Glossary
The defined terms that appear across Noah Predict output.
Every term below is drawn directly from Noah Predict output. Definitions are deliberately short — a single decision-ready sentence — with cross-links to related terms. For the underlying methodology, see methodology. For the product FAQ, see faq.
Backtest window
The historical interval used to measure a model's predictive performance against known events. Noah Predict publishes rolling 90-day backtests for every benchmark-grade vertical.
Benchmark-grade
A coverage tier where source density, backtest history and calibration meet published thresholds sufficient for use in regulated decision support.
Chokepoint risk
The probability of material disruption at a maritime bottleneck (e.g. Strait of Hormuz, Bab-el-Mandeb, Suez, Malacca) over a defined horizon.
Civil-unrest change-point
The moment at which protest dynamics shift from background level to an escalatory regime, detected through acceleration in frequency, geography and participant composition.
Composite risk score
A single 0–10 score combining weighted drivers from a vertical into one decision-ready number. Always accompanied by its confidence band and the list of contributing drivers.
Confidence band
The interval within which the true score is expected to fall given current source density and model calibration. Narrower bands indicate higher certainty.
Contrarian read
A mandatory field on every Noah output naming the strongest case against the headline view. "There is no contrarian case" is a validator rejection.
Coupling
The degree to which two independent signal streams move together. When port closure and road closure couple at Jaccard 1.0, cargo movement has stopped.
Divergence
The gap between what official sources say and what the broader information system is saying. Large divergence is a strong predictive signal.
Driver
An individual input signal that contributes to a composite risk score. Every composite publishes its driver decomposition and the weight applied to each.
Equal-source vote
The doctrine that a wire service, a court listing, an academic pre-print and a parliamentary transcript each count once. Tier-one prestige does not confer additional weight. Weight is earned by corroboration. Empirically attested via the Polymarket 30% drop.
Exploratory beta
A capability tier where the engine produces directionally useful output but lacks the source density or backtest history to claim benchmark-grade reliability. Clearly labelled as such in the Labs page.
External anchors
Six independent public references attached to every country-level risk bundle: FCDO, US State Department, Global Peace Index, World Bank WGI, OFAC & UK HMT sanctions registries, and ACLED. Broad agreement is confirmation; divergence is a prompt to read Noah's blind spot.
Held-out set
A sample of historical data withheld from model training used to measure out-of-sample performance. Each vertical publishes its held-out composition alongside the backtest.
Jaccard coupling
A coupling score between 0 and 1 measuring how frequently two signal streams co-occur across a time window. 1.0 means the two always fire together in the sample; 0.0 means never.
Narrative lead
The median elapsed time between a Noah Predict signal firing and the same directional change being reflected in broker consensus pricing. Currently 6.4 hours across verticals.
Noise is substrate
One of the five doctrine rules: a sentiment expressed by many low-prestige sources is a real sentiment. Conventional tools filter it as fringe; Noah measures it as substrate.
Per-paragraph sentiment
Classification performed at the paragraph level rather than the article level. An 800-word article typically contains 6–12 structured signal units. Article-level classification loses them.
Polymarket attestation
An empirical field travelling with every bundle showing that upweighting tier-one sources reduced Noah's predictive accuracy on resolved Polymarket markets by 30% (Brier score). The load-bearing evidence behind the equal-source-vote doctrine.
Port closure risk
The probability of a named port entering an operational-stop state over a defined horizon. A maritime sub-signal that often couples with road closure at Jaccard 1.0 when logistics shut down.
Positive predictive value
The share of fired signals that were followed by the predicted directional change within the stated horizon. Noah's current composite PPV is 0.71.
Progressive narrowing
The doctrine of running multiple retrieval passes — typically three or four — each narrowing the search on what the previous pass surfaced. Enables reducing a 20,000-item corpus to a 300-signal answer without losing the signals you wanted.
Provenance
The property that every conclusion is traceable to the underlying drivers, thresholds and source documents that produced it. Noah publishes 100% provenance; no output is rendered without an audit trail.
Reduction funnel
The five-stage pipeline every output passes through: Ingest, Classify, Score, Reduce, Render. Documented publicly.
Regime stress velocity
The rate of change in sovereign-regime stress indicators over a rolling window. Used as an early-warning driver for regime transition and default risk.
Shadow-fleet exposure
Exposure to vessels operating outside mainstream AIS reporting, sanctioned ownership structures, or flag-of-convenience arrangements. A maritime underwriting signal with its own scoring lane.
Signal fire
The moment at which an underlying driver crosses a calibrated threshold indicating a directional change in risk. Timestamped, source-attributed, auditable. Approximately 1,847 signal fires per 90-day window across all verticals at current coverage.
Trajectory
Where velocity is heading over a defined window. The second derivative of attention. Positive trajectory is an accelerating narrative; negative is one settling down.
Velocity
Rate of change in mentions, severity and actor involvement. The first derivative of attention. One of five physics measurements applied continuously to every signal.
Volatility
Noise around a signal. High-volatility signals are noted but weighted lower until they stabilise. One of five physics measurements.
Every term is live in the product.
Open the workspace to see them in Noah output.