Adaptive grid, tested — not fitted
Research console for a Bybit USDT-perp grid. Price path is OKX linear swaps (Bybit REST is geo-blocked here). 1H candles, conservative trade-through fills. 2–5% daily and 30% monthly are different magnitudes. Locked holdout does not print 30% a month.
Risk actions
Pause does not abandon open risk. Flatten requires a second click. LIVE is not wired.
Holdout equity — H2 vs buy & hold (BTC)
Locked period 2025-10-01 → 2026-09-05. H2 is the frozen candidate. Buy & hold is the market, not a strategy claim.
Verdict
NO-GO
- Loading locked holdout…
H2 drawdown (holdout)
H2 monthly return (holdout)
Holdout, BTC, frozen params — comparable risk budget
Geometric monthly vs 30% target. Cycle count is completed grid cycles, not fills. Many fills in one grid are one bet.
| Strategy | End equity | Net | Geo month | Max DD | Calmar | Cycles | Fills | Fees |
|---|
30% / month target
Missed
Need +0.88% per day compounded. H2 holdout geometric monthly is —. Hitting 30% every month would be +2230% annualized.
2–5% daily target
0/0 days ≥ 2%
2% daily compounds to +81.00% per month — a different problem than 30% monthly. Observed daily distribution does not contain that regime. Sample of daily closes: median —.
PnL decomposition (H2 holdout)
- Net—
- Fees—
- Funding—
- Slippage—
Costs are real. Tiny net on three cycles is not an edge.
Find 30% — spin the top book
8,468 rotators on full-history names never compounded 30%/mo. The universe still has a pile of eligible names ≥30% trailing 30 days. PAPER holds that book equal-weight — not the listing lottery.
Keep — frozen winners
Best rules and full parameters on a shelf so the next hunt cannot bury them. Optimization: 2,784 rotators, 0×30% holdout, turnover-aware honest +8.59%/mo, best three-window +9.26%/mo. Copy JSON from any card.
Hypos — mechanism battery
13 stories, 28 variants. 0×30%. Anti-chase (skip names already +50%/30d) is the only new paper edge at +14.1%/mo. Spin-when-hot is rejected.
Universe scan — top 100 coins, frozen grid
Same BTC params, no per-coin fit. Honest count is names that hit 30%/month in development and still did on holdout. Loading batch…
Hunt — cluster params, then freeze
Frozen BTC params found 0/55 at 30%. Cluster search on 1H and 15m: paper bar 8%/mo never appears, even as fishing. Ceiling ~1.6%/mo. Hunt page has both waves.
Ceiling — is 30% in the tape?
Drop the grid. 1× buy-hold of every full-history name: 0 at 8%/mo, 0 at 30%. 3× longs: 55/56 liquidated in holdout. Frozen momentum rotator +1.5%/mo out of sample.