Game totals (FT)
Official only when the model states ≥56% — the zone where its hit rate is proven over two backtest seasons. Below that, the read is published as a trend.
AI Trends Journal — MLB
An AI trend is a model signal that doesn't meet every requirement of an official pick: confidence below the validated threshold, unconfirmed pitchers, or a zone still under observation. We publish all of them, frozen at pick time with their stated probability, and settle them against the final score — won or lost. This is our public proving ground: once a zone shows over time that “stated X% = realized X%” at a paying price, it graduates to official.
Why is a trend not an official pick?
First-5 (F5)
Baseball's noisiest market (≈5 runs average, ±3.3 SD): our F5 model is still being stabilized — everything stays a trend for now.
Run line (±1.5)
Official inside the two-season-validated 53-60% confidence zone, with confirmed pitchers and a price that pays the probability. Very high confidences (60%+) are under observation: they win often… at prices that don't always pay.
First inning (NRFI/YRFI)
New engine (July 2026), very well calibrated in backtest — but no zone has yet proven a net profit at real prices across two seasons. Everything is published as a trend, settled nightly, and the zone goes official only if live results confirm.
No settled trends in the last 30 days.
Frequently asked questions
What is the “stated probability”?
The calibrated probability the model gave its trend at pick time. Our transparency commitment: over time, a zone stated at 56% must win ~56% of the time — this page lets you verify it.
Why publish losing trends?
Because a complete record is the only honest one. Every trend is frozen at pick time and settled against the final score, no filtering.
When does a trend become an official pick?
When its confidence zone has proven — in a two-season backtest AND live — that the stated rate is realized at a price that pays it. Official MLB picks follow those validated zones.
