On Manifold, a binary market can end without Yes or No
An audit of 500 recent play-money resolutions found 25 cancellations and eight probability-based outcomes. A Yes/No-only dataset leaves them out.
Prediction Contracts editorial · Updated · 3 min read
A yes-or-no question doesn’t always get a Yes or No answer. Of the 500 most recently resolved binary markets returned by Manifold’s public search on October 4, 2026, 33 (6.6%) ended another way. These are play-money markets: Mana has no cash redemption value.
How 500 markets ended
Manifold’s API documents four resolution labels for binary markets. Most of the sample resolved Yes or No.
| Resolution | Markets | Share |
|---|---|---|
| Yes | 189 | 37.8% |
| No | 278 | 55.6% |
| Cancelled (CANCEL) | 25 | 5.0% |
| Probability payout (MKT) | 8 | 1.6% |
| Total | 500 | 100% |
A Yes/No-only filter keeps 467 markets and drops 33.
Source: calculated results
A cancelled market gets cancellation/refund treatment. An MKT market pays out at a resolution probability; all eight here used fractions from 0.26 to 0.72. Neither label is automatically an error, and neither says why a market ended that way.
What a Yes/No filter leaves out
Some accuracy studies explicitly keep only YES and NO labels. Expertium’s open-source calibration collector, for example, excludes MKT and CANCEL as a stated rule.
Applied to our sample, that rule drops 33 markets, including every market from 11 of the 238 creator IDs. That is not Expertium’s historical exclusion rate, and it can’t show whether leaving these markets out raises or lowers any accuracy score.
A few creators move the number
One creator ID accounts for 11 of the 25 cancellations. The busiest, with 35 markets, had none outside Yes/No. Counting the sample other ways moves the share:
| How the sample is counted | Outside Yes/No |
|---|---|
| Each market once (headline) | 6.6% (33 of 500) |
| Without the creator ID behind 11 cancellations | 4.5% (22 of 489) |
| Without the busiest creator ID | 7.1% (33 of 465) |
| Each creator ID weighted equally | 5.6% (average of creator shares) |
| Weighted by lifetime Mana volume | 4.0% of reported volume |
Five descriptions of one sample, not a margin of error or a likely range for Manifold. Mana volume is play money, not dollars.
Source: creator sensitivity checks
Creator IDs only group markets. They aren’t verified identities, or necessarily the accounts that resolved them.
What this does and doesn’t show
This is one bounded snapshot of visible search results, not a census or random sample of Manifold, and it says nothing about cash platforms such as Kalshi or Polymarket. It measures which records a label filter removes, not forecast accuracy.
If you publish a study built on Yes/No-filtered data, report how many CANCEL and MKT records, and how many creators, the filter removed.
Methods, evidence and how to verify
Collection. We made one request to Manifold’s public API at 03:36 UTC on October 4, 2026, for 500 binary markets sorted by most recent resolution. All 500 records were unique, resolved and explicitly marked Mana, arrived in descending resolution order and passed every eligibility check. Recorded resolutions span September 19, 13:21 UTC, to October 4, 02:47 UTC.
Protocol. The plan was fixed before collection, but this was exploratory research, not a preregistered experiment. A two-record feasibility check, which showed two cancellations by one creator, helped motivate the creator checks.
Notes. The API’s probability for a resolved market is read after the fact, not a forecast at a fixed horizon, so we scored no forecasts. Cancelled markets had a median age of 312 days from creation to resolution, versus 30 for Yes and 33 for No. That describes this sample; it is not a settlement delay or evidence that older markets are more likely to be cancelled.
Verify it yourself. Read the methods and verification instructions, then check the market-level evidence and calculated results with the offline verification script. Creator IDs are replaced with consistent codes. A fresh API request would return a new sample.