Methodology
What this terminal measures, how explanations are ranked, and exactly what it refuses to claim. This page is deliberately specific about limits: a research tool is worth trusting only if its absence states are as precise as its numbers.
Units
Probability is stored as a decimal between 0 and 1 and displayed as a percentage. Movement is expressed in probability points — a move shown as +6.5 pts means the probability rose by 0.065. Volume is the served figure for the stated window; where a value is absent the field reads “not reported” rather than zero.
Movement detection
Venue prices are collected on a one-minute resolution and stored as append-only observations. A detector compares recent observations against a baseline and records a move with its detection time, reaction window, pre- and post-probability, volume and a z-score against that baseline. Nothing on the measurement side of a Catalyst Card is generated — it is read back from those stored observations.
Candidate ranking
When a move is detected, timestamped publications from a curated registry of sources are aligned against the move window. A publication can only be a candidate if it was published before the move started; anything published later is excluded. Candidates are scored and ranked, never collapsed into a single verdict — when two candidates score close together, both are shown and neither is promoted.
The score expresses how well timing, entities and market reaction line up for that candidate. It is a ranking signal, not a probability that the explanation is true.
Unattributed moves
When no candidate crosses the confidence floor, every surface renders the same final statement: Unattributed move — no candidate catalyst found. That is an answer, not a pending state — there is no spinner, no deferred analysis, and no later rewrite. The measurement remains complete and useful either way.
Cross-venue comparison
Two venues rarely price identical questions: thresholds versus buckets, different resolution sources, different close times. A spread is printed only when the pair has been confirmed comparable; otherwise both values are shown and the spread is suppressed with the reason. Comparison must be earned.
Data gaps and staleness
Collectors report their own freshness. When a collector falls behind, values remain visible but are labelled stale with their age. Recorded gaps in any time-series view are drawn as gaps — a line drawn through minutes nobody observed would invent a trend that never happened. Until a served time-series endpoint exists, history panes draw nothing at all.
Licences and attribution
Venue data carries its required attribution wherever it appears — currently Market Data provided by Kalshi and Market Data provided by Polymarket (venue codes KAL / PM). A venue whose licence is disabled disappears from serving entirely rather than appearing unattributed.
What this product is not
EventFrame holds no positions, places no orders, touches no customer money and never asks for exchange credentials. It does not advise. Explanations describe statistical association between publications and moves — they are not statements of causation or recommendations of any action.
Statistical association, not causation. Not investment advice.