Unusual Activity

Every coin is measured against itself

For each coin and each metric the radar takes the last closed 1d candle (UTC) and compares it with the 30 closed days before it — that coin's own baseline, so a $37B market and a $30M market are judged on the same footing. A baseline needs at least 24 of 30 days of data or the signal is dropped rather than guessed. Lookback for percentiles and "last time this extreme" is 200 days. The day scored on this page is 2026-10-02 (UTC) — the candle still running is never scored.

The live 24-hour frame

The Live 24h toggle swaps the daily candle for a rolling 24-hour window that ends on the last closed hour and is rebuilt every hour. Each window is compared with the window ending at the same hour of each of the previous 30 days — 14:00 against 14:00 — so the normal Asia/Europe/US session rhythm and funding settlement times do not register as anomalies. Funding in this frame is the latest hourly close of the OI-weighted rate, not a 24-hour average: an early warning is only useful before it is diluted. The gates, the robust statistics and the ranking are identical to the daily frame; percentiles run over the last 35 same-hour windows.

What is in the scan, and what is not

311 of the 537 perp coins tracked clear the $10M open-interest minimum where a 30-day baseline still means something. This snapshot scans all 311 of them (smallest book in the scan: $10.1M). Coins below $10M are left out on purpose: on a $2M book a single trade clears 3σ. Non-crypto instruments the site tracks (equities, metals, indices) are out of scope.

Unusual is not enough — it has to be material

A 7σ funding move that shifts the annualised cost by 3 points is noise. Each signal also has to clear a size floor:

• Taker buy/sell: ratio at least 0.15 points from the median
• Liquidations: at least $200K and 1.6× the 30-day median
• Open interest: day-over-day move of at least 3%
• Volume spike: at least $2M and 1.6× the median
• Volume dry-up: 0.6× the median or less, on a $2M+ baseline. On a Saturday or Sunday it must also undercut the coin's own weekend median — the normal weekend lull is not an anomaly
• Funding: at least 10 annualised points away from the baseline

Why not plain standard deviations

Volume, liquidations and open interest are multiplicative and right-skewed, so they are scored on a log scale. Spread is measured with robust statistics — the median deviation, backstopped by the width of the middle 80% of the baseline — so the one outlier we are trying to detect cannot inflate its own yardstick, and quiet stretches with fat tails do not produce fake extremes. σ is capped at 10; past that the number stops meaning anything, so the percentile is the honest read. Ranking still uses the raw, uncapped σ to break ties — the detail panel shows the raw value.

How the ranking works

The radar score starts from the coin's strongest σ (capped at 10), with smaller bonuses for every additional signal clearing 2σ and for each cross-metric tag. It rewards coins where several independent metrics moved together, not just one loud column.

Cross-metric tags

One column jumping happens every day; the tags mark days when columns disagree. OI up · price down — new positions opening into a falling price, usually shorts pressing in or longs averaging down. OI up · price up — new longs funding the move rather than short covering. Forced deleveraging — open interest flushing out while liquidations spike, positions closed by the engine rather than by choice.

What it is not

The radar score is a heuristic ranking, not a probability, and not a forecast. Long/short here is the taker buy/sell volume ratio — not exchange account ratios. Nothing on this page is investment advice.

Source: Coinfuty's own aggregated exchange data — the same series the site's charts read.