Methods and error rates
Every method behind the record, and how often it is wrong
Each field on a Market Record comes from a method listed here, with how it was validated and its error rate where one is measured. A method we have not measured says not yet measured.
| Method | Used by | Validation | Error rate | Last audit | Doc |
|---|---|---|---|---|---|
| Fees | Cost | Every fee comes from fees.py, pinned to real fills in tests/test_fees.py. Each of the account's real Kalshi fills is repriced with it and compared with the fee the exchange charged, to $0.0001; the rate is the share that did not match. Not yet measured. | Not yet measured | Not audited yet | docs/kb/01-market-mechanics.md |
| Settlement audit | RuleRecord | Each settled 15-minute window is reproduced from the published rule (the 60-second average of the index) and compared with what the exchange paid; the rate is the share that disagreed. | 0.0002 | Nov 13, 2026 | docs/kb/20-settlement-rule.md |
| Calibration | Record | How often a market quoted at a price settled YES, by venue, family and time to close, with the Brier score against the base rate and the markets in each bin. A measurement, not a test, so no error rate applies. | Not yet measured | Not audited yet | docs/kb/42-centring-and-the-two-calibrations.md |
| Equivalence grades | Twins | Each pair of markets on different venues is graded A to D from both rules texts and the settlements behind it; A and B need settlement history. Each quarter 100 graded pairs are drawn at random and checked by hand, and the rate is the share graded wrong. Not yet measured. | Not yet measured | Not audited yet | docs/kb/220-equivalence-grades.md |
| Integrity detector: ahead | Integrity | Validated on planted cases in copies of real windows; the rate is the share of matched clean windows the detector flagged (70 measured). Definition: clean windows flagged: integrity_ahead.study against the family's placebo events. | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: book | Integrity | Looks for: Orders pulled before they can trade, layering, quote stuffing, momentum ignition. Measured over the last days of real tape by integrity_measure; the rate is the detector's own false-positive rate (65075 measured). Definition: spoof flags per market-hour on synthetic normal order flow (calibration). | 0.0149 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: chain | Integrity | Measured over the last days of real tape by integrity_measure; the rate is the detector's own false-positive rate (274517 measured). Definition: chance rate stated on each flag (placebo wallets for new-wallet concentration). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: consistency | Integrity | Looks for: Mutually exclusive outcomes that do not sum to one; ladder strikes out of order. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: consistency_pulled | Integrity | Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: consistency_xv | Integrity | Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: crossbook | Integrity | Looks for: Orders pulled before they can trade, layering, quote stuffing, momentum ignition. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: data | Integrity | Looks for: Payouts that disagree with the exchange's own published figures or the stated index. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: data_pulled | Integrity | Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: echo | Integrity | Looks for: Price moves before the first public report of the news. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: embargo | Integrity | Looks for: Price moves before the first public report of the news. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: entity_graph | Integrity | Looks for: Wash trading, circular flow, clusters trading with themselves. Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: a group of four co-buying one long shot, a privately funded group of four co-buying one long shot). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: family-settlement | Integrity | Looks for: Payouts that disagree with the exchange's own published figures or the stated index. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: flow | Integrity | Looks for: Price moves before the first public report of the news. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: index_push | Integrity | Looks for: Payouts that disagree with the exchange's own published figures or the stated index. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: information | Integrity | Looks for: Price moves before the first public report of the news. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_ahead | Integrity | Measured on the clean copies of the evasion study's planted cases (26 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: drift ahead of an event). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_book | Integrity | Measured on the clean copies of the evasion study's planted cases (54 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: cancel after fill, place and cancel near the touch, support behind a fast move pulled once traded into, support behind a fast move pulled once traded into, the shape before the flag). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_chain | Integrity | Measured on the clean copies of the evasion study's planted cases (24 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: new-wallet concentration before the close, self-dealing cluster). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_crossbook | Integrity | Measured on the clean copies of the evasion study's planted cases (30 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: one wallet cluster holds most of an outcome, one-sided buying in the last minutes before the close, order on one strike, trade on the next). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_echo | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: the anchor's trade echoed on another venue). