How Signals grades prospects.

Prospect evaluation is one of fantasy football's hardest forecasting problems. Every other model we build gets to lean on NFL evidence — real routes against real corners, real workloads in real offenses. A prospect has none of that. He has a college resume, a stopwatch, a birthday, and an incoming draft slot. From that thin file, dynasty managers are asked to price rookie picks, win devy leagues, and decide whether the 1.02 is a franchise cornerstone or a famous mistake.

We answer it the same way we answer everything: no film-watching, no favorites, no vibes — a single backtested grade per prospect, built only from what is actually knowable about him today, validated against how hundreds of real prospects' NFL careers actually turned out.

And this guide isn't just for the devy die-hards. If you play devy, you're literally drafting these players while they're still on campus — this grade is your draft board. If you're a dynasty manager, the incoming rookie class and every rookie pick you own or trade for are priced by it. And even if you only play redraft, your league drafts rookies too — the same grade feeds how we project what a first-year player can give you right now. Whether you're shopping three years from the NFL or three weeks from your own draft, this is the model doing the pricing.

And here, more than anywhere else in Signals, the override matters. Prospect evaluation is where conviction lives — everyone remembers the rookie they banged the table for. Our grade is the most rigorous starting point we know how to build, but if you believe in a player, pin him where your board says and every pick value and trade verdict follows you.

Here's the short version, then the deep dive:

Every prospect gets one grade, and that grade is the currency everywhere. It's built only from pre-NFL information — production adjusted for age and competition, breakout timing, recruiting pedigree, athletic testing, size, charted college data, and projected or realized draft capital — and validated against real NFL outcomes with a hard no-hindsight rule. That one grade powers the Prospect rankings board, prices every rookie pick in dynasty, values every devy stash, seeds every young player's dynasty value, and informs rookie projections in redraft — so a devy freshman, a rookie 1.03, and a second-year NFL player all sit on one comparable scale.

One grade, everywhere it matters

The first thing to understand about our prospect model is what the grade does. It isn't a standalone scouting toy — it's load-bearing across the whole platform:

• The Prospect rankings board is the grades in ranked order, by position, with a confidence reading on each.
• Rookie pick values come from the grades: as our dynasty guide covers, a pick is priced as the prospect you should expect at that slot, so the 1.03's value is the third-best grade in the class, priced for your league.
• Devy leagues run on the same pipeline — a college player you can roster is valued through the identical grades, so a devy stash and a rookie pick are directly comparable, on one scale, with no translation layer.
• Young NFL players carry their grade into the dynasty model: for a player's first couple of pro seasons, the grade fills the gap where NFL production history is still too thin to judge him on.
• Rookie redraft projections consult it too: the redraft engine's rookie forecasts lean on draft capital plus the prospect grade — at the positions where the backtest proved the grade predictive — to size year-one roles and production.

One grade, spent five ways. That's deliberate — the moment rookies, devy players, and picks get valued by different systems, they stop being comparable and every trade involving them gets murky. It's also why we keep saying there's no separate rookie board: the class is already ranked, everywhere, by the same number.

The no-hindsight rule

The signature discipline of this model is the strictest one we run anywhere at Signals: a prospect grade may never use information from beyond the moment it's issued.

No NFL outcomes. No NFL production, no dynasty market values, no "everyone knows how this turned out." When we backtest a grading idea, we replay history and grade each prospect using only