How Accurate Is Polymarket? And How Accurate Are We?

Published platform figures, attributed — and our own graded track record below.

201 / 289
BUY verdicts hit the outcome (April – September 2026)
96.7%
Polymarket self-reported, 4h before resolution
0.0843
Polymarket self-reported Brier score
5
Tier-1 wallets (Sep 2026)
Track record updates as tracked markets resolve · wallet tiers recomputed daily

OddsShift's AI verdicts hit the resolved outcome in 201 of 289 graded BUY calls (April – September 2026). How accurate is Polymarket? The platform reports 96.7% accuracy four hours before resolution and 90.4% one month out (Brier score 0.0843) — its own figures, not an independent audit. We publish a different number: our own calls, frozen when published and graded when the market resolves, misses included. The full list is in the Track Record section below. Last updated:

OddsShift does not audit platform accuracy at scale. What we publish is our own graded track record: each AI verdict is frozen at publication and compared with the resolved outcome — 201 of 289 BUY calls hit so far (April – September 2026), misses listed. Polymarket's self-reported accuracy is 96.7% four hours before resolution and 90.4% one month out (Brier 0.0843).

Polymarket Analytics Tools Compared our approach vs others

Several tools exist for tracking Polymarket activity. They differ in methodology, what they verify, and whether they publish a track record. Here's the factual side-by-side.

Tool Approach Published
Track Record
On-Chain
Smart Money
Cross-
Platform
Resolved
Hindsight
Free
OddsShift Tier-1 wallet agreement + AI verdict with reasoning 201/289 graded
PolymarketAnalytics.com Per-market analytics + trader leaderboard
Dune Analytics Raw on-chain SQL queries — DIY analytics Partial
Nansen Smart-money wallet labels (all chains) $$$

What sets OddsShift apart: we publish a graded track record (every AI verdict frozen at publication and graded against the resolved outcome — 201 of 289 BUY calls hit so far) and we surface resolved-market hindsight on every prediction and a weekly graded scoreboard — so you can audit our calls, not just trust them.

TRACK RECORD 201 of 289 BUY verdicts hit the outcome · April – September 2026
AI verdicts · accuracy
70%
201 of 289 BUY verdicts frozen before the close · 164 HOLD not counted
ROI, $1,000 per call
-2.3%
held to resolution · −$6,685 total on 289 calls
Open BUY verdicts
115
on live markets — they enter the record as they resolve
Market
Outcome
AI verdict
NO
NO ✓BUY NO @ 87¢ · +$149
NO
NO ✓BUY NO @ 92¢ · +$87
NO
NO ✓BUY NO @ 72¢ · +$389
NO
NO ✓BUY NO @ 92¢ · +$87
NO
NO ✓BUY NO @ 88¢ · +$136
NO
NO ✓BUY NO @ 88¢ · +$136
NO
NO ✓BUY NO @ 60¢ · +$667
NO
NO ✓BUY NO @ 90¢ · +$111
NO
NO ✓BUY NO @ 91¢ · +$99
NO
NO ✓BUY NO @ 83¢ · +$205
NO
NO ✓BUY NO @ 91¢ · +$99
NO
NO ✓BUY NO @ 77¢ · +$299
NO
YES ×BUY YES @ 73¢ · −$1,000
NO
YES ×BUY YES @ 64¢ · −$1,000
NO
YES ×BUY YES @ 64¢ · −$1,000
YES
NO ×BUY NO @ 71¢ · −$1,000
NO
NO ✓BUY NO @ 42¢ · +$1,381
NO
NO ✓BUY NO @ 92¢ · +$87
YES
NO ×BUY NO @ 22¢ · −$1,000
NO
NO ✓BUY NO @ 92¢ · +$87
YES
YES ✓BUY YES @ 72¢ · +$389
YES
YES ✓BUY YES @ 54¢ · +$852
NO
NO ✓BUY NO @ 63¢ · +$587
NO
NO ✓BUY NO @ 54¢ · +$852
NO
NO ✓BUY NO @ 92¢ · +$87
NO
NO ✓BUY NO @ 56¢ · +$786
NO
NO ✓BUY NO @ 82¢ · +$220

Latest 40 of 289 graded BUY verdicts — the full list, month by month with misses, is on the scoreboards.

289 BUY verdicts frozen before resolution · 164 HOLD verdicts (no entry) not counted · methodology: /accuracy · scoreboards by month → · all resolved markets

All verdicts were published before resolution and frozen at publication. See resolved market verdicts, smart-money wallets, or live signals.

How We Measure Our Accuracy methodology

1

Freeze the call

Every AI verdict (side, price, conviction tier) is written to a frozen record the moment it is published. It is never edited afterwards.

2

Wait for resolution

When Polymarket resolves the market, we record the winning outcome from the resolution data, not from the last traded price.

3

Grade it

Matched if the verdict's side is the winning side, missed otherwise. Both are published in the table above. Markets without a frozen verdict are not counted — nothing is scored in hindsight.

4

Wallet tiers

Tracked wallets are ranked on their own resolved-market record: tier-1 means 80% or better on at least 15 resolved markets; tier-2 is the next band. Tiers are recomputed daily from on-chain positions. See the wallets →

5

Independence

OddsShift is not affiliated with Polymarket or Kalshi. Our data comes from public APIs and on-chain records, and platform accuracy figures quoted on this page are attributed to their source.

Frequently Asked Questions accuracy & methodology

How accurate is Polymarket?

Polymarket's own accuracy page reports 96.7% accuracy four hours before resolution and 90.4% one month out, with a Brier score of 0.0843 across its resolved markets. Those are the platform's figures, not an independent audit. Accuracy varies by category and by how close to resolution you look.

Are prediction markets accurate?

Broadly yes. Academic work (Wharton, the Iowa Electronic Markets, CEPR) finds prediction markets competitive with or better than polls and expert panels, and accuracy improves sharply as resolution approaches. Prices are calibrated probabilities, not certainties: events priced at 70% should happen roughly 70% of the time.

Which is more accurate, Polymarket or Kalshi?

There is no independent large-sample answer yet. Both platforms publish their own calibration. OddsShift records both prices where the same event is listed on both, but our matched sample is still too small to support category-level comparisons, so we do not publish them.

What is a Brier score?

The mean squared error between a predicted probability and the binary outcome (0 or 1). A score of 0 is perfect, 0.25 is a coin flip, 1 is always wrong. Below 0.1 is generally considered excellent.

What is OddsShift's own track record?

We freeze each AI verdict when it is published and grade it against the resolved outcome: BUY YES / BUY NO with $1,000 per call held to resolution, HOLD shown but not counted. The current count, hit rate and ROI are in the Track Record block on this page and grow as tracked markets resolve; the full list, including the misses, is on the monthly scoreboards.

How does OddsShift measure its accuracy?

Each verdict is frozen at publication (side and price) and later compared with the market's resolved outcome. Nothing is re-scored after the fact. Markets without a frozen verdict are not counted, and misses are published alongside hits.

What is a tier-1 wallet?

A tracked on-chain Polymarket wallet with 80% or better accuracy on at least 15 resolved markets. Tiers are recomputed daily from on-chain positions; 5 wallets qualify as tier-1 and 16 as tier-2 as of September 2026.

Is a prediction market basically gambling?

Both involve risk, but prediction markets aggregate dispersed information into calibrated probabilities and are used as forecasting tools; regulated exchanges such as Kalshi are supervised by the CFTC. Whether a given trade is a bet or an informed position depends on the edge behind it.

Get a Telegram alert when a tracked wallet trades

Subscribe — $29/month