Open notebook / Coverage desk
Know what is here.
Know what is missing.
A research resource earns trust by naming its boundaries. This page records the active data editions, model scope, recruiting review and market status behind the published site. Counts are source records or retained identities, not claims about every player or game.
2,287Published 2026–27 matchup forecasts
887,748Current-edition player box rows
366Basketball programs with ratings
484Verified pregame market observations
01 / The active editions
Two sports, two honest baselines.
Generated Sep 12, 2026Football / 2026
Scores, players and units.
Schedule records · 2022–2618,758
Player archive rows · 2018–26693,774
NCAA player-game evidence · 2013–251,015,076
Upcoming FBS forecasts708
2025 holdout / margin MAE14.2 pts
The player archive spans 2018–26 and keeps defense/specialist event rows separate when the source has no stable athlete ID. Forecasts use team identities and home field; they do not use injuries, depth charts or recruiting.
Open football methods →
Basketball / 2027
Possessions, players and context.
Schedule records across seasons26,756
Paired completed box games25,041
Upcoming 2026–27 forecasts1,579
2025–26 holdout / margin MAE10.3 pts
Historical player logs span 24 published seasons. The independent ratings use opponent-adjusted efficiency, tempo and pooled Four Factors; NCAA RAPM stays in its own source identity namespace.
Open basketball methods →
02 / Football player evidence
Every role has a visible source boundary.
Generated Sep 12, 2026The identified archive and the event notebook are complementary. Box rows use source athlete IDs and support player profiles; defensive and specialist releases carry names and game context without stable athlete IDs, so they remain separate and are never name-joined. The NCAA-derived player-game release is listed separately because it carries contest context but no stable athlete ID.
The two event notebooks contain 251,774 source records across 27,274 game contexts. A repeated name is kept as a separate source row and is not a career total.
03 / Basketball evidence inventory
Every stat layer has a receipt.
19 published layersRow counts are table-local source records. They are not deduplicated person counts, and identities from NCAA releases are kept separate from ESPN-derived records unless an exact source key is available.
Live Cloudflare check
Confirm the warehouse behind the page.
Checking D1…This read-only check queries the deployed Cloudflare D1 database, rather than the bundled static files. It gives the current remote row counts and the latest source receipt clocks used by the research publisher.
Loading remote coverage…
04 / Supplemental research archives
The deeper player and possession files.
Source-native and derived layersThese archives power the historical player, shooting, lineup, team-box and impact desks. Their rows are not interchangeable identities: NCAA records stay in the NCAA namespace, and derived profiles retain their source season and edition.
05 / Identity review queue
Missing IDs do not mean missing observations.
174,521 rows retain source valuesSource rows are withheld from player and team joins when a required identifier is absent. The original payload stays in the private warehouse; this summary counts whether useful source fields remain, without guessing an identity or promoting the row into a ranking.
These rows remain excluded from player rankings, career totals and forecast features until the source supplies a stable join key. A source value is not silently attributed to a nearby player.
Open the retained source rows → Search and download a bounded page when auditing a source release.
06 / Recruiting file
Useful evidence, clearly partial.
Programs with reviewed announcements14
People with dated records96
Announcement events98
Observed 2026–27 roster listings5,459
2024–25 source roster snapshot12,931 · 699 programs
2025–26 source roster snapshot9,988 · 727 programs
2026–27 source roster snapshot5,459 · 354 programs
Programs in roster source view354
Roster publisher profiles linked5,459
Roster rows with publisher Box BPM4,580 · 83.9%
School announcements are retained with source links and dates. A signing does not establish eligibility or availability, and an absent listing does not establish departure. Box BPM is source- attributed prior-season context and stays blank when its exact athlete/team row is unavailable. The review is not a national recruiting census.
Read the recruiting file →
07 / Forecast record
Predictions have a clock.
Football registered versions6,457
Basketball registered versions55,265
Confirmed-start games361
Football market observations216 · 74 games
Basketball market observations0 · 0 games
Settled games in the prospective record36
The archive currently retains 216football observations across 74 games from the prospective ESPN Summary capture. Basketball has 0qualifying observations in this edition. Historical imported lines without a publisher clock remain outside market evaluation, and retained reading snapshots stay separate from the scorecard.
Open the prospective record →
08 / Basketball data library
Choose the evidence layer.
Each dataset keeps its own source identity.Player archive
3,676,397 identified box rows
24 source seasons and 98,830 archived player identities, with game logs and field-level coverage.
Open player statistics →Roster workload
5,459 source-listed players
Rank the 2026–27 roster release by prior minutes, scoring, playmaking or shooting efficiency while keeping source status and player evidence visible.
Open the roster workload board →Publisher context
8 source-native leaderboards
Preserve the publisher’s labels and display values alongside Silvermine’s derived production rates.
Open national leaders →NCAA snapshot
3,842 national records
Final individual tables across 3 divisions, with publisher ranks and missing-field coverage visible.
Open NCAA leaderboards →Boutique context
4,974 RAPM records · 738,233 shots
Read possession impact and shot-location evidence without joining separate source IDs by name.
Player impact → · Shooting lab →
Shot-location validation
733,313 located attempts · 99.3% usable
Coordinates are retained only when they fall inside the court bounds and agree with the stated shot type, value and distance. 4,920 attempts remain visible as rejected location evidence.
Player games reconciled to box scores111,293 / 113,675
Inconsistent coordinates2,276
Placeholder coordinates2,644
Missing coordinates0
A located attempt is event evidence, not optical tracking. Rejected or unmatched rows stay out of location-derived summaries rather than being repaired by inference.
Open the shooting evidence →Play by play
2,915,731 events · 6,275 games
A searchable game index connects the retained event releases to the publisher's complete source pages and shot reconciliation.
Open the play-by-play archive →09 / Source boundary
Attribution is part of the statistic.
Current releases come from the attributed SportsDataverse bulk store. Direct ESPN extraction is disabled under source terms, and NCAA requests stop when robots policy disallows crawling. Missing values remain missing; names are never used to invent identities.
Browse retained briefs ↗Football edition Sep 12, 2026 · basketball edition Sep 12, 2026 · ledger edition Sep 15, 2026. Source receipt timestamps and hashes are available from each desk’s methodology page.