AI vs Market Scoreboard

Name: VoxOdds AI vs Market Forecast Scoreboard
Creator: VoxOdds
License: https://creativecommons.org/licenses/by/4.0/

Every hourly AI forecast is stored with the market price captured at the same moment, then scored when the market resolves. Append-only receipts, deterministic math, losses published. Lower Brier is better. Read the full research writeup → Submit your own forecasts →

Forecasts scored

6336

9705 awaiting resolution

AI Brier (mean)

0.182

sample: 6336

Market Brier (mean)

0.132

same timestamps

AI accuracy

74%

market: 81%

Across 6336 scored forecasts, VoxOdds AI (Brier 0.182) trailed the market price (Brier 0.132) taken at the same moment. A sweep over every AI/market mixture weight (n=6336) finds no weight of AI improves on the price — the market alone is optimal so far.

Recently scored forecasts

Market	AI said	Resolved	AI Brier	Market Brier	Verdict
Valorant: Sentinels vs 2GAME Esports (BO3) - VCT Americas Stage 2 Play-In	Sentinels @ 86%	Sentinels	0.020	0.018	Market closer
Valorant: Sentinels vs 2GAME Esports (BO3) - VCT Americas Stage 2 Play-In	Sentinels @ 87%	Sentinels	0.018	0.018	AI closer
Valorant: Sentinels vs 2GAME Esports (BO3) - VCT Americas Stage 2 Play-In	Sentinels @ 87%	Sentinels	0.017	0.018	AI closer
Valorant: Sentinels vs 2GAME Esports (BO3) - VCT Americas Stage 2 Play-In	Sentinels @ 87%	Sentinels	0.016	0.018	AI closer
Valorant: Sentinels vs 2GAME Esports (BO3) - VCT Americas Stage 2 Play-In	Sentinels @ 88%	Sentinels	0.014	0.018	AI closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Flavio Cobolli @ 70%	Flavio Cobolli	0.090	0.087	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 32%	Flavio Cobolli	0.105	0.087	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 37%	Flavio Cobolli	0.134	0.087	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 44%	Flavio Cobolli	0.191	0.087	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 58%	Flavio Cobolli	0.336	0.087	Market closer
Valorant: Sentinels vs 2GAME Esports - Map 1 Winner	Sentinels @ 82%	Sentinels	0.032	0.034	AI closer
Valorant: Sentinels vs 2GAME Esports - Map 1 Winner	Sentinels @ 82%	Sentinels	0.034	0.034	AI closer
Valorant: Sentinels vs 2GAME Esports - Map 1 Winner	Sentinels @ 82%	Sentinels	0.034	0.034	AI closer
Valorant: Sentinels vs 2GAME Esports - Map 1 Winner	Sentinels @ 82%	Sentinels	0.033	0.034	AI closer
Valorant: Sentinels vs 2GAME Esports - Map 1 Winner	Sentinels @ 82%	Sentinels	0.032	0.034	AI closer
Will Elon Musk post 220-239 tweets from August 14 to August 21, 2026?	No @ 98%	No	0.000	0.001	AI closer
Will Elon Musk post 220-239 tweets from August 14 to August 21, 2026?	No @ 97%	No	0.001	0.001	Market closer
Will Elon Musk post 220-239 tweets from August 14 to August 21, 2026?	No @ 97%	No	0.001	0.001	Market closer
Will Elon Musk post 220-239 tweets from August 14 to August 21, 2026?	No @ 97%	No	0.001	0.001	Market closer
Will Elon Musk post 220-239 tweets from August 14 to August 21, 2026?	No @ 97%	No	0.001	0.001	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 56%	Flavio Cobolli	0.314	0.319	AI closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 57%	Flavio Cobolli	0.325	0.319	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 58%	Flavio Cobolli	0.335	0.319	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 59%	Flavio Cobolli	0.351	0.319	Market closer
Cincinnati Open: Tommy Paul vs Flavio Cobolli	Tommy Paul @ 62%	Flavio Cobolli	0.384	0.319	Market closer

Model variants — the self-correction experiment

Model	Forecasts	Brier	Market Brier	Beats market	Accuracy
blend-mkt50-v1	5604	0.141	0.128	50%	78%
blend-mkt75-v1	4946	0.130	0.127	50%	81%
blend-mkt90-v1	4789	0.127	0.126	50%	81%
gemini-2.5-flash	6336	0.182	0.132	54%	74%
gemini-2.5-flash-cal2	3789	0.135	0.126	45%	80%
gemini-3.7-flash	1337	0.172	0.154	44%	76%
gemini-3.7-flash-cal2	1334	0.155	0.154	45%	79%

blend-mkt50-v1 = the raw AI probability blended 50/50 with the market price at forecast time — a deterministic self-correction deployed after the calibration analysis showed mid-range overconfidence. If the blend's Brier beats the market's, the AI adds information beyond price.

Calibration — when the AI says X%, how often does it happen?

AI confidence	Forecasts	Realized	AI Brier	Market Brier
0-10%	8	100%	0.941	0.000
20-30%	1	100%	0.640	0.922
50-60%	419	57%	0.252	0.173
60-70%	1049	45%	0.262	0.213
70-80%	1193	59%	0.229	0.191
80-90%	976	59%	0.232	0.219
90-100%	2690	86%	0.098	0.037

Red = overconfident (events happened less often than the AI's stated probability).

AI vs market, by market line

Market line (winner's price at forecast time)	Forecasts	AI Brier	Market Brier	AI edge
mkt 95%+	1653	0.083	0.001	-0.083
mkt 80-95%	1625	0.079	0.016	-0.062
mkt 60-80%	1187	0.107	0.093	-0.014
mkt 40-60%	1058	0.274	0.231	-0.043
mkt <40% (upset)	813	0.578	0.559	-0.019

Methodology

Hourly AI forecasts are stored with the contemporaneous market price (append-only receipts). At resolution every receipt is scored: Brier = (1 - probability assigned to the eventual winner)^2, for both the AI and the market price captured at the same timestamp. Lower is better. Brier is computed only for 2-outcome markets; a winner requires one outcome priced >=0.95 at resolution. No receipt is dropped silently.

Unscorable receipts: 12034 ({'market_expired_unresolved': 11975, 'ambiguous_resolution': 15, 'unmatched_outcome_text': 44}) — counted, never silently dropped.

Raw JSON: /api/v1/scoreboard · Updated 2026-08-20T23:30:02.751746+00:00 · Market-implied odds, not financial advice.