Squeeze Radar shows you what the candle chart hides: where the resting money is, which of it is real, who is being aggressive, and the measured probability that a move in progress extends. One principle governs everything here: we display what we measure, with the sample size, and we never claim direction — we measured direction predictability at these horizons repeatedly, and it is not there.
The main canvas scrolls right-to-left at ~1px/60ms. Heat bands = resting liquidity in the order
book (brighter = more $; levels are aggregated into ~0.2bps price buckets). The white line is price.
Dots are aggressive market orders — 🟢 buys, 🔴 sells, radius ∝ $. White-core bubbles are
liquidations. Use the BOOK pill to set the heatmap to normal / dim / off — signals keep full
strength in every mode. Drag the canvas or use the slider to look back ~5 minutes; the ⏪ REPLAY
button rewinds 30 minutes (PRO).
The right gutter labels only walls that are special: absorbing right now, within 0.4% of price, or with session history. Max 3 per side, nearest first. The label shows the wall's $ size (green bid / red ask), its price, and the level's history badges:
A cyan diamond with 🧊N.Nx marks a level that is executing N× more than it displays right now (volume at the level, correct side, ±0.01%, trailing 60s) while price is in contact. The cyan dashed trail is its history. ·Nt marks a confirmed iceberg: its displayed size has been fully consumed and refilled N times (multi-tranche confirmation — the standard in the published detection literature; a single pass stays a candidate).
When a big wall (≥$500K) vanishes as price approaches without the tape eating it — we check the executed volume before flagging — the radar marks the spot with a ⊕ ring, the amount pulled ("👻 $1.1M"), and a dashed ghost trail that runs on for 45s where the screen liquidity used to be.
When price ignites (≥3bps in 20s), a fixed logistic model scores the probability that the move completes ±0.16% (≈$100 on BTC), re-scored at +20/60/120/180/240/300s. The chip also shows the measured odds that the move runs to ±0.32% and ±0.48% within 30min, conditioned on the score:
| BTC, out-of-sample (~4,000 bursts/symbol) | reaches ±0.32% | ±0.48% | ±0.80% |
|---|---|---|---|
| top-scored quintile | 82% | 56% | 26% |
| bottom-scored quintile | 13% | 4% | 0% |
Live calibration is public at /stats — every prediction and outcome is logged automatically, misses included. Current live top bucket: predicted ~90% → realized ~91%.
The violet lines are today's value area: the price band holding 70% of the session's traded volume (auction theory's "fair value"), with the POC (point of control) at the most-traded price. Recomputed live from kline turnover. Price inside the VA = balance; price outside = the market auctioning for a new fair value — watch whether it gets accepted (volume builds outside) or fails.
The classic auction-school setup, detected live: price pushes beyond the value-area edge, the aggressors get absorbed by big trades at the extreme, then initiative returns toward the area. The violet chip shows the stage (forming → confirmed) and the POC target. Every signal and its outcome (reached POC / invalidated) is being logged; the measured stats will be published once the sample is honest. Until then the chip says what it is: calibrating.
Seven nested trailing ranges (6h → 16d) with live break detection. Measured on 13 months of BTC tick data: after a 24h-range break, P(>1% move within 6h) = 78% vs 66% base (n=448), and the range expands ×1.23 on average. The 4d–16d rungs are rare fat-tail events. Expansion is measured; direction is not implied — we tested directional continuation after breaks across 5 years and it is indistinguishable from chance.
The shaded cone projects the probable price RANGE at +1m and +5m from measured conditional quantiles on our own recorded data. In validation the 80% band held the price 79.7% of the time. Width only — the cone deliberately has no tilt: directional tilt failed validation.
A composite of book imbalance (OBI), 20s taker flow, CVD slope and net liquidations. It describes who is leaning on the market right now — info, not a prediction. Note that OBI reads the visible book, which is gameable (see Ghost 👻); weight the executed components (taker, CVD) above it.
A Coinglass-style ladder of estimated liquidation clusters computed from leverage brackets, consumed live as levels get swept. Clearly labeled an estimate — it is a model of where forced flow may sit, not observed orders.
Everything quantitative on this product carries its sample size and lives in a log you can audit at /stats. When something fails validation we say so on the page (the cone has no tilt; bursts carry no direction; the failed-auction detector is labeled calibrating). When the only serious paper on a phenomenon contradicts the popular reading — as with iceberg satiation — the manual tells you the paper's version, not the popular one. We sell measurement, not prophecy.
Squeeze Radar is an information and visualization tool. Nothing here is financial advice. Trading crypto derivatives involves substantial risk of loss. References: Frey & Sandås (QJF 2017); Christensen & Woodmansey (J. Trading 2013); Zotikov & Antonov (QF 2021); Bessembinder, Panayides & Venkataraman (JFE 2009); Hasbrouck & Saar (2009); Eisler, Bouchaud & Kockelkoren (QF 2012); Avellaneda, Reed & Stoikov (2011). ← Squeeze Radar