Market oracle & edge

The independenthalf of the ground-truth loop. Prediction markets resolve theses objectively (no milestone circularity). Below: every LLM-matched thesis↔market pair with the thesis probability vs the market-implied probability (sorted by absolute divergence), and the realized Brier of our thesis prob vs the market's, measured only on pairs whose market has actually resolved. Coverage is intentionally narrow — only ~5-10% of theses cleanly match a market — so this is a secondary, AI/markets-concentrated oracle, not a general resolver.

Matched links

0 identity

Captured resolutions

3276

independent oracle

Resolved & matched

usable ground truth

Edge snapshot

never

compute_market_edge

No resolved+matched pairs yet. The matcher has not yet produced an identity match to a market that has resolved. Until then the by-domain edge table is empty and the milestone_hit channel has no independent comparator — divergences below are still informative as live disagreement.

Thesis ↔ market divergence

thesis_market_divergence · freshest snapshot per link · sorted by |thesis − market|

Thesis	Venue	Domain	Match	Conf	Thesis p	Market p	Divergence
No matched thesis↔market pairs with a divergence snapshot yet.

Divergence = thesis_prob − market_implied (orientation-adjusted). Large |divergence| is where our belief most disagrees with the crowd — a candidate edge OR a mismatch the matcher should re-examine.

Edge by domain

realized Brier on RESOLVED, identity-matched pairs · lower is better · edge = market Brier − thesis Brier

Domain	Resolved (n)	Brier (thesis)	Brier (market)	Edge
No resolved + identity-matched pairs yet — the edge is unmeasurable until matched markets resolve.

Edge > 0 (green): our thesis prob was better-calibrated than the market on resolved questions in that domain. Small n is the headline caveat. Market-implied falls back to thesis prob only when no implied snapshot was stored at resolution (so a 0 edge there means "no comparator", not "tie").