Live calibration · updated continuously
When we say 80%, you can
check whether it happens 80% of the time.
This page scores AIM's regime-probability engine in public, with no edits. Every forecast is timestamped and locked before the outcome is known, then automatically graded ~20 trading days later against the regime that actually materialized. Hits and misses are both published. No chatbot can make this claim — verify it yourself below.
The declarative version of this headline — "it happens 80% of the time" — unlocks automatically at 50+ scored forecasts with calibration holding. Hold us to it.
Watch the gate re-earn itself: 975 of 30 matured forecasts (classifier v2.0.0)
gate closedWe found and fixed a flaw in our own regime classifier, so we reset the clock on ourselves: 362 pre-fix forecasts stay visible in the ledger below but are not counted toward the gate that must open before any probability reaches a decision screen. Nothing deleted, everything re-earned. Here's why →
Reliability: what we said vs. what happened
Forecasts grouped by stated confidence. A calibrated system's two bars match (±10 pts). Buckets render their bars only at n≥10.
Does the transition warning actually warn?
A meaningful gap between these numbers means the early-warning signal carries real information.
Edge alert receipts — did the call hold?
When a watchlist alert says a ticker gained or lost a validated edge, the claim doesn't get to stand unexamined: ~14 days later the ticker is re-raced on fresh data and the outcome is published here — hit or miss, no mechanism to exclude a bad call.
Edge anatomy — does the regime read predict survival?
Each graded "gained an edge" call is joined with the ticker's regime read at call time (from the hash-chained forecast ledger). If edges born in stable regimes survive their ~14-day re-race more often than edges born mid-transition, the regime read carries real information about edge durability — not just direction.
How this works — and why it can't be faked
Every regime read is written to the scoreboard the moment it's computed — with its full probability vector — and SHA-256-chained into the public ledger below.
After the horizon passes, the realized regime is labeled by the same deterministic rules, leak-free, and the forecast is Brier-scored.
The aggregates above include every scored forecast. There is no mechanism to exclude a bad call.
Confidence numbers only appear inside the product once this public scoreboard proves they're honest.
Every forecast is SHA-256-chained into a public ledger at lock time — currently 1800 links, tip 514e36e429e60236…. Recompute the chain yourself: any edit, deletion, or backfill after a link is written breaks every hash downstream. The exact verification recipe ships inside the file.
✓ Chain self-audited nightly — last verified Nov 11 · 1780 links recomputed clean
Trade with probabilities that answer to reality.
Backtests, regime intelligence, and risk management — all held to the same standard as this page.
Start freeRegime probabilities are statistical classifications of historical price behavior — not investment advice or price targets. Data updated Thu, 12 Nov 2026 23:05:44 GMT · cached 5 min.