Data & Sources
The repaired publication boundary requires every value to trace to an admitted, byte-verified source capture with complete transitive lineage, rights posture, schema evidence, and receipt identity. Missing or stale evidence blocks publication. Nothing is scraped. Upstream changes are staged for review and never silently replace an admitted object.
▸ Attribution
- Play-by-play, schedules, rosters, players and supporting datasets: nflverse, CC BY 4.0.
- Participation data 2016–2022: NFL NextGenStats via nflverse, CC BY 4.0.
- Participation data 2023 onward: FTN Data via nflverse, CC BY-SA 4.0.
- Charting descriptors 2022 onward: ftndata.com via nflverse, CC BY-SA 4.0.
- Snap counts: Pro Football Reference, via nflverse's licensed redistribution.
- Contracts: OverTheCap.com via nflverse.
Proprietary systems (PFF, SIS Total Points, ESPN Total QBR, FTN DVOA) are referenced for methodology and used — where permitted — only as private validation comparators. Their values are never scraped, displayed, or used as training labels. Big Data Bowl tracking data is contractually restricted and is never used.
Conditional-rights rule: the current artifact-class policy withholds 2023–2025 FTN participation before read from public team, player, biography, coverage, and research artifacts. It allows 2016–2022 NGS participation for declared artifact classes only with verified CC BY 4.0 lineage and exact NFL NextGenStats attribution. Any future FTN allow decision must require exact FTN attribution, complete lineage, and compatible CC BY-SA 4.0 output terms. If the decision or any required term is missing, stale, or incompatible, publication fails closed. This is an operational compliance posture, not a legal ruling.
▸ Reproducibility and durable restoration
The frozen completed-season evaluation contains 104 source bindings spanning 1999–2025; independent strict URL replay recovered 101 exact historical byte streams. Current capture storage contains 109 immutable receipt files for 107 active index entries: three additional 2026 identity/role bindings and two superseded player/schedule receipts retained for history.
The repository holds eight content-addressed exact snapshots: the three original evaluation restorations, the DOA-7 schedules snapshot, and four frozen 2026 identity/role inputs. Every SHA-256 matches its receipt. These disclosed snapshots restore the chain without pretending that mutable URLs still serve old bytes or that the August 8 roster evidence describes today's roster.
▸ Source register
The live register below is generated from the machine-readable rights registry — the same file the capture tooling enforces. Raw JSON: sources.json · freshness.json
| Source | Rights status | License | Attribution | Captured |
|---|---|---|---|---|
| Enable JavaScript for the live table, or read sources.json directly. | ||||
▸ Field coverage by season
Non-null rates for the participation fields models depend on — measured, not assumed. A field a season never recorded bounds what that season's models may claim. Raw JSON: coverage-map.json
| Enable JavaScript for the live table, or read coverage-map.json directly. |
▸ Publication validation
The gate checks completed schedule games in both directions, refuses fatal or error findings, and binds each report to the exact canonical and capture hashes used for publication. Season counts below are generated from those reports, never written by hand.
Enable JavaScript for the coverage detail, or read coverage.json directly.
▸ Known source limits, stated plainly
- Participation data for 2023 onward arrives only after each postseason ends — it never updates in-season. In-season features rely on nightly play-by-play and ~48-hour charting descriptors instead.
- Participation reveals who was on the field, never who was responsible. Assignment-level evidence is not lawfully available to a public product — which is why trench and coverage estimates carry wider uncertainty and competing-model disagreement.
- Rows listing more than 11 offensive or defensive participants are retained in source evidence but excluded from derived participation profiles with explicit counts. The audited anomalies were 2,977 rows in 2018 and 2,250 in 2019.
- NGS aggregate stats currently cover 2016–2024 upstream.
- Completion probability and CPOE fields exist only from 2006 onward.