Methodology

Where the data comes from

Combine measurables and draft results are facts of public record, published by the NFL and reported widely. This dataset was compiled and cleaned by the maintainer over several years, cross-referenced against public references. The canonical per-year CSVs, the cleaning decisions, and the full pipeline are in the open-source repository — data under CC BY 4.0, code under MIT.

Derived metrics

PPI — pounds per inch

weight ÷ height in inches: how much mass an athlete carries per inch of frame.

Power Index

√(weight × vertical): mass-adjusted lower-body power. A 260 lb athlete jumping 33″ and a 200 lb athlete jumping 43″ score similarly — that's the point.

Percentile tiers

“Top 10%” always means better than 90% of combine athletes at that position who tested that drill. For timed drills (40, 3-cone, shuttle) the mapping is flipped internally so faster = higher tier — you never have to remember which end of the scale is good.

The athleticism rating

The rating is a true percentile, top to bottom. Every test becomes an exact percentile vs the athlete's position group (2000–2026, timed drills flipped so higher is always better). Tests roll into five equal domains — Speed, Explosion, Agility, Strength, and Size — and the domain average is re-ranked within the position group, so a 94 means this athlete's overall profile beats 94% of his position. Ratings carry one decimal and are floored, never rounded up — a 99.0 is genuinely top 1% of the position, and 99.9 belongs to the single best profile in a large group.

• Size counts, deliberately. The league drafts size, so it's one domain of five — never more.
• Partial testing is worn openly. A rating requires size plus at least two performance domains; every card shows what was and wasn't tested. Athletes below the threshold are unrated, never quietly scored.
• Ratings are position-relative. A 95 cornerback and a 95 tackle carry the same meaning: top 5% of their own position.

Position mapping

Raw combine labels are era-dependent. Every athlete keeps their raw label; norms are computed on canonical groups mapped like this:

227 athletes (83 DB, 141 OL, 3 DL) carry only a generic label and appear in unit-level data, not position-group norms. Every heuristic assignment is flagged on the athlete record.

Honesty caveats

• 2021 is a pro-day year. No combine was held; numbers were recorded at school pro days under less standardized conditions.
• Missing values aren’t random. Top prospects routinely skip drills, so tested samples skew away from the very best. Participation rates are shown on every stat.
• Small groups stay small. There are only 32 long snappers in the dataset. Tiers are suppressed under n=10 and flagged under n=30 — a median of a handful of athletes is a reference point, not a norm.