Sigma-1: self-relative anomaly detection, on the record.

Most attention metrics are popularity contests: cross-asset rankings that structurally favor whatever is already large and loud. Sigma-1 inverts the frame. Each asset is compared only to itself, so the unit of measurement is departure from baseline, not size. A quiet asset waking up outscores a loud asset doing what it always does. That single design choice removes cap bias, venue bias, and most of the reflexivity that makes trending lists self-fulfilling.

Each snapshot consumes 24-hour traded volume and the absolute 24-hour price move. Deliberately minimal: attention expresses itself first as volume and displacement, both inputs are public and checkable, and a score's credibility begins with the auditability of what feeds it. Richer signals (derivatives positioning, stablecoin flows, exchange-level order flow) are collected by the Terminal as context and reserved as candidate features for future Sigma models. They do not touch Sigma-1.

Each input is converted to a z-score against that asset's own trailing 30-day distribution. Z-scores, per ticker. Yes, that is where the name comes from. Two guardrails make the statistic honest. Z-scores are clamped at four standard deviations, because beyond four sigma everything is maximally unusual and one insane print would otherwise pin the scale and flatten the board. And no asset is scored until 14 days of baseline exist, because a z-score on three days of data is an opinion. New listings appear unscored rather than artificially hot.

The 30-day window adapts to regime change with deliberate lag: an asset that becomes permanently louder normalizes within weeks, so Heat keeps measuring novelty rather than a new steady state.

The clamped z-scores combine in a fixed weighted blend, normalized so 100 requires both inputs at the clamp simultaneously. The blend weights volume anomaly above move anomaly: volume is the cleaner attention proxy, while move anomaly alone rewards ordinary volatility. The exact weights are proprietary and stated as such; the principle is public and testable against the record.

Heat at or above 70 opens an event and freezes the asset's price at that instant. Closing requires hysteresis: Heat must fall below a cooler band and stay there across consecutive readings, so one surge produces one event rather than a dozen border flickers. Peak Heat and the triggering z-scores are recorded with the event.

Every event resolves at +24h, +7d, and +30d as the forward return from the frozen open price, computed from Tickerz's own stored history so nothing can drift after the fact. Receipts publish wins and losses identically. This is the point of the whole system: an attention score whose historical meaning anyone can check at any time, including the finding that attention frequently precedes drawdowns. See the live record on the receipts page or the mechanism on history in the Sandbox.

The top 100 crypto assets by market capitalization refresh every 15 minutes around the clock. A curated set of major US equities runs on the identical mechanism at 10-minute intervals during regular market hours, each against its own baseline. When an upstream feed halts, the board says HALTED. Machine access to everything is free via the API.

Public: the inputs, per-asset standardization, the four-sigma clamp, the 14-day minimum baseline, the event threshold at 70, the existence of hysteresis, and the receipt horizons. Proprietary: the blend weights, the close band, the confirmation count, and any future calibration parameters. The record exists so you don't have to take the hidden parts on faith: judge the model by its receipts.

The Terminal already stores market context beyond Sigma-1's inputs: derivatives positioning, stablecoin supply, deployed DeFi capital, US spot flow. As the receipt record matures it becomes labeled training data, and Sigma-2 will use it to move from a fixed blend to a calibrated score with published calibration. Sigma-1 stays frozen and versioned beside it, because a track record only means something if the model that produced it doesn't quietly change underneath.

Heat is not a prediction, a buy signal, or an endorsement. Nothing on this site is financial advice; the standing terms live on the disclaimer page. Heat measures attention, not merit.