A public, timestamped, tamper-evident record of every Filth+ prediction PitchIQ makes — scored honestly as outcomes resolve. Hits and misses both shown. Nothing backdated, nothing hidden.
Loading record…
How this works. Every night PitchIQ logs a prediction for each qualified pitcher (75+ pitches): a near-term call (Filth+ stays within a time-scaled cone over 30 days, plus a directional read) and, where age is known, a season-horizon aging projection. Each prediction is written once and never altered. When its target date passes, the actual Filth+ is compared to the prediction and scored hit or miss.
Why the calibration target is ~68%, not 100%. A 1-SD cone should contain the outcome about 68% of the time by definition. A system that "hits" 100% has cones so wide they say nothing. Landing near 68% is the sign the uncertainty is honestly sized — being right the right amount, not all the time.
What this does and doesn't predict. Filth+ measures pitch nastiness from physical shape — it predicts Filth+ and the bat-missing outcomes downstream of it, not ERA or box-score lines, which carry defense, luck, and park noise that stuff metrics exist to strip out.
Verification. Each day's complete prediction set is fingerprinted with SHA-256 and published externally. Re-hashing any day's recorded predictions and matching the published fingerprint proves the record was not altered or backdated since that date.