Methodology
How we measure roster construction
Every metric on this site is reproducible. The dataset is open-source, the definitions are explicit, and we cite the prior art our work extends.
Data sources
All data is public. Joins, time-series alignment, and snap-weighting are ours.
- Draft picks: every selection from 1980 to present, sourced from Pro-Football-Reference and nflverse.
- Weekly rosters: active and injured-reserve status by week, 2002 to present, sourced from nflverse.
- Snap counts: per-game offensive and defensive snap participation, 2012 to present, sourced from Pro-Football-Reference.
- Storage: SQLite, ~1.3M rows. The repository is public: github.com/pete-builds/nfl-draft-analysis.
Retention definition
A player is retained in year N if they were rostered for at least one regular-season week of that team in year N. Snap-weighted retention multiplies that by the player's share of team snaps in year N, so a contributor counts for more than a practice-squad call-up. The drafting team is the team that originally selected the player; trades reset the clock.
Drafting definition
A team's draft output is the cumulative snap share earned by its drafted players in their first four NFL seasons, weighted by pick value (Jimmy Johnson chart, normalized). Pro Bowl and All-Pro selections are tracked separately and reported but not used in the headline ranking.
Time window
Where the data supports it, we report rolling 12-year windows (2012 to current season). For pre-2012 work, snap counts are unavailable; we fall back to roster presence and games played. Pre-2002 work uses season-end roster only.
Position weighting
A starting left tackle staying for five years is not the same as a seventh-round special-teamer staying for two. Where relevant, we report position-weighted retention using snap share within position group. The unweighted version is always available alongside.
Snap thresholds
"Contributor" is defined as a player with ≥20% of their team's offensive or defensive snaps in a given week. "Starter" is ≥50%. Special-teams snaps are tracked separately and not included in the contributor threshold.
Comparison to prior published work
Two pieces have published team-level draft retention or related work in the last twelve months. We extend both with wider time coverage, snap-level granularity, and open data. The table below maps the differences directly.
| Aspect | Dubow / AP (2025) | SIS (2026) | The 53 Report |
|---|---|---|---|
| Time window | 2021–2024 (4 years) | 4-year endpoint, since 2016 | 1980+ picks; 2002+ rosters; 2012+ snaps |
| Retention metric | On active roster or IR after cutdowns (binary) | Players still on team after fourth year | Snap-weighted retention, multi-year, all CBA eras |
| Data source | Sportradar | Not stated publicly | nflverse / Pro-Football-Reference (open) |
| Position weighting | No | Yes (positional value, R1 only) | Yes |
| Draft capital weighting | Day groups (R1 / R2-3 / R4-7) | R1 only | Pick-level (Jimmy Johnson chart) |
| Open data | No | No | Yes — full SQLite, MIT |
Read Josh Dubow's AP piece (Sept 2025) and the SIS positional value study (April 2026) .
Reproducibility
Every published article links to the exact query that produced its findings. The repository contains a notebook for each headline metric. If you find a methodology question we haven't answered here, tell us: [email protected].
Updates
This page is versioned. Last updated April 2026.