +1.17 | +1.61 | +0.43 |
UPDATE
- Status: MEASURED.
- What it means: flow in the hour before entry is close to balanced for both sides; against-flow trades are -0.17 pp pooled and -0.21 pp within engine, both CIs include zero, weaker in 3 of 6 engines; without deepput_strike_C14_G09 the within-engine difference is +0.07 pp. Absence of evidence, not evidence of no effect: the CI still allows an effect near a typical trade's mean.
- Rejected:
lb6-against-flow-weaker(before 2026-02-01 the difference was -0.21 pp within engine with a 95% CI from -0.75 to +0.41, weaker in only 3 of 6 engines; one engine (deepput_strike_C14_G09, -1.56 pp on 34 trades) carries it, and picking it after seeing six would not be a blind bet (decided 2026-11-18))