targeting.ai lab

Paper Bets · a live experiment

Two AI judges read the news and bet play money against the crowd.

Every few hours each judge reads the newest headlines about a real prediction market and says how likely a yes is. The crowd's price is the bar to beat. When the market resolves, everyone is scored.

  • The crowd the market's own price
  • Jev TypeSafe's judge model
  • Decisions OpenAI's judge model
markets followed
11
rounds of reading
44
headlines read
178
opinions given
220
resolved
0

Open right now

Where the judges and the crowd stand, soonest to close first.

closes in 1 day

Next Astra Model (6.1+) released by November 13, 2026?

The crowdmarket price 17%
JevAI judge 9%
DecisionsAI judge 3%
0% 50% 100% 4 new headlines 1 new headline 9 new headlines 8 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

22 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 17% · Jev 9% · Decisions 3%

Nov 11, 21:00 UTC crowd 67% · Jev 11% · Decisions 9%

Nov 11, 20:11 UTC crowd 65% · Jev 12% · Decisions 13%

Nov 11, 16:01 UTC crowd 60% · Jev 10% · Decisions 8%

closes in 1 day

Will Trump's approval rating be between 39.0 and 39.4 on November 13, 2026?

The crowdmarket price 20%
JevAI judge 10%
DecisionsAI judge 3%
0% 50% 100% 1 new headline 10 new headlines 10 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

21 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 20% · Jev 10% · Decisions 3%

Nov 11, 21:00 UTC crowd 17% · Jev 10% · Decisions 5%

Nov 11, 20:11 UTC crowd 17% · Jev 16% · Decisions 18%

Nov 11, 16:01 UTC crowd 16% · Jev 23% · Decisions 49%

No new headline in this round.

closes in 3 days Decisions is 47 points below the crowd

US x Iran Effective Ceasefire by November 15?

The crowdmarket price 54%
JevAI judge 43%
DecisionsAI judge 7%
0% 50% 100% 1 new headline 4 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

5 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 54% · Jev 43% · Decisions 7%

Nov 11, 21:00 UTC crowd 55% · Jev 66% · Decisions 42%

No new headline in this round.

Nov 11, 20:11 UTC crowd 55% · Jev 64% · Decisions 57%

No new headline in this round.

Nov 11, 16:01 UTC crowd 55% · Jev 63% · Decisions 57%

closes in 3 days

Will Zoran Janković win the 2026 Ljubljana mayoral election?

The crowdmarket price 59%
JevAI judge 54%
DecisionsAI judge 43%
0% 50% 100% 5 new headlines 7 new headlines 1 new headline Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

13 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 59% · Jev 54% · Decisions 43%

Nov 11, 21:00 UTC crowd 59% · Jev 63% · Decisions 66%

Nov 11, 20:11 UTC crowd 59% · Jev 41% · Decisions 52%

Nov 11, 16:01 UTC crowd 58% · Jev 30% · Decisions 71%

No new headline in this round.

closes in 3 days

Will GPT Astra 6.1+ be released between November 9 and November 15, 2026?

The crowdmarket price 19%
JevAI judge 7%
DecisionsAI judge 7%
0% 50% 100% 3 new headlines 4 new headlines 10 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

17 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 19% · Jev 7% · Decisions 7%

Nov 11, 21:00 UTC crowd 68% · Jev 9% · Decisions 6%

Nov 11, 20:11 UTC crowd 68% · Jev 7% · Decisions 12%

Nov 11, 16:01 UTC crowd 67% · Jev 8% · Decisions 5%

No new headline in this round.

closes in 3 days

Next GPT Sol (6.2+) released by November 15, 2026?

The crowdmarket price 19%
JevAI judge 14%
DecisionsAI judge 7%
0% 50% 100% 2 new headlines 4 new headlines 3 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

9 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 19% · Jev 14% · Decisions 7%

Nov 11, 21:00 UTC crowd 69% · Jev 19% · Decisions 27%

No new headline in this round.

Nov 11, 20:11 UTC crowd 66% · Jev 19% · Decisions 27%

Nov 11, 16:01 UTC crowd 70% · Jev 33% · Decisions 56%

closes in 5 days

Will Helen Zille be the next mayor of Johannesburg?

