What is IDP DMX? Scoring Defensive Prospects

Data Dictionary · the defensive parallel to the DMX rookie score

Same skeleton, defensive muscles

IDP DMX keeps the offensive model's build — three independent pre-draft signals, each a position-normed z-score, blended into one composite — and swaps out the one piece that doesn't carry over to defense:

• ATH — athleticism. The same combine and pro-day battery as offense (speed, agility, lower-body power, strength), z-scored within defensive position. A 4.5 forty means one thing for a safety and something else entirely for a defensive tackle.
• DPOS — draft capital. Identical treatment to offense: pick number as a within-position z-score, the NFL's aggregate verdict compressed into one number. It stays the single strongest predictor on defense, too.
• Havoc share — production. The defensive replacement for AWP: a prospect's share of his team's total disruption — sacks, tackles for loss, forced fumbles, interceptions, and passes defensed. Havoc share travels across schemes and defensive quality where raw counting stats don't.

The weights are calibrated separately for defense — front-seven and secondary careers don't reward the same things — and just like the offensive weights, I keep them to myself. Everything past the weights (scores, deciles, hit rates) is out in the open.

What the deciles produced

Deciles get graded within position and draft year, D1 at the top through D10 at the bottom. Across defensive prospects from the 2004 through 2021 classes with a finished labeling window (n = 1,976), here's the startable-career hit rate by decile:

Defensive prospects, 2004–2021 classes, n = 1,976. A “hit” is a startable-caliber IDP career, benchmarked at the 80th percentile of positional scoring. Live query of the public database.

The slope tells the same story as offense: a D1 defensive prospect hits at 58.7%, roughly eight times the ~7% floor the bottom three deciles share. And no, the curve isn't perfectly clean at the very bottom — D8 through D10 all bunch up near 7% — but that's just small-sample noise in the tail, not me claiming D10 beats D8. Read the shape, not any one row.

A worked example

Here's a Decile 1 defender from the 2026 board — Dani Dennis-Sutton, DE, Penn State:

Live scores from the IDP Draft Board, 2026 class.

This is the defensive profile the model rewards: three independent signals — the workout, the NFL's investment, and the college tape — all pointing the same way. A one-trick edge rusher (elite testing, no production) or a stat-compiler on a bad defense (havoc puffed up by raw volume) can't get to D1 the way a guy backed by all three does.

The defensive board — 2,987 prospects

IDP DMX covers every defensive class from 2004 through 2026, across the five scored positions:

Live counts from the public IDP database. Total: 2,987 defensive prospects.

Front-seven and secondary prospects run through the same three-piece pipeline, just with position-specific norms — havoc share means something different for an edge rusher than for a corner, and the scoring accounts for that.

The 2004 line

IDP coverage starts at the 2004 draft class for an honest reason: clean, TOS-respecting defensive college stats don't exist in any public source before then. Rather than make up production, pre-2004 defenders would carry an ATH+DPOS-only score with the havoc piece left null and a reduced-confidence flag — so IDP DMX just starts where the real data does. The most recent classes (2022–2026) are scored but not labeled yet, because a career needs a few NFL seasons before you know how it turned out.

Using it

Flip the Draft Board over to its DEF view for the full IDP board — scores, deciles, and havoc-share components for every defensive class since 2004. Want the production metric in detail? See havoc share. For the offensive model this mirrors, see the DMX guide.

Frequently asked questions