EventHorizonIQ
Live calls →The Proof
Published calls are dated, and resolved calls are graded on the public ledger — the wins and the misses. Held calls are listed with their numbers withheld. This page is the math behind the record: not just a hit-rate, but whether the probabilities are actually calibrated, and whether they beat the prediction markets. Real-time calls only; backtests are excluded by construction.
Record (real-time)
25-13
rate withheld — see methodology
Brier score
0.215
rate withheld — see methodology
Discrimination (AUC)
0.71
separates hits from misses
Vs. recorded market price
86%
7 disagreements · +41.5pp avg realized · market leg not venue-sourced
Call #67 was disputed and adjudicated on 2026-10-19; the record above counts it as ruled. Under the other reading the record would show 24 hits and 14 misses of 65 resolved calls. Its history includes a July 23 change made without recorded approval; the ruling, the other reading and the governance notes are on the call's page.
Are the probabilities honest?
When we say 70%, does it happen ~70% of the time? Each dot is a decile of stated probability; the dashed line is perfect calibration. Dots on the line = honest forecasts, not just lucky ones. (48 scored calls)
When we disagreed with the market
Of the 13 calls made against a live market price (Kalshi / Polymarket), we disagreed materially 7 times. When we took the other side of the market, we were right 86% of the time (95% CI 49%–97%) — an average realized edge of +41.5pp per call over trading the market at its own price. The market prices here are the ones we recorded at call time, not venue-sourced history, so this is not an independently verified edge over the market.
By methodology
| Method | N | Hit | Brier |
|---|---|---|---|
| unlabeled | 30 | — | 0.206 |
| sem_only | 11 | — | 0.281 |
| manual | 4 | — | 0.114 |
| tiq_sem | 1 | — | 0.116 |
| multi_engine | 1 | — | 0.116 |
| base_rate | 1 | — | 0.360 |
Performance & equity curve
Performance and the equity curve are private to EHIQ members. The predictions record above is graded fully in public.
Append-only. Calls are never edited or deleted after posting. Calibration is computed over real-time calls only; a right outcome reached via wrong reasoning is graded a calibration note, not a clean hit. This is not financial advice.