Yuno Ball normalizes public NFL records from nflverse-data, licensed under CC BY 4.0. We combine source identities, normalize franchises and positions, aggregate regular-season records, and derive game questions and ratings. nflverse and the NFL do not endorse Yuno Ball.
What’s in this snapshot
29,810 player identities and 131,576 player/team/season records. Rosters cover 1960–2026; supported performance statistics cover 1960–2025; snap counts cover 2013–2025. Incomplete 2026 game statistics are excluded. Version: nfl-v2.1-0ff2f1cf757c.
Roster membership is not proof of a game appearance. Team and era questions require recorded appearances or a verified season-specific performance honor. Recorded game counts are observed minimums; missing historical appearances do not prove a player never played for a team. Searches include roster-only identities, but those identities cannot satisfy unsupported game facts.
Relocated teams share a franchise identity. The Houston Oilers belong to the Titans franchise, separately from the Texans. The Baltimore Colts belong to the Colts franchise, separately from the Ravens. Browns records remain distinct.
Where facts come from
- Player identity and biography: players.csv, supplemented by season rosters.
- Roster history: roster_1960.csv through roster_2026.csv.
- Historical game records: Zack Thoutt’s 2017 public Pro Football Reference archive supplies earlier offense and participation. Dated schedules separate regular season and playoffs. Unreliable historical sacks, tackles and targets are excluded; the archive’s reversed passing-attempt/completion fields are corrected against its published extraction code.
- PFWA honors and historical AP selections retain the year, selector, team and position. Mismatched names, birth dates and teams remain quarantined. An honor can prove a season appearance without supplying an invented game count.
- Regular-season production: stats_player_week_1999.csv through stats_player_week_2025.csv. Traded players’ season totals combine teams; postseason statistics remain separate.
- Observed appearances and workload: snap_counts_2013.csv through snap_counts_2025.csv.
- Draft facts, career totals and awards: draft_picks.csv. All-Pro means AP first-team selections. Award totals cover drafted players and are snapshot totals; missing awards are unknown, not zero. We do not infer MVPs, championships or award seasons from those career totals. Separately sourced season honors retain their actual dates and selectors.
Imports retain source URLs, timestamps and SHA-256 hashes. Identity conflicts and unsupported rows are quarantined rather than silently merged. This import quarantines 4,419 source rows. Missing and inconsistent statistics are omitted rather than treated as measured zeros.
How player OVR works
OVR is a game estimate derived from recorded production. Each season is compared with players at the same position in that season, with at least eight eligible peers. Quarterbacks need 100 attempts, running backs 40 carries, receivers and tight ends 12 catches, and linemen 200 offensive snaps. Statistical ratings need at least four observed games.
Quarterbacks combine passing volume, efficiency, touchdown rate, interception control, completion rate and rushing. Running backs combine rushing volume, efficiency, touchdowns and receiving. Receivers and tight ends use separate positional cohorts. Defenders combine position-specific sacks, tackles, interceptions, passes defended and forced fumbles where supported.
Missing metrics are omitted and remaining weights are rescaled. The weighted percentile maps to OVR as 45 + 50 × (percentile / 100)0.75, with a box-score ceiling of 95. Incomplete component coverage regresses toward 75 rather than inventing missing statistics. A player’s franchise-era rating combines the best three distinct supported seasons at 45%, 35% and 20%, rescaled when fewer seasons are available. A single-season estimate above 92 is regressed toward 92. Position eligibility is checked against those historical seasons, not a current biography label.
Historical line and defensive ratings use supported participation or season-specific honors when detailed metrics are unavailable. An ordinary participation estimate is 60 + 25 × same-season positional participation percentile. Role-verified first-team recognition starts at 94, with one point per distinct prior first-team year at that position in the last seven years, capped at 99; second-team recognition is 90. Multiple selectors in the same year count once. A punting or return honor cannot boost a different roster position. Dated honors can support a rating when game counts are unknown. These estimates carry limited confidence. The 1960s currently have only three verified centers.
Offensive-line data measures workload rather than blocking quality. Its snap-only estimate is 55 + 0.3 × snap-volume percentile, capped at 85. Coverage, blocking, scheme and tracking data are incomplete. Limited confidence is shown where these omissions matter. No player receives a hand-assigned star rating. Unsupported eras and positions have no rating.
How a team becomes a record
A 22-player team has eleven offensive and eleven defensive slots. Offense weights quarterback 34%, skill positions 33% and line 33%. Defense weights pass rush 30%, run defense 25% and secondary 45%. The average of those units is Team OVR. Team Score additionally accounts for offense/defense imbalance, the weakest unit and the weakest three starters, allowing two OVR points before the two gap penalties. Gaps are penalized more heavily as the roster approaches the elite ceiling.
Strength maps from 40 to the verified canonical roster reference 97.66521 onto a 0 to 1000 Yuno Team Score, clamped at either end. Projected wins are round(17 / (1 + exp(-(Team Score - 660) / 92))). This nonlinear game model reserves perfection for exceptionally strong, balanced rosters. These are game-balance parameters, not fitted NFL forecasts. Detailed weighting appears under How We Calculated This on the result screen. There is no wagering or prize implied.
Grid, Trivia and rarity
Each Grid has nine verified, nonempty intersections and a solution with nine different players. Difficulty reflects the median answer-pool size: Expert up to 8, Hard up to 25, Normal up to 100, otherwise Easy. Daily boards share a UTC date and dataset version. Practice boards are randomized.
Trivia combines verified facts with original question templates. Distractors use contemporary positional peers, nearby draft classes or plausible numeric alternatives. Difficulty reflects prominence, draft depth, historical era and statistical rank. Questions identify their supporting source after the answer.
Grid rarity remains unavailable until real persistence is connected. There are no simulated community counts. The prepared adapter requires at least 50 recorded picks in a cell before showing a percentage; storage failures return unavailable. Guest sessions are signed, expire within 24 hours and are unranked. A signed round is not a competitive anti-cheat ledger.