v3 · Pattern matching, MLB
Most games don’t clear.That’s the whole idea.
A pattern-matching model that grades every matchup it has data for and publishes a forecast only when the patterns clear a fixed threshold. Most of what it looks at never gets published.
Thirty seconds on how this works
Model performance summary
Games held back
324 of 423
77% of everything the model produced a view on over 34 days never got published.
Accuracy on published forecasts
60.6%
60 correct, 39 incorrect, across 99 published forecasts.
Days with no output at all
4
Nothing met the bar, so nothing was published.
Window: July 24, 2026 through August 26, 2026. Past performance does not predict future results. Model accuracy varies by sample and by period, and any published record reflects a limited window.
The filter is the product
Every matchup gets graded. Almost none clear. A forecast is published only when enough historical patterns match and no danger signal fires — and on plenty of days that means nothing goes out at all. What the model declines to say is as much of the system as what it says.
One tile per game the model produced an output for between July 24, 2026 and August 26, 2026. Filled gold tiles cleared threshold and were published. Hollow tiles did not — the model logged a view and kept it to itself.
One model architecture, more sports coming
Each sport gets the same treatment: pattern matching, a fixed threshold, and no launch until there’s a public record to judge it by.
Access opens this fall
First 100 people to reach my inbox get 25% off the first four months. Until then the daily outputs are free and the record on this site keeps updating either way.
What a subscription includes
- The daily model card, published before first pitch
- Every forecast timestamped and added to the public record
- The grade and the patterns behind each output
- The full held-back log — every matchup the model passed on, and why
- Full performance tracking, including the figures kept off the public pages
- NFL and college football forecasts included at no extra charge during their proving period
What it is not: this is not wagering advice, and no part of the subscription accepts wagers, places them for you, holds your money, or takes a share of any outcome. It is access to the output of a statistical model.