The venue trust board

We audit our own liquidation feed. Every venue below is checked on a ~5-minute cadence against an independent exchange reference, and the verdicts, sample sizes and confidence floors are published exactly as measured, including the venues still accruing history.

11 of 15 venues measured. Liquidation-event reconciliation currently covers 11 of 15 tracked venues; the venues listed here are the complete measured set. Cross-venue open interest and funding are aggregated across all 15 venues, and that broader coverage is not reconciliation-measured and is never implied to be.

Rolling 7-day window · reconciler last ran 2026-08-16T22:10:02Z · measured history: 66 days (2026-06-12 → 2026-08-16) · generated 2026-08-16T22:14:05Z

binance

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

99.93%

assessable share of runs

sample

50,367

assessable runs · rolling 7d

measured

66d

days with reconciliation rows

Independent check: open-interest-signature confidence-interval check; Binance's public REST liquidation endpoints are geo-blocked/discontinued from this VPS, and the OI-history endpoint is independent of the WebSocket broker.

bitfinex

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

100.00%

assessable share of runs

sample

50,400

assessable runs · rolling 7d

measured

22d

days with reconciliation rows

Independent check: independent reconciliation (method per venue).

bitget

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

100.00%

assessable share of runs

sample

50,400

assessable runs · rolling 7d

measured

22d

days with reconciliation rows

Independent check: independent reconciliation (method per venue).

bitmex

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

100.00%

assessable share of runs

sample

50,400

assessable runs · rolling 7d

measured

35d

days with reconciliation rows

Independent check: real-time integrity spot-check: our event-level WS tape is cross-checked against BitMEX's public REST active-liquidation snapshot.

bybit

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

99.96%

assessable share of runs

sample

50,382

assessable runs · rolling 7d

measured

66d

days with reconciliation rows

Independent check: open-interest-signature confidence-interval check; no clean public REST liquidation endpoint exists at any tier.

deribit

CLEAN → stable

reconcile accuracy

99.00%

Wilson 95% floor 98.88%

coverage

99.97%

assessable share of runs

sample

30,232

assessable runs · rolling 7d

measured

65d

days with reconciliation rows

Independent check: event-level REST reconciliation against the public trade tape (liquidation flag) plus feed-liveness verified silence.

dydx

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

99.85%

assessable share of runs

sample

50,326

assessable runs · rolling 7d

measured

22d

days with reconciliation rows

Independent check: independent reconciliation (method per venue).

gateio

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

99.97%

assessable share of runs

sample

50,387

assessable runs · rolling 7d

measured

26d

days with reconciliation rows

Independent check: independent reconciliation (method per venue).

hyperliquid

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

100.00%

assessable share of runs

sample

50,400

assessable runs · rolling 7d

measured

35d

days with reconciliation rows

Independent check: addressable non-HLP vault-fill capture with independent OI corroboration; the HLP-vault side is structurally unmeasurable via public APIs and carried as a coverage caveat, never as a failure.

kraken_fut

CLEAN → stable

reconcile accuracy

100.00%

Wilson 95% floor 99.99%

coverage

100.00%

assessable share of runs

sample

50,400

assessable runs · rolling 7d

measured

62d

days with reconciliation rows

Independent check: event-level public feed capture (type=liquidation) with independent OI corroboration via Kraken's public tickers REST (separate infra from the WS).

okx

CLEAN → stable

reconcile accuracy

99.95%

Wilson 95% floor 99.93%

coverage

100.00%

assessable share of runs

sample

50,400

assessable runs · rolling 7d

measured

66d

days with reconciliation rows

Independent check: direct REST liquidation-orders event-level reconciliation.

What this means

What reconciliation is

Liquidation feeds are easy to claim and hard to prove; exchange APIs are known to underreport liquidations. So we measure ours: on a ~5-minute cadence, each venue's ingested events are checked against an independent surface the venue's own WebSocket broker cannot influence (the method is disclosed per venue above). reconcile accuracy is the share of assessable checks that came back clean over the rolling 7-day window. Every score carries its sample size n and, where the sample supports it, a Wilson 95% lower bound: the floor, not just the headline.

What clean and drift mean

clean = the independent check corroborated what our feed ingested. drift = the check found a verified miss (for example events visible on the exchange's REST surface that our feed never ingested). Drift counts against the score. Windows we cannot verify at all (a reference endpoint down, a structurally unmeasurable surface) are not counted as failures; they show up as coverage gaps instead.

Why coverage matters

coverage is the share of checks that were assessable at all. A high accuracy on low coverage would mean "clean when we can check", so the two numbers are always published together. An accuracy figure without its coverage is not an honest figure.

What the trend arrow is (and is not)

trend compares the trailing 3 days' measured clean-rate with the preceding up-to-27-day baseline (assessable-weighted; ±1.0pp bands; minimum-sample guards). 'accruing' means the measured history is still too short to grade; it is shown honestly, never hidden. A trend describes the measured history only; it says nothing about the future.

The honesty contract

Every venue's ingested liquidation events are reconciled against an independent exchange reference surface on a ~5-minute cadence. reconcile_accuracy_pct = the share of assessable reconciliation runs whose verdict was clean over the rolling 7-day window; coverage_pct = the share of runs that were assessable at all. Unverifiable windows are coverage gaps, never counted as failures. This is a descriptive measurement of our own data quality, independently reconciled against exchange references; it is not a guarantee of future accuracy, and it is not a trading signal.

Machine-readable forms: /feeds/trust (this board as JSON) and /proof/feed_trust (the rolling verdicts). The track-record page for the prediction side of the product is /proof; metric definitions are frozen at /methodology; the Data API (including tiered trust history) is documented at /docs.