What powers the intelligence
Every reading on the scanner is a real measurement. Behind it sit six independent data systems — positioning, price, options, macro, and a volatility model — each computed the same way, point-in-time, from the data our published research is built on.
No black box and no playbook: we tell you which systems are stretched and what has typically followed. We never tell you what to buy or sell. Want the exact definitions — windows, thresholds, formulas? They're all on the methodology & glossary page.
The six systems
Speculative Sentiment Index (SSI)
How the retail FX crowd is positioned — net long vs short — and how extreme that positioning is versus its own history. Extremes flag stretched conditions, not a direction to fade.
Commitments of Traders (COT)
Where large speculators sit in FX futures, as a z-score against their trailing year. A slower, institutional counterpart to the retail read.
Options-dealer positioning (GEX)
Where options dealers are likely to dampen or amplify moves. Positive gamma pins price toward key levels; negative gamma amplifies away from them. We surface the regime and the nearest call/put walls, correctly scaled per pair.
Where price actually sits
Percentiles within every trailing window (7d to all-time), distance to 52-week highs and lows, round numbers, weekly/monthly fibs, and average daily range — the objective geometry of a market, in pips.
Economic-calendar surprise
How far a release landed from consensus, as a standardized surprise. The surprises — not the scheduled prints — are what move FX.
The RV forecaster
A machine-learning model that forecasts how MUCH a market is likely to move over the next minutes-to-day — the expected move size, never a direction. It is the one input we score continuously, in the open. Its calibration is below.
The context layer
None of these systems ships raw. We pull from a dozen independent sources — retail sentiment feeds, the CFTC, options chains, central-bank rate data, an economic calendar, tick-level prices — each on its own schedule, then clean, align, and normalize every series into a single point-in-time fact layer. A COT print from Friday, an options snapshot from this hour, and a price from a second ago all resolve to what was actually knowable then — so nothing is contaminated by hindsight.
That normalized layer — every reading expressed as a z-score or percentile against its own history, comparable across 35+ instruments — is the product. It is not a broker feed, not an indicator, not a signal service repackaging a single dataset. It is an independent context layer built to answer one question the rest of the market skips: how unusual is this, and versus what?
How well the model is calibrated
Five of the six systems are direct measurements — there is nothing to “score.” The sixth, the volatility forecaster, makes a prediction, so we grade it continuously against what actually happened and publish the result. This is the quality of one input, not a trade record: the model says how much a market may move, never which way.
How we find what happened last time
When a market reaches an extreme, we don't ask what an indicator says — we ask when this exact situation has happened before, and what followed. Each moment is reduced to a multi-factor fingerprint: where price sits, how the crowd and the futures market are positioned, the volatility regime, the options backdrop — dozens of dimensions at once, not a single line crossing a threshold.
We then search decades of history for the closest matches to that whole fingerprint and surface what actually happened next in each — the distribution of outcomes and the typical path, with its real spread. Every analogue is scored point-in-time, so each one was genuinely knowable at its own moment. We show the record and how widely it varied; we never dress it up as a prediction. The past is not the future — but knowing how a setup has resolved before, and how much it scattered, is context you will not find anywhere else.
How it fits together
Each system produces an honest reading. The scanner counts how many are at a statistical extreme for a given instrument — a number we call extremity, a count of stretched factors, not a direction. Click any instrument and the same inputs assemble into a dossier: where it sits, who's positioned, what it moves with, and — for members — what has typically happened the last times it looked like this.
Everything is deterministic and reproducible. No opinions, no model guessing at direction — just the same facts, computed the same way, every hour.
See the systems at work on today's market.