What we publish
Different sports, different outputs
Connected feeds provide game data and market observations. A separate model artifact provides a decision, and a separate settlement record provides its grade. One cannot stand in for another.
A selected-side probability,
with its record.
When an MLB selection is supplied, the board identifies its selected team, moneyline selection and selected-side probability. A missing selection or probability stays unavailable instead of becoming a pick. A missing market reference leaves the market comparison unavailable; it does not change a supplied selection.
Before an MLB game starts, Model vs market compares the selected-side probability with a current, margin-adjusted market probability from one book. We use a complete home-and-away moneyline pair from the same capture, choosing the eligible book with the lowest nonnegative overround. The book is chosen independently of the selected side; ties use provider and book identifiers.
We normalize the two implied probabilities to sum to 100%, then subtract the selected side’s market probability from its model probability. The result is in percentage points, not expected return or a betting recommendation. The displayed best price is separate from this reference. No Pick, a started game, prices older than five minutes, or no eligible complete pair leaves the comparison unavailable.
MLB boardA weekly-current selection,
with retained records.
NFL weekly-current predictions can refresh before kickoff. When supplied, the board identifies the selected team, moneyline selection and selected-side probability. Market edge is not published.
Displayed retained selections and results remain bound to their source records. A current view does not rewrite the original selection or its result.
NFL boardA selected-side probability,
with its source inputs.
NCAAF selections can name either the home or away team. The displayed probability is for that selected side, not always the home team. Prior-season team support and the home or neutral-site setting provide input context; they are not measured contributions from each factor.
An absent selection is not a model pick. An observed price alone is not a model selection, and a result is shown only when its saved record supports it.
NCAAF boardTime is part of the record
Pick-generation time, market-observation time and result-source time describe different events. A prediction can refresh before kickoff while a market observation remains missing or records a different time.
Current views follow the timing supplied by their source. A later quote or result does not rewrite a retained selection. Missing or unusable source timing stays unavailable.
Saved records preserve the timing supplied with that record. Freshly retrieving an old artifact does not make its contents current.
How to read an entry
- No selection
- An absent selection is not a model pick.
- Pregame selection
- A selection before the game is a current source record, not a promise about the final result.
- Historical selection
- A retained selection remains tied to its saved source record after the game starts.
- Ungraded result
- A final score without the exact saved selection and matching final does not grade a pick.
- Verified outcome
- A grade appears only when the saved selection and matching final support it.
Why a selection can be absent
A selection can be absent when the current source does not supply one or cannot support it. That absence remains visible as no model pick; the page does not infer a selection or a reason that the source did not provide.
How grading works
The separate MLB aggregate archive counts only the published eligible cohort: graded records equal eligible records and wins plus losses. Historical quarantine is separate. Its reason counts overlap, so they are never summed. Voided or postponed exclusions do not identify individual game dispositions in this aggregate feed. Coverage abstentions on the forecast board remain separate from graded performance.
A win rate needs eligible graded observations. A saved-price return needs price coverage and uses the source’s priced cohort; it is not an executed wager or a single-book result. Zero denominators mean unavailable, never 0% performance. Individual grades, voids, dated performance and settlement histories are not reconstructed from team names or dates.
In that MLB archive, published output is recorded with freshness unverified. The model generation time, feed generation time and retrieval time are separate. The game file becomes historical after its recorded start; elapsed time alone never establishes a final result. Model version, serving receipt and captured spread, total, moneyline, book and line timestamp are unavailable when absent. An artifact digest is not a model receipt. Stable game addresses retrieve only the selected feed, not an archive.
The separate MLB archive price snapshot has no canonical game identity, quote observation time or timezone on event times. It stays unverified and separate from the source odds comparison. No current best-price rank, pick join or model edge is inferred from these individual prices.
A grade requires the exact saved selection and a matching final. The NFL record can retain graded selections, but retained history can be incomplete; an absent entry does not establish a complete record. Dated September 10 snapshots and the MLB aggregate archive describe only their own cohorts. A score alone does not grade a model record.
The current model feed does not supply a complete out-of-sample evaluation, so its record pages do not infer a model-quality conclusion. A described source-bound saved-price summary remains limited to its priced cohort and is not a model-quality conclusion. Any descriptive evaluation in a saved snapshot applies only to that snapshot and its stated sample, not to the current board.
We do not turn a missing price into zero, an incomplete feed into a prediction, or an illustrative example into a track record.
Understand the game. Keep the uncertainty.
BetMetrics is not betting advice. Outcomes remain uncertain. Nothing on these pages is a promise or a recommendation to act.