| gate | what it requires | status |
|---|---|---|
| Forecast calibration | Ensemble Brier must beat the market/base rate on held-out questions | BLOCKED · re-run 08-18 failed uniformly |
| A significant book | One paper book profitable beyond noise over a multi-week window | BLOCKED · drift is the best book and still only t=+0.85 over 20 days; needs ~106. AND DO NOT RANK ON THE BOARD'S LIFETIME EQUITY: driftNEW (+347%) and driftD15NEW (+177%) are FORKS that inherited their parent's history at the fork, so most of that was never earned by those arms; 89% of drift's own lifetime P&L predates 2026-08-05. Rank on post-fork earnings only. |
| Break-even win rate | Realised win rate above the payoff-implied bar, not above 50% | CLEARED · 9 of 11 drift books beat their own bar; the 2 below it are the 2 that lost money |
| Execution cost | Fills modelled on real depth, not snapshot/mid | MEASURED · bookwalk arms pay real VWAP |
| Fee model | Taker fee charged on the same side live charges it | MODELLED · maker and taker are equal here |
| Position sizing | A size rule that survives its own capacity test | PARTIAL · DIVE-314: cap decided at $5 with a live-only size (books keep amt 10), but NOT implemented — and it FAILS its own capacity test: $44.91 funds ~6 slots against the 25 configured, and MIN_SHARES=5 puts a 95c contract at $4.75, so $5 admits ONE minimum order in the 90c+ bucket. A decision is not a size rule. |
| Kill switch | One action that halts entries and is provable from outside the process | CLEARED · DIVE-314: executor refuses by default (EXEC_ARMED only, no endpoint can flip it); sim/live_kill.sh covers BOTH processes and exits 3 not 0 while resting orders are unverifiable — entries halted and provable, pre-existing resting orders are NOT |
| Funding + limits | Wallet funded to a stated cap with a per-day loss limit | NOT STARTED · needs a human decision |