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_embargo | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: trading inside a release's lockup window). | 0.1667 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_flow | Integrity | Measured on the clean copies of the evasion study's planted cases (240 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: long-shot buying before the close). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_index_push | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: one constituent pushes the settlement index). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_leadlag | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: global Polymarket leads the other venues). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_pseudo | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: one actor's slices before a release). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_resolution | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: voters linked to the market's holders cast the deciding vote). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_rfq_leak | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: leg buying between an RFQ and its block). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_source_pages | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: a source page edited near the rule). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_sports | Integrity | Measured on the clean copies of the evasion study's planted cases (6 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: the venue's line leads the books). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_stream | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: a fresh wallet's long shot during an open market). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: integrity_xv_frontrun | Integrity | Measured on the clean copies of the evasion study's planted cases (12 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: same-way flow on one venue before a large print on the other). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: leadlag | Integrity | Looks for: Price moves before the first public report of the news. Validated on planted cases in copies of real windows; the rate is the share of matched clean windows the detector flagged (36 measured). Definition: clean windows flagged: integrity_leadlag.study_group; a hit is a candidate at the planted moment. | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: market_risk | Integrity | Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: onchain | Integrity | Looks for: Wash trading, circular flow, clusters trading with themselves. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: pseudo | Integrity | Looks for: Price moves before the first public report of the news. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: registry-chain | Integrity | Looks for: Wash trading, circular flow, clusters trading with themselves. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: resolution | Integrity | Looks for: Rule text changed after listing; disputed resolutions; resolution voters who hold the market. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: rfqleak | Integrity | Looks for: Orders pulled before they can trade, layering, quote stuffing, momentum ignition. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: settle-agreement | Integrity | Looks for: Payouts that disagree with the exchange's own published figures or the stated index. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: settlement | Integrity | Looks for: Payouts that disagree with the exchange's own published figures or the stated index. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: settlement_pulled | Integrity | Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: source | Integrity | Looks for: Rule text changed after listing; disputed resolutions; resolution voters who hold the market. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: source_pages | Integrity | Looks for: Rule text changed after listing; disputed resolutions; resolution voters who hold the market. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: source_pulled | Integrity | Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: sports | Integrity | Looks for: Price moves before the first public report of the news. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: stream | Integrity | Looks for: Orders pulled before they can trade, layering, quote stuffing, momentum ignition. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: wallet_features | Integrity | Measured on the clean copies of the evasion study's planted cases (6 measured). Definition: clean copies flagged on the synthetic baseline (integrity_evasion, plain pattern: long-shot wallet). | 0.0000 | Nov 7, 2026 | docs/kb/170-detector-validation.md |
| Integrity detector: wallets | Integrity | Looks for: Wash trading, circular flow, clusters trading with themselves. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Integrity detector: xvfrontrun | Integrity | Looks for: Price moves before the first public report of the news. Not yet measured. | Not yet measured | Not audited yet | docs/kb/170-detector-validation.md |
| Toxicity score | Integrity | The chance a fill is marked out against the passive side by more than the half-spread at 60 seconds, from a logistic regression fitted on the explore segment, chosen once on the test segment and reported on forward tape only. The rate is the false-alarm rate on placebo windows: forward fills whose markout stayed inside the half-spread that the score still put in the informed band (0.75 or more). Trial count 1. Model v1: fitted on days before 2026-09-22, tested once on 2026-09-22 to 2026-10-01, reported from 2026-10-02. On those days 4 of its 7 feature families had no data (open integrity flags, skilled-wallet flow on the global polymarket twin (aggregate), minutes since the first public report, cross-venue echo in the last minute), because the jobs that feed them started in October. Their weights are zero and they never move this score, which in effect scores on order imbalance and price band. Not chosen yet. Not yet measured: 0 of 1,000 forward fills so far. | Not yet measured | Not audited yet | docs/kb/224-toxicity-score.md |
Registry built Nov 13, 2026, 09:46 AM ET.