What we measure, and what we do not pretend to measure.
TheHomeRuns.org is moving its public track record onto durable model snapshots: a saved record of what the model actually showed before games were played. This page defines the language before any public performance number changes.
Current public track-record pages still show legacy replay metrics unless a gated preview is explicitly enabled. The stricter snapshot-forward numbers will not become the default until join quality, audit output, and review gates pass.
The official daily snapshot
For public claims, one slate snapshot is official: the first successful scheduled or watchdog full-slate snapshot before 11:00am ET. This is the morning slate users and emails are meant to reflect.
Later refreshes are still valuable. They are better for research and last-minute decision support, but they answer a different question: how good was the model with more information closer to first pitch?
Metric definitions
Pick-hit rate
The percentage of picks that hit at least one home run. Example: 2 hits on 10 picks is a 20% pick-hit rate.
Baseline lift
How a ranked pool performed against random expectation for that slate. Example: if 10 picks were expected to hit 1 HR and they hit 4, that is 4.0x baseline.
Capture rate
The share of scored-slate home runs captured inside a ranked pool. Capture is useful context, but it does not lead the Season Dashboard because high-HR days make this number look artificially lower.
Morning capture rate
Capture rate using the official morning snapshot only. This is the public promise metric.
Live capture rate
Capture rate using later pre-game information. This is a research metric, not the main marketed number.
Replay estimate
A historical recomputation from files or logs when no timestamped official snapshot exists. Useful for research; labeled separately from snapshot-forward truth.
Prime-card pick-hit rate
The hit rate of explicit prime-card picks only. Legacy first-six-lock guesses are not part of the primary metric.
Top-N lift formula
This is the headline frame for the Season Dashboard. It answers the founder and user question directly: did the model's ranked pool beat a random sample of the same size from that slate?
Top-N capture formula
The denominator is the home runs hit by batters who appeared in the scored model snapshot. Pinch hitters and bench players the model never scored are excluded from the capture-rate denominator, but they remain visible in audit output as unmatched or out-of-scope events.
Ranking and ties
Official rank is the unfiltered full-slate rank. User filters can change what a person sees on a working page, but filters do not change the public rank used for measurement.
- Top-N pools contain exactly N rows. Ties do not expand the pool.
- Ties are broken deterministically by final score, base score, HR/FB matchup, season HR rate, batter name, then pitcher name.
- Display rounding does not decide rank. Raw score does.
Migration policy
When public numbers move from legacy replay to official morning snapshots, the site will show old and new methodology side by side for a transition window instead of silently changing the numbers.
If stricter measurement lowers a number, we will say that directly. If it raises a number, we will still show the method change. The goal is trust, not metric shopping.