What is DMX (Dynasty MetriX)?

Data Dictionary · the complete guide to the DMX rookie score

Why it matters

Rookie drafts are won on your priors…what you believe about a guy before he ever takes an NFL snap. The problem is, most rankings run on film takes and off-season hype, and they move every time somebody drops a new mock. DMX doesn't work that way. I score every prospect the same way going back to 2000, so you can check the math yourself. It won't tell you who to fall in love with. It tells you what happened to 25 years of guys who looked just like this one, once they actually hit the NFL. History, not hype.

How DMX is calculated

Three pieces, and each one is a z-score against other prospects at the same position:

To get the final DMX score, I weight those three pieces by position and blend them. The weights aren't a guess — they're fit against 25 years of actual career outcomes, and I refit them every few years as more careers finish. I keep the exact weights to myself, and here's why: anybody can average three z-scores. Figuring out which blend actually predicts NFL production is 25 years of work, and it's the one thing I don't give away. Everything else is out in the open — the scores, the deciles, the regressions, the calibration — all on the Model Transparency page.

What the deciles mean

The raw DMX score is a continuous number, but I also drop every prospect into a decile — ten quality bands within his position, cut off the full history rather than off his own class. So what does each bucket actually produce? Here's every offensive prospect from the 2001 through 2020 classes, grouped by decile, with what their careers became (the Star / Starter / Bust tiers come from value-over-replacement, ranked within position):

Offensive prospects, draft classes 2001–2020, n = 2,026. Star / Starter / Contributor / Bust tiers from career VBD percentiles within position cohort.

Don't get hung up on any one row — read the slope. A D1 rookie is about 16× more likely to turn into a Star than a guy down in D8–D10, and better than three in four D10 prospects flat-out bust. No model calls any single player for you. The whole point is stacking these odds in your favor, pick after pick, year after year. Do that for a decade, and that's how dynasties get built.

One honest note about the middle of that table. The top and the bottom sort hard — D1 hits four times out of five, D10 hits once in thirteen. But look at D6 through D9: 14.2%, 12.6%, 14.1%, 12.1%. That's four deciles inside a two-point band, and D8 actually edges D7. I'm not going to dress that up. Once a prospect lands in the bottom half, the model is telling you "unlikely," and it is not reliably telling you much more than that. The precision is at the ends.

A worked example

Let me show you a real one. Here's Jeremiyah Love, a Decile 1 running back from the 2026 class:

Live scores from the DMX Draft Board.

This is exactly the kind of profile DMX is built to catch. His testing is nothing special on its own. But the NFL spent big capital on him and he produced at an elite level for his age — two signals pointing the same way. That's what the score rewards. A workout warrior with no production and no draft capital can't fake his way into D1, and that's the point.

The dataset behind it

A model is only as good as the history behind it, and this is where DMX pulls away from the one-season stuff. The database runs 3,100+ offensive prospects across 27 classes back to 2000, plus 2,900+ defenders back to 2004. Every one of them is matched across his college stats, combine numbers, draft record, and NFL career — then audited, so the same guy is the same guy in every table. That last part is unglamorous, and it's most of the work. Public football data is a mess: duplicate names, transfers, guys who share a name. Build your model on a bad join and you're quietly scoring the wrong career, and you'd never know it.

That history is also what makes the hit-rate table up above worth anything. To even quote a hit rate, you have to wait five NFL seasons to see how a class turned out. So a model that launched on two years of data has no track record — it can't, by definition. Mine goes back to 2001. And where there are gaps, I say so: clean defensive college stats just don't exist before 2004, so the early IDP classes get lower-confidence scores instead of me making up numbers to fill the hole.

How well does it predict?

Honestly, and out in the open. DMX explains somewhere between 10% and 42% of what happens to a career, depending on the position and what you're measuring. If that sounds low, it isn't. Remember what a pre-draft model can't see: which coach he lands with, whether the scheme fits, injuries, who's ahead of him on the depth chart. None of that exists on draft day, and it caps what any model can do. For reference, draft capital on its own runs a 16–22% R² against career VBD. The blend beats it at every position when the target is career tier — Star vs Starter vs Bust, which is the call I actually care about. On raw VBD magnitude, though, draft capital alone is a shade ahead at running back, tight end and receiver. I'd rather print that than bury it. It tells me the position weights are due a refit against the current component scales.

Every regression behind those numbers is published on the Model Transparency page — the coefficients, the R², the sample sizes, the years — along with calibration plots by position and a straight list of what the model struggles with (era drift, small-school prospects, the pre-2004 IDP gap). If you're the type who wants to check my work, you can reproduce the regressions from the public scores. The underlying NFL data comes from sources like nflverse.

A few things DMX leaves out on purpose: landing spot, injury history, character stuff. It's a clean draft-day starting point. Use it with your own read on a guy's situation — just don't let it replace it.

DMX vs DPX

DMX locks in on draft day and never moves after that. The second a rookie starts taking real NFL snaps, I hand him off to DPX (Dynasty Performance Index) — a separate in-career model built on volume, scoring, and efficiency that updates every season. Simple way to think about it: DMX tells you what to pay on draft day; DPX tells you what he's worth right now.

Using DMX on the platform

• DMX Draft Board — every class since 2000, scored and ranked, with the component breakdowns and career outcomes.
• Model Transparency — the regressions, calibration, methodology, and where the model falls short.
• Analytics — residual analysis, historical comps, and the Player Profiles archetypes.

Frequently asked questions