Research you can trust has to grade itself. Every verdict and forecast Magnisto publishes is snapshotted at that moment — the score, the rating, the forecast direction and the share price — and then graded automatically against the realized return 21, 63 and 126 trading days later. Nothing is graded in hindsight, nothing is quietly dropped: misses resolve on the same schedule as hits, and the numbers below update as each prediction ages past its horizon.
What you'll find
Every prediction recorded at publish time — no hindsight edits
Misses and pending calls shown alongside the hits
Directional accuracy and edge measured against the base rate
Calibration (Brier score) — was the stated confidence honest?
Accuracy split by model version, so improvements are measured, not asserted
Frequently asked questions
How is the Magnisto track record measured?
Each time Magnisto generates a verdict, the prediction — composite score, rating, forecast direction and the share price at that moment — is written to a ledger. A scheduled job later grades every entry against the actual stock return 21, 63 and 126 trading days after the prediction date. The page shows directional accuracy, the edge over the period's base rate, and a Brier calibration score.
Does Magnisto hide its bad calls?
No — that is the point of this page. Grading is automatic and runs for every ticker ever predicted, whether or not anyone looks at it again, so a wrong call cannot be dropped from the record. Recent calls are listed with their realized returns, including the misses.
What do directional accuracy and edge mean?
Directional accuracy is how often a bullish or bearish call matched the direction the stock actually moved. Edge subtracts the base rate — the share of all graded windows that were positive anyway — so a number above zero means the calls beat a coin flip weighted to that market period.
Why do some predictions show as pending?
A prediction can only be graded honestly after its horizon has passed. A call made recently has no 63-trading-day return yet, so it is shown as pending rather than being counted — the denominator only includes fully aged predictions.