No side ships until both gates clear.
Gate 1 proves replay is real. Gate 1b proves accuracy beats the baseline. Five sides frozen until then.
Determinism first. Then calibration. Then the sides wake.
Gate 1 (bit-for-bit replay, 30 consecutive days) unlocks only the B2B wedge — with the consumer surface off. Gate 1b (beat the disclosed Brier/log-loss baseline) unlocks the consumer sides. Miss Gate 1b by month 6 and the forecast/oracle/intel/fund sides are cut, not shipped.
Replay green · 30 consecutive days.
Every simulation must reproduce exactly — same inputs, same output, same bytes — for 30 consecutive days. No approximation. No rounding. Bit-for-bit. This is the proof that the oracle is not a dashboard — it is an audit instrument.
Beat the disclosed Brier/log-loss baseline on ≥1 series.
Beat the disclosed Brier/log-loss baseline on ≥1 forecast series, plus one signed paid pilot before month 6. Gate 1b is the commercial proof gate — accuracy that survives comparison to a public benchmark, not just internal green-flag.
Five sides. All frozen.
The five sides — Citizens, Verified Actors, Intel Subscribers, Market Prediction, Investment Funds — are all ROADMAP and frozen behind these gates. No side ships until the gates clear.
Certified provenance ≠ certified truth.
Every output replayable. Not every output correct.