Accuracy
How well each venue's prices and our fair values predicted what happened. Every figure says how much data it rests on; early on, that is not much.
Loading…
How well each venue's prices and our fair values predicted what happened. Every figure says how much data it rests on; early on, that is not much.
Loading…
The sections below come from the Bitcoin tracker's minute-by-minute snapshots.
Kalshi settles on the CF Benchmarks BTC index (BRTI) averaged over the final 60 seconds, which isn't a free feed. Our fair values use Deribit's index instead. Here each settlement is compared with Deribit's index and Coinbase over the same minute (averaged from the snapshots at its start and end). A steady offset would shift every fair value near the close.
| close | Kalshi settlement | Deribit − it | Coinbase − it |
|---|---|---|---|
| Nov 21, 18:00 UTC | $59,954.82 | +$13.70 | +$10.07 |
| Nov 21, 17:00 UTC | $59,922.84 | −$0.64 | −$8.95 |
| Nov 21, 16:00 UTC | $59,799.42 | +$9.01 | −$2.72 |
| Nov 21, 15:00 UTC | $60,217.97 | +$13.71 | +$11.74 |
| Nov 21, 14:00 UTC | $60,510.94 | −$42.15 | −$50.61 |
| Nov 21, 13:00 UTC | $58,761.17 | −$2.65 | −$5.31 |
| Nov 21, 12:00 UTC | $58,106.17 | +$0.64 | +$0.96 |
| Nov 21, 11:00 UTC | $57,872.09 | −$17.20 | −$21.73 |
| Nov 21, 10:00 UTC | $57,923.26 | +$6.31 | −$4.77 |
| Nov 21, 09:00 UTC | $57,694.64 | +$20.55 | +$12.94 |
| Nov 21, 08:00 UTC | $57,884.78 | +$7.50 | −$1.38 |
| Nov 21, 07:00 UTC | $57,940.92 | +$26.38 | +$5.83 |
| Nov 21, 06:00 UTC | $55,713.04 | +$6.60 | +$0.46 |
| Nov 21, 05:00 UTC | $56,233.13 | −$5.47 | −$6.09 |
| Nov 21, 04:00 UTC | $56,617.54 | +$18.08 | +$13.81 |
| Nov 21, 03:00 UTC | $58,507.53 | −$4.80 | −$9.93 |
| Nov 21, 02:00 UTC | $58,897.86 | +$5.97 | −$4.50 |
| Nov 21, 01:00 UTC | $58,915.41 | −$2.38 | −$12.34 |
| Nov 21, 00:00 UTC | $58,790.92 | +$9.69 | +$2.72 |
| Nov 20, 23:00 UTC | $59,045.37 | −$1.39 | −$7.38 |
| Nov 20, 22:00 UTC | $58,928.88 | −$3.28 | −$35.27 |
| Nov 20, 20:00 UTC | $58,478.20 | – | – |
| Nov 20, 19:00 UTC | $58,397.20 | −$5.49 | −$7.68 |
| Nov 20, 18:00 UTC | $59,080.71 | −$4.62 | −$13.08 |
For every settled contract, one prediction at each of 5 minutes, 1 hour, 1 day and 1 week before it was decided (when it was listed that early): the venue's mid (halfway between bid and ask) and our fair value, scored against what happened. The Brier score is the average of (probability − result)², where the result is 1 if the contract paid out and 0 if not: lower is better, and always guessing 50¢ scores 0.25. Only contracts in play are scored (the mid or fair value between 3¢ and 97¢); the rest are all but decided and would flatter every score.
Contracts of one close are decided by the same number and win or lose together, so the sample size that counts is the number of closes. The last column is our fair value's score minus the venue's, on the same contracts: negative means ours was better. Its range comes from resampling whole closes 1,000 times and keeping the middle 90% of the results; a range that includes zero means no measurable difference. It needs at least 5 closes. Each horizon scores a different set of contracts (an hour out, more strikes are in play), so compare scores across a row, not down a column.
