Published levels are logged before their outcomes are scored.
Published levels are logged before their outcomes are scored. The record includes misses and simple baselines, including when a baseline does as well or better. Only evidence cleared for public release is shown.
Complete chronological hit and miss rows for the current provisional window. Released headline figures stay separate.
Loading the complete scored row record…
Misses included: reached and missed levels both stay in the record. Dated record status
Loading the recorded dollars
Data age unavailable. A dated snapshot of recorded dollars is not available.
Public data is at least 2 hours behind. Missing evidence is never counted as zero.
Retrospective observations. Timestamp submitted, awaiting verification.
The first complete 15 minute bar after a wall was served that spans its price opens a one hour window. Recorded dollars use the frozen wall band and position side across the registered venues, with repeats excluded. Each event belongs to at most one episode. Both dollar totals include only complete, attested episodes. Gaps remain unresolved. These observations do not establish causation.
A liquidation map estimates where leveraged positions could be forced closed. Published-level outcomes and the historical band-containment backtest answer different questions. The backtest measures whether price stayed inside a band, rather than whether a published level was reached.
Loading the scored record. See the dated record status.
More records as they score.
The sections below are the longer record behind those lines: the symbols we switched off, the band containment track, the deep-magnet capture test and its baselines, the per-symbol ledger and the daily cryptographic seal. Every figure carries its sample size and a confidence bound.
Want the raw table? Download the per-symbol ledger (CSV) and check it yourself.
| Major | 1D band (±) | Held (up) | Held (down) | 95% low (worst side) | Disjoint days |
|---|---|---|---|---|---|
| Loading… | |||||
| Date | Hypothesis | Why it was killed |
|---|---|---|
| Loading… | ||
| Symbol | Grade | precision@N (day-pooled) | day-cluster 95% low | Disjoint days | Window |
|---|---|---|---|---|---|
| Loading… | |||||
Publishing what we auto-disabled is the whole point: survivorship editing would be the dishonest move. precision@N is the near-spot (easy-bar) measure; the honest confidence floor is the day-cluster lower bound from the pre-registered track above, merged in per symbol. Hover any header for its meaning.
| Symbol | Status | Grade | precision@N (all) | last 7d | day-cluster 95% low | disjoint days | events (n) | tracked since |
|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||
Most vendors show reached levels and quietly delete whatever stopped working. This page is built to make that impossible here: when a symbol's live precision falls below our auto-disable bar, its published levels switch off automatically, and the symbol stays on this page, marked AUTO-DISABLED, with its measured history intact. Nothing is removed from the ledger; there is no survivorship editing. The box below is the live state of that circuit, read from the same data as the table above.
Every day, the complete record ledger and outcome journals behind this page are canonicalized and hashed with SHA-256, and each day's seal includes the previous day's seal: a hash chain, the standard construction behind append-only audit logs.
Why it matters: a track record the vendor can edit after the fact is marketing, not evidence. With the chain, any retroactive edit to the history (one character, by anyone, including us) breaks every seal written after it, and anyone holding a copy of the chain can prove it. We cannot quietly rewrite a bad week.
Check it yourself: the full chain is public as machine-readable JSON at /proof/anchors, and the payload itself embeds the exact step-by-step verification recipe (canonicalization, digest, and chain rule), so re-checking every seal needs nothing but a copy of the chain and a SHA-256 implementation. The chain lets anyone check whether the retained history has changed.
"Near-spot" means placing a liquidation close to the current price. Many liquidations happen there, so even a trivial band can score well. These figures are the near-spot measure, published in full for completeness; they are not proof of skill, and we never present them as the edge. The hard tests are containment and deep-magnet capture above.
| Within | Reaches a magnet ~1% away | Sample (n) |
|---|---|---|
| Loading… | ||
Price zones where leveraged positions are forced to liquidate (the "magnets" shown in the product), and a historical 1-day containment band.
For each major we publish a 1-day band and score, on the strictly-later historical segment, the share of disjoint days price closed inside it. We report the band width and the containment together, plus the conservative worse-of-both-sides Wilson 95% lower bound, never the better side. Because a wider band trivially contains more, the honest read is width + containment jointly. It is a descriptive containment record, never a target or forecast.
Of the liquidation events landing more than 3% from spot, what fraction landed in one of our top-ranked published bands, scored historical and de-overlapped. We publish it next to its honest comparator: the same capture rate for a trivial near-boundary bet scored on the same events, shown as a separate plain comparison (never as a multiple), including the cells where that bet matches or beats the model. The 95% lower bound is computed on the independent-effective sample, which is smaller than the raw event count by design.
Raw scoring windows overlap, so a Wilson CI on the raw count overstates certainty. The pre-registered track scores each completed UTC day once, resampling whole disjoint days as the independent unit: a day-cluster bootstrap 95% lower bound that tightens with genuine historical days. That is the confidence floor shown in the day-cluster track. The near-spot companion is published separately.
Near-spot precision@N is a historical companion measure. See the near-spot table for the stored rates, sample sizes and dates. See the containment table for its disjoint-day counts and confidence bounds. No paired excess is quoted here.
See the deep-magnet capture cards for the stored baseline comparisons, confidence bounds and effective sample sizes. Each card states the snapshot date. A historical capture measurement is not a forecast.
The liquidation event feed itself is reconciled every 5 minutes against sources we don't control: open-interest signatures on Bybit and Binance, direct REST liquidation records on OKX, a structural expected-zero check on Deribit, and event-level capture of Kraken Futures' public liquidation feed cross-corroborated by Kraken's independent public open-interest REST. Verification ran 7 days in shadow before any badge went live, and has run continuously since 2026-05-10.
When a symbol's live precision drops below our auto-disable bar, its published levels are switched off automatically and it appears here as auto-disabled until it re-qualifies. Nothing is removed from the ledger; there is no survivorship editing.
A historical measurement, not a forecast. Past precision does not guarantee future precision, and nothing here is financial advice.
The frozen, versioned definition of every metric on this page lives at Full frozen metric contract to /methodology.