The crowdmarket price 45%
JevAI judge 42%
DecisionsAI judge 49%
0% 50% 100% 3 new headlines 1 new headline 8 new headlines 10 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

22 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 45% · Jev 42% · Decisions 49%

Nov 11, 21:00 UTC crowd 38% · Jev 42% · Decisions 38%

Nov 11, 20:11 UTC crowd 29% · Jev 45% · Decisions 40%

Nov 11, 16:01 UTC crowd 51% · Jev 48% · Decisions 68%

closes in 7 days Decisions is 85 points below the crowd

Will Anthropic IPO by November 19, 2026?

The crowdmarket price 90%
JevAI judge 53%
DecisionsAI judge 5%
0% 50% 100% 5 new headlines 9 new headlines 10 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

24 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 90% · Jev 53% · Decisions 5%

Nov 11, 21:00 UTC crowd 91% · Jev 64% · Decisions 89%

No new headline in this round.

Nov 11, 20:11 UTC crowd 90% · Jev 62% · Decisions 89%

Nov 11, 16:01 UTC crowd 87% · Jev 64% · Decisions 96%

closes in 7 days Jev is 50 points below the crowd

Will Anthropic IPO on November 19, 2026?

The crowdmarket price 86%
JevAI judge 36%
DecisionsAI judge 94%
0% 50% 100% 6 new headlines 9 new headlines 10 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

25 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 86% · Jev 36% · Decisions 94%

Nov 11, 21:00 UTC crowd 87% · Jev 37% · Decisions 80%

No new headline in this round.

Nov 11, 20:11 UTC crowd 86% · Jev 33% · Decisions 80%

Nov 11, 16:01 UTC crowd 85% · Jev 64% · Decisions 89%

closes in 10 days Decisions is 87 points below the crowd

Israel x Iran ceasefire continues through November 22?

The crowdmarket price 92%
JevAI judge 56%
DecisionsAI judge 5%
0% 50% 100% 3 new headlines 2 new headlines 5 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

10 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 92% · Jev 56% · Decisions 5%

Nov 11, 21:00 UTC crowd 92% · Jev 63% · Decisions 17%

No new headline in this round.

Nov 11, 20:11 UTC crowd 92% · Jev 64% · Decisions 17%

Nov 11, 16:01 UTC crowd 91% · Jev 59% · Decisions 24%

closes in 13 days

Will Core PCE MoM be 0.3% in October?

The crowdmarket price 53%
JevAI judge 43%
DecisionsAI judge 57%
0% 50% 100% 1 new headline 2 new headlines 7 new headlines Nov 11, 16:01 UTC Nov 12, 06:56 UTC

Grey dots under the lines: new headlines arrived in that round.

10 headlines read in 4 rounds

Nov 12, 06:56 UTC crowd 53% · Jev 43% · Decisions 57%

Nov 11, 21:00 UTC crowd 53% · Jev 47% · Decisions 56%

No new headline in this round.

Nov 11, 20:11 UTC crowd 55% · Jev 46% · Decisions 56%

Nov 11, 16:01 UTC crowd 49% · Jev 40% · Decisions 50%

The score

Scored on what each one said last before the market closed. Brier score: 0 is perfect, 0.25 is a coin flip.

No market we follow has resolved yet. The next one closes in 1 day.

The dress rehearsal

Before going live, both judges replayed 37 markets that had already resolved (19 yes, 18 no), reading the headlines of 14 days, 7 days and 1 day before each close.

Jev

no headlines 0.272
two weeks before the close 0.172
the day before the close 0.161

Brier score, shorter is better. The line marks a coin flip.

Decisions

no headlines 0.297
two weeks before the close 0.178
the day before the close 0.195

Brier score, shorter is better. The line marks a coin flip.

  • Any evidence beats none. Without headlines both judges did worse than a coin.
  • Newer evidence on top of the first batch: not detectable. Jev -0.025, Decisions +0.001 in Brier score, both inside their own margin of error.
  • A rehearsal, not a result. Search dates are rough and a page edited later can carry the outcome. The live markets above are the clean test.

How it works

  1. Pick. News-driven yes/no markets on Polymarket that close within two weeks. One market per event. No sports, no price ladders.
  2. Read. Every six hours, hourly on the last day, the newest headlines about the question are collected: a title and one sentence each. Up to three of the newest articles are also read in full. Pages that quote betting odds are thrown out.
  3. Judge. Both judges get the same question, rules and headlines and answer with one number: the probability of yes. A second answer, on the headlines plus the full articles, is kept beside it. Neither sees the crowd's price.
  4. Settle. When the market resolves, everyone is scored on what they said last. The purse is play money staked on each judge's own opinion.