Do LB6 entries taken against another engine's open position earn less per trade?
FRAME
- Subject: LB6 Config B, a BTC strategy of 6 engines. Each engine reads hourly options data and opens only long or only short (3 long, 3 short). 337 simulated trades, 2025-04-23 to 2026-08-22, net of 0.05% fee per side.
- Why it matters: Config B is the version planned for live. If its engines often trade against each other, that interaction is a lever no single engine shows.
- Prior work: none on engine interaction. Earlier single-feature scans of this tradelist found no options metric that separates results across the portfolio (all BH q >= 0.94).
- Current best: Config B as is: mean +0.710% per trade.
- Scoreboard: pak Yon judges by drawdown and Calmar: frozen-core3 was killed on drawdown budget, live runs a 15% drawdown kill switch, the 34-sleeve roster was picked by Calmar. [not confirmed for LB6]
- Question: 6 engines often hold positions at the same time, some in opposite directions. Is an entry taken against another engine's open position weaker than other entries, and is that more than a mix of weak engines?
- Scope: descriptive only; no rule is changed here. In-sample. Data:
trading-infracommit53c6993^,docs/lb6_tweaked/reference_tradelists/*_oracle_B_trades.csv, joined to hourly options without look-ahead (poc/join_b.py). - Terms: "opposite entry" = at entry time, at least one other engine holds an open position on the other side (no engine overlaps itself; if both same-side and opposite-side positions are open, the entry counts as opposite).
BET
Opposite entries earn less per trade than other entries, within engines and not driven by any one engine.
- Adjective: weaker
- why: the question is whether trading against another engine costs return; weaker per trade is that cost.
- Measured by: mean net % per trade, opposite minus other; more negative = weaker.
- why: per-trade mean is the unit the pattern was first seen in, and it does not depend on how many trades each group has.
- Because: in the pooled tradelist, opposite entries average +0.348% vs +1.062%.
- Falsified if: within-engine difference, n-weighted, > -0.30 pp; or the difference is >= 0 with any one engine removed. Locked in
docs/bets/2026-11-17-lb6-f1.md, commitc272e6e, written before any test ran.
KILL
Within-engine: difference computed inside each engine, then averaged weighted by engine trade count. Leave-one-out: opposite labels recomputed after removing one engine.
| test | falsified if | result | status |
|---|---|---|---|
| within-engine, n-weighted | > -0.30 pp | -0.769 | pass |
| leave-one-engine-out, worst | >= 0 | -0.490 | pass |
Pooled:
| group | n | mean | total | win_rate |
|---|---|---|---|---|
| opposite | 166 | +0.348 | +57.8 | 56.6% |
| other | 171 | +1.062 | +181.6 | 68.4% |
Per engine:
| engine | side | n | n_opposite | mean_opposite | mean_other | diff |
|---|---|---|---|---|---|---|
| FOCUSED_TOP_SHARPE | short | 36 | 23 | +0.852 | +1.944 | -1.092 |
| deepdw_iv_M04 | long | 33 | 13 | +0.692 | +0.613 | +0.079 |
| deepput_strike_C14_G09 | long | 67 | 39 | -0.148 | +1.364 | -1.512 |
| dw_tail_C04 | short | 41 | 10 | +0.075 | +1.089 | -1.014 |
| iv_deepcall_0804_SL8 | long | 58 | 33 | +0.691 | +0.705 | -0.014 |
| netdw_skew_C01 | short | 102 | 48 | +0.239 | +1.010 | -0.771 |
Leave-one-engine-out:
| engine_removed | n | diff |
|---|---|---|
| FOCUSED_TOP_SHARPE | 301 | -0.724 |
| deepdw_iv_M04 | 304 | -0.803 |
| deepput_strike_C14_G09 | 270 | -0.558 |
| dw_tail_C04 | 296 | -0.576 |
| iv_deepcall_0804_SL8 | 279 | -0.898 |
| netdw_skew_C01 | 235 | -0.490 |
Pooled difference (-0.714) and n-weighted within-engine difference (-0.769) differ because the second is computed per engine first.
UPDATE
- Status: SURVIVED.
- What it means: trading against another engine's open position goes with lower per-trade return in 4 of 6 engines;
deepdw_iv_M04(+0.079) andiv_deepcall_0804_SL8(-0.014) are the exceptions. - Next bets:
lb6-skip-opposite(Should LB6 skip trades that go against another engine?);lb6-half-size-opposite(Does trading opposite entries at half size make Config B more efficient?) - Open: why the two exceptions are unaffected; whether the pattern holds after 2026-08-26.