Fair values for Bitcoin and Ether come from Deribit's options, independent of the venue. Everywhere else they are a consensus of the venues' own order books, so they can only differ from the venue's mid by smoothing it or by drawing on the other venue: expect little difference there. Polymarket's economics markets keep trading after the figure is published, so they are timed from the paired Kalshi contract's close. Predictions are stored from 2026-11-17; Kalshi's Bitcoin contracts go back to 2026-11-14 from the tracker's history.
| venue, category | before decided | closes | contracts | venue Brier | fair method | venue Brier (same contracts) | fair Brier | fair − venue, 90% range |
|---|---|---|---|---|---|---|---|---|
| Kalshi crypto | 5 min | 1121 | 2,609 | 0.1313 | options | 0.1319 | 0.1320 | −0.0026 to +0.0031 |
| Kalshi crypto | 1 hour | 536 | 6,788 | 0.1059 | options | 0.1062 | 0.1044 | −0.0038 to −0.0001 |
| Kalshi crypto | 1 day | 18 | 339 | 0.0958 | options | 0.0958 | 0.0998 | +0.0005 to +0.0079 |
| Polymarket crypto | 5 min | 3 | 5 | 0.0668 | options | 0.0668 | 0.0296 | 3 closes |
| Polymarket crypto | 1 hour | 14 | 24 | 0.0498 | options | 0.0498 | 0.0394 | −0.0198 to −0.0038 |
| Polymarket crypto | 1 day | 12 | 46 | 0.1529 | options | 0.1529 | 0.1504 | −0.0099 to +0.0052 |
| Kalshi financials | 5 min | 52 | 607 | 0.1746 | book consensus | 0.1724 | 0.1647 | −0.0195 to +0.0040 |
| Kalshi financials | 1 hour | 52 | 2,291 | 0.1445 | book consensus | 0.1425 | 0.1409 | −0.0059 to +0.0014 |
| Kalshi financials | 1 day | 24 | 2,637 | 0.1823 | book consensus | 0.1692 | 0.1618 | −0.0169 to −0.0015 |
When a price said this much, how often did the contract pay out? From 1191 closes, every horizon pooled. A well-calibrated price sits on the dashed diagonal: of contracts priced around 30¢, about 30% pay out. Marker size grows with the number of contracts; hover for the numbers.
| price | Kalshi mid: n · said · paid out | Polymarket mid: n · said · paid out | Our fair value: n · said · paid out |
|---|---|---|---|
| 0–10¢ | 4181 · 5% · 5% | 20 · 5% · 5% | 4208 · 5% · 5% |
| 10–20¢ | 2110 · 14% · 11% | 11 · 16% · 9% | 2083 · 14% · 10% |
| 20–30¢ | 1156 · 25% · 17% | 8 · 24% · 0% | 1058 · 25% · 15% |
| 30–40¢ | 808 · 35% · 23% | 6 · 34% · 33% | 909 · 35% · 23% |
| 40–50¢ | 775 · 45% · 29% | 4 · 42% · 75% | 707 · 45% · 37% |
| 50–60¢ | 1326 · 53% · 29% | 4 · 56% · 50% | 764 · 55% · 40% |
| 60–70¢ | 784 · 65% · 55% | 4 · 67% · 75% | 746 · 65% · 55% |
| 70–80¢ | 1040 · 75% · 64% | 3 · 77% · 100% | 1027 · 75% · 67% |
| 80–90¢ | 1174 · 86% · 82% | 4 · 87% · 75% | 1249 · 86% · 82% |
| 90–100¢ | 1917 · 95% · 91% | 11 · 95% · 91% | 1862 · 95% · 91% |
A disagreement is a snapshot where our fair value is outside Kalshi's bid–ask: buying at Kalshi's ask or selling at its bid would, by our model, have been a good trade. Counted over every snapshot, so these are shares of time, not independent events.
Data note: until Nov 13, 23:45 UTC, Kalshi's bid and ask came from its market list, which we later measured trailing the live order book by up to a minute; since then they come from the live order book. Early figures overstate disagreement somewhat.
| time to close | snapshots | fair outside bid–ask | by (¢, when outside) | mean |gap| (¢) |
|---|---|---|---|---|
| under 1 h | 311,564 | 53% | 3.1 | 4.4 |
| 1–3 h | 12,774 | 73% | 2.0 | 2.1 |
| 3–6 h | 35,589 | 69% | 1.7 | 1.7 |
| 6–24 h | 68,249 | 66% | 1.8 | 2.0 |
| 24–48 h | 14,295 | 51% | 1.8 | 3.2 |
| hour of day (UTC) | snapshots | fair outside bid–ask | by (¢, when outside) | mean |gap| (¢) |
|---|---|---|---|---|
| 00:00 | 18,148 | 55% | 2.5 | 3.5 |
| 01:00 | 19,500 | 48% | 2.1 | 3.4 |
| 02:00 | 20,525 | 50% | 2.5 | 3.5 |
| 03:00 | 20,274 | 49% | 2.4 | 3.7 |
| 04:00 | 19,475 | 50% | 2.1 | 3.5 |
| 05:00 | 18,411 | 54% | 2.5 | 3.6 |
| 06:00 | 16,565 | 57% | 2.7 | 4.1 |
| 07:00 | 17,220 | 50% | 2.5 | 3.7 |
| 08:00 | 16,286 | 50% | 2.6 | 4.8 |
| 09:00 | 17,181 | 48% | 3.0 | 5.1 |
| 10:00 | 16,329 | 57% | 3.3 | 4.3 |
| 11:00 | 16,257 | 59% | 3.7 | 4.5 |
| 12:00 | 16,681 | 57% | 3.0 | 4.7 |
| 13:00 | 15,390 | 61% | 2.6 | 3.7 |
| 14:00 | 18,151 | 66% | 3.7 | 4.3 |
| 15:00 | 17,717 | 64% | 3.3 | 3.7 |
| 16:00 | 27,673 | 63% | 2.2 | 2.5 |
| 17:00 | 25,437 | 61% | 2.0 | 2.5 |
| 18:00 | 22,954 | 61% | 2.2 | 3.0 |
| 19:00 | 17,386 | 65% | 2.3 | 2.9 |
| 20:00 | 14,833 | 67% | 2.9 | 3.7 |
| 21:00 | 8,285 | 62% | 3.2 | 4.6 |
| 22:00 | 22,355 | 53% | 3.0 | 4.0 |
| 23:00 | 19,438 | 52% | 2.7 | 4.0 |
For contracts that have settled: one prediction each at 5, 15, 30 and 60 minutes before the close, from Kalshi's mid and from our fair value, scored against what happened. The Brier score is the average of (probability − result)², where the result is 1 or 0: lower is better, and always guessing 50¢ scores 0.25. Contracts in one close all settle on the same price, so the number of closes is the real sample size.
| minutes before close | closes | contracts | Kalshi Brier | fair Brier |
|---|---|---|---|---|
| 5 | 364 | 5868 | 0.0356 | 0.0352 |
| 15 | 364 | 3452 | 0.0884 | 0.0904 |
| 30 | 366 | 5132 | 0.0789 | 0.0794 |
| 60 | 17 | 154 | 0.1507 | 0.1361 |
Calibration, from 366 closes: when a price said this much, how often did the contract pay out? A well-calibrated price pays out about as often as it says.
| price | Kalshi: n | said | paid out | fair: n | said | paid out |
|---|---|---|---|---|---|---|
| 0–10¢ | 9010 | 4% | 2% | 8805 | 2% | 2% |
| 10–20¢ | 1671 | 14% | 12% | 1904 | 14% | 13% |
| 20–30¢ | 801 | 24% | 26% | 764 | 24% | 25% |
| 30–40¢ | 397 | 35% | 32% | 453 | 35% | 31% |
| 40–50¢ | 327 | 45% | 48% | 302 | 45% | 46% |
| 50–60¢ | 251 | 55% | 51% | 264 | 55% | 53% |
| 60–70¢ | 271 | 65% | 65% | 264 | 65% | 67% |
| 70–80¢ | 317 | 75% | 75% | 326 | 75% | 76% |
| 80–90¢ | 414 | 86% | 84% | 446 | 86% | 86% |
| 90–100¢ | 1147 | 95% | 95% | 1078 | 96% | 95% |