How much does a simulated grid's result depend on when it starts?
FRAME
- Subject: the grids of grid-neutral and grid-strike-lb6 (
lib/grid.py), Binance Vision BTCUSDT USDT-M perpetual 1-minute bars. - Why it matters: every grid report so far simulated one path from the first minute of its window. Grids re-anchor where the previous one ended, so the first anchor shapes all later ones.
- Scope: 24 hourly start times on the first day of each window. Before 2026-02-01: starts 2025-04-24 00:00..23:00 UTC to 2026-02-01. Holdout, post-hoc robustness check of grid-strike-lb6 (status unchanged): starts 2026-02-01 00:00..23:00 to 2026-08-26. Method locked in
docs/bets/2026-11-18-grid-start-sensitivity.md, commitb1ac306; not blind (six starts before 2026-02-01 seen first). Parameters: step 0.5% (median BTC 1h range before 2026-02-01: 0.46%); half-range 3%, 13 levels (median 24h range 2.63%; LB6 median hold 24h); take-profit one step; stop one step beyond the far end; at most 2 resting open orders; maker fee 0.02%, taker fee 0.05%; 1,000 USDT capital, leverage 1; re-anchors at the current price when an executor ends. - Terms: halves in % of the 1,000 USDT total, as in grid-neutral. Calmar = total / max drawdown (1-minute mark-to-market).
MEASURES
Before 2026-02-01, total %:
| grid | min | median | max |
|---|---|---|---|
| neutral | -15.51 | -6.48 | -3.71 |
| BUY half | -8.62 | -7.11 | -5.85 |
| SELL half | -9.66 | +0.79 | +2.63 |
Max drawdown (pts):
| grid | min | median | max |
|---|---|---|---|
| neutral | 10.87 | 11.68 | 18.76 |
| BUY half | 9.97 | 10.73 | 11.29 |
| SELL half | 5.93 | 7.09 | 15.29 |
Calmar:
| grid | min | median | max |
|---|---|---|---|
| neutral | -0.83 | -0.55 | -0.33 |
| BUY half | -0.79 | -0.69 | -0.59 |
| SELL half | -0.63 | +0.15 | +0.44 |
Holdout (post-hoc), Calmar:
| grid | min | median | max |
|---|---|---|---|
| LB6-directed | +1.96 | +2.22 | +2.22 |
| neutral | -0.58 | +0.67 | +0.82 |
| long-only | -0.44 | +0.30 | +0.60 |
Total %:
| grid | min | median | max |
|---|---|---|---|
| LB6-directed | +7.81 | +8.88 | +8.88 |
| neutral | -5.98 | +5.07 | +6.15 |
| long-only | -5.83 | +3.20 | +6.27 |
LB6-directed Calmar beats the better plain grid at the same start in 24 of 24 starts. Starts 00:00..22:00 give the LB6-directed grid one identical path (LB6 first takes a side at 22:00 on 2026-02-01), so this mostly varies the plain grids.
Post-hoc, added after a cold read: LB6-directed grid with LB6's direction acted on 0 to 55 minutes late (step 5), which moves its own anchors:
| delay_min | total_pct | max_dd_pts | calmar |
|---|---|---|---|
| 0 | +8.88 | 4.00 | 2.22 |
| 5 | +9.81 | 3.75 | 2.62 |
| 10 | +4.78 | 6.40 | 0.75 |
| 15 | +13.27 | 4.27 | 3.11 |
| 20 | +15.48 | 4.64 | 3.34 |
| 25 | +21.17 | 4.38 | 4.83 |
| 30 | +18.59 | 4.67 | 3.98 |
| 35 | +13.35 | 5.26 | 2.54 |
| 40 | +12.82 | 4.53 | 2.83 |
| 45 | +11.03 | 5.39 | 2.05 |
| 50 | +9.98 | 5.51 | 1.81 |
| 55 | +7.37 | 5.99 | 1.23 |
Delayed runs above the best plain-grid Calmar over all 24 starts (0.82): 11 of 12. The locked run (start 2026-02-01 00:00) gave LB6-directed 2.22, neutral 0.65, long-only 0.39.
UPDATE
- Status: MEASURED.
- What it means: before 2026-02-01 the neutral grid's total ranges -15.5% to -3.7% by start hour alone; single-path grid numbers are one draw from this spread. On the holdout, LB6-directed beats the better plain grid in 24 of 24 starts.