How the Oracle works.
Five mechanisms. One invariant: every output replayable.
Not five products. Five mechanisms over one shared core.
Each mechanism reuses the same Rust engine and the same per-output audit manifest, and powers a specific side. One codebase — not six.
Verified actors affirm or dispute claims on the record.
Each affirmation is an AxisOverride: a new immutable NodeVersion, hash-chained into the audit manifest. The doubt that triggered it is sealed and resolved.
Every output seals a per-output hash-chained manifest.
Bit-for-bit reproducible. Not certified truth — certified provenance. The manifest proves who affirmed what, when, and that it replays exactly.
Certified provenance ≠ certified truth.
Every output replayable. Not every output correct.
Realized outcomes score each Sister. The reward shapes future simulations.
The calibration loop is the moat. Every realized outcome updates the Brier/log-loss score for each Sister. Gate 1b: beat the disclosed baseline.
23 A2A CRDT agents. 5 personality sisters. One merged ensemble.
Analyst, Contrarian, Disruptor, Historian, Institutionalist — each imagines plausible futures independently via A2A protocol. The Oracle merges into a normalized ensemble.
No side ships until both gates clear.
Gate 1: replay green · 30 consecutive days. Gate 1b: beat disclosed Brier/log-loss baseline on ≥1 series + one signed paid pilot before month 6.