How accurate are the ETH liquidation magnets?

For ETH, the walk-forward precision@N is 100.0% (Wilson 95% lower bound: 91.6%). The random-label baseline is 20.0%. Measurement span: 10d of out-of-sample data. If the symbol drops below our threshold, it's automatically disabled and we show you that.

Precision@N is the share of our top-N predicted magnet bands where a real liquidation actually materialized within the measured horizon window. It is computed out-of-sample (future data the model never saw), walk-forward (using only the calibration that was live at the time of the prediction). We publish the Wilson 95% confidence interval lower bound, the conservative floor, rather than the mean.

Why is the ETH data delayed 2 hours on this page?

The real-time tape and full graded market intelligence are paid-tier features inside the dashboard. The delay here (2 hours) lets you verify the data is real without giving away the live product. Real-time ETH data plus all tracked symbols is available with a free account.

Which exchanges does the ETH data cover?

The ETH liquidation tape aggregates data from Bybit, Binance, OKX, Kraken, BitMEX, and Deribit (where applicable), with OI and funding across up to 15 venues. Each venue's liquidation feed is independently reconciled every 5 minutes.

Does Hunter Killer publish accuracy scores for all coins?

We publish graded accuracy for every symbol we actively track, currently 25 perp instruments. Symbols below the precision threshold are auto-disabled and their state is shown honestly. We do not inflate scores by hiding failing symbols.

ETH Liquidation Levels & Accuracy

Cross-exchange ETH perp liquidation data — walk-forward graded, delayed 2 hours for the public snapshot — with the measured accuracy of our magnet predictions for ETH.

Magnet accuracy grade — ETH

Every grade is walk-forward, out-of-sample, with n + Wilson 95% CI. Below the threshold, the model is auto-disabled and that state is shown here honestly. Failing grades withhold the product.

Tracked & Graded Walk-forward magnet accuracy — ETH

100.0% precision@N (walk-forward)

91.6% Wilson 95% lower bound

42 effective independent samples

10d walk-forward measurement span

Metric: out-of-sample walk-forward precision@N (share of the top-N predicted magnet bands where a liquidation actually materialized within the horizon) · baseline 20.0% (random label) · Wilson 95% CI lower bound is the conservative floor we publish, never the mean · raw scored windows: 212 — the CI above is computed on the smaller de-overlapped effective sample of 42, so the interval never reads tighter than the evidence supports (larger raw counts on /proof are the same data before de-overlapping) · measurement began 2026-08-06.

Full scored ledger with all symbols →

Measured · serving ETH levels: 100.0% walk-forward precision [Wilson low 79.6%]

In plain English: ETH levels hit 100% of the time - checked over 356 windows across 15 independent days. The cautious number (95% floor): 79.6%. That is 6.8x better than guessing (the no-model mass-weighted baseline: 14.5%). proof → · how we measure →

100.0% walk-forward OOS precision

79.6% Wilson 95% lower bound

15 day clusters (the independence unit)

356 graded windows

+85.5pp excess vs 14.5% baseline · ci95 floor +65.1pp

Basis anchor-invariant (the phase-free fixed UTC grid) · the published floor is the artifact's own Wilson 95% lower bound computed on held-out day clusters — the DAY, not the correlated intra-day window, is the independent unit · the excess rides against the governed pool's mass-weighted trivial reference with its own ci95 floor, never a bare excess · measured live · calibration artifact stamped 2026-08-16T21:54:35Z · this block re-derives from the live artifact on every page render (5-minute page cache) · governance state: serving · descriptive history, not a forecast · full governed table →

Cascade autopsies — ETH

After major ETH liquidation cascades we publish what the map showed before the event, with the timestamped evidence attached. Descriptive records, never forecasts.

Liquidation snapshot — ETH · delayed 2 h

Observed cross-exchange liquidation tape for ETH perps, delayed 2 hours. Descriptive history only — not a forecast.

350 events · 24h

$3.0M total liq'd · 24h

$785K longs liq'd

$2.2M shorts liq'd

$575K largest single

Time	Exchange	Side	Price	Size
18:21 UTC	okx	SHORT	$1,885.56	$377K
18:13 UTC	gateio	LONG	$1,881.85	$13K
18:11 UTC	bybit	SHORT	$1,876.04	$6K
17:34 UTC	binance	LONG	$1,883.06	$6K
17:34 UTC	binance	LONG	$1,882.87	$7K
17:34 UTC	binance	LONG	$1,883.30	$6K
17:34 UTC	binance	LONG	$1,883.48	$7K
17:34 UTC	binance	LONG	$1,883.88	$10K
17:18 UTC	gateio	LONG	$1,885.67	$11K
16:41 UTC	binance	SHORT	$1,889.30	$27K

Last 10 events ≥$5K · older events available inside the dashboard

🕑 Delayed 2 hours · observed cross-exchange tape · descriptive history only · real-time data inside the Money Map

How accurate is this? — ETH FAQ

How accurate are the ETH liquidation magnets?

For ETH, the walk-forward precision@N is 100.0% (Wilson 95% lower bound: 91.6%). The random-label baseline is 20.0%. Measurement span: 10d of out-of-sample data. If the symbol drops below our threshold, it's automatically disabled and we show you that.
What is precision@N?

Precision@N is the share of our top-N predicted magnet bands where a real liquidation actually materialized within the measured horizon window. It's computed out-of-sample (future data the model never saw), walk-forward (using only the calibration that was live at the time of the prediction). We publish the Wilson 95% confidence interval lower bound — the conservative floor — rather than the mean, so the number is always the pessimistic honest estimate. Full methodology →
Why is the data delayed 2 hours on this page?

The real-time tape and full graded market intelligence are paid-tier features inside the dashboard. The delay here (2 hours) follows the same doctrine as our public liquidation teaser: it lets you verify the data is real without giving away the live product. Real-time ETH data plus all tracked symbols is available with a free account.
Which exchanges does the ETH data cover?

The ETH liquidation tape aggregates data from Bybit, Binance, OKX, Kraken, BitMEX, and Deribit (where applicable), with OI and funding across up to 15 venues. Each venue's liquidation feed is independently reconciled every 5 minutes — the per-venue reconciliation record is public on the proof page.
Does Hunter Killer publish accuracy scores for all coins?

We publish graded accuracy for every symbol we actively track — currently 25 perp instruments. Symbols below the precision threshold are auto-disabled and their state is shown honestly on these pages. We don't inflate scores by hiding failing symbols — the full ledger is on the proof page, wins and failures together.

See ETH in real time

Real-time ETH magnet map, position overlay, and graded alerts — free during beta.

Get the daily BTC & ETH liquidation levels, free

One short email a day: the strongest clusters above and below price. Delayed published data, no trade signals, unsubscribe instantly. No card, just an email.

Free trading tools

No-signup calculators for sizing and pricing ETH perp risk — the same tools linked from every page of the site.

Learn how to read this

Plain-English guides on liquidation heatmaps, maps, and the derivatives structure behind ETH.

Other tracked coins

Liquidation data delayed 2 h · walk-forward accuracy grades updated on calibration schedule · auto-disabled symbols shown honestly · no forecast claims · Full scored ledger