Methodology & data sources

Implemented source layer

• CollegeFootballData (opens in a new tab) supplies the configured schedules, teams, games, play-by-play, player/team statistics, lines, weather, rosters, transfers, draft, and related fields used by current jobs.
• The Odds API (opens in a new tab) supplies the configured event and book-market snapshots. Coverage is limited to the current plan, requested regions/books/markets, and captures SharpDataOS actually retained.
• ESPN team endpoints supply current roster observations where the response passes team, season, identity, and completeness checks. Depth-chart availability is reported separately and is not inferred from a missing response.

Provider names, marks, and data remain the property of their respective owners. SharpDataOS uses configured provider access for subscriber research subject to the applicable provider terms; it does not offer raw provider-data resale.

Raw to serving contract

Provider JSON is archived with request parameters, response metadata, content hash, row count, and fetch time before normalization. Derived layers move from provider facts to game facts, season rollups, and serving caches. Corrections and official-total reconciliation stay separate from the original raw evidence.

New raw fetches use unique capture paths and insert-only metadata. Cache reads choose the latest capture for the exact request and verify its SHA-256 digest before parsing.

Market receipt rules

• Every displayed captured outcome retains snapshot, book, market, selection, point when applicable, price, and capture time.
• “First captured” is SharpDataOS's first stored observation for that event—not a claimed sportsbook opener.
• Slate consensus is the median of available book points in the stored frame. It is not weighted by limits, handle, or an invented “sharp” designation.
• Event links must pass team-pair, kickoff-window, confidence, and ambiguity checks before a receipt can appear or be saved.
• A saved exact receipt identifies one stored outcome. Manual terms remain explicitly manual and carry no snapshot ID.

Football evidence rules

Pace, PPA, game-script, and explosive metrics in the dossier use the all-script, garbage-time-excluded FBS-opponent tendency sample. Red Zone aggregates identify their separate game count and can include an FCS opponent. Prior-season player roles are historical evidence, not current availability. Roster capture time and current team coverage are shown independently.

Claims intentionally withheld

The current product does not promote a score, margin, win/cover/total probability, fair price, +EV label, best bet, public-money signal, or player projection. Those require an immutable pregame feature/model ledger, leakage controls, walk-forward evaluation, calibration, uncertainty, and a current executable quote. Descriptive evidence or line movement alone is not enough.

Legacy same-season matchup blends and touchdown-allocation heuristics are not exposed as projections or grades. They can include results that were unknown before the target game and have no immutable pregame ledger or out-of-sample calibration.

Known operating limits

Schedule, odds, roster, and weather data can be stale or unavailable; the UI and readiness endpoint expose those failures. Comprehensive injuries and depth charts are not currently available. Historical market testing is limited to forward captures that SharpDataOS actually stored. These limitations should widen uncertainty or result in “insufficient evidence,” never fabricated precision.