Prediction metrics & accuracy
How well does Defaxon predict? This dashboard tracks our self-correcting prediction engine in the open — every forecast’s earned, out-of-sample accuracy, the patterns we have discovered, and the live event corpus the algorithms run on. The numbers are honest and data-gated: they sharpen as the historical archive deepens, and we never assert an accuracy we haven’t earned. The figures below update live as events flow in.
Metrics are loading…
How these metrics work
Are these metrics live?
Yes. The page renders the current numbers on every load and then auto-updates in your browser every 15 seconds — no refresh needed. Ingestion is bursty: our primary news feed lands in ~15-minute batches while sensor feeds trickle continuously, so totals step up rather than stream. The "last event" timer shows how recently the newest event arrived.
How is Defaxon's prediction accuracy measured?
Every forecast is recorded before its outcome window opens and later scored against what actually happened. The headline accuracy is the share of those out-of-sample predictions that came true — earned from outcomes, never asserted. A backfilled or in-sample result can never count toward it.
What is a discovered pattern?
The engine continuously mines recurring signatures across every domain — a flight route, or a category of event in a country that recurs on a regular rhythm. A pattern becomes 'predictable' only when it is daily-ish, regular, and time-concentrated; scheduled feeds are excluded so a fixture or timetable is never dressed up as a forecast.
Why are some metrics zero or still accruing?
Predictions are data-gated, not code-gated. Cross-pattern links only appear once a co-occurrence clears a statistical-significance gate, and earned accuracy needs forecasts whose windows have already closed. These sharpen as the historical archive deepens — we show the honest current state rather than padding it with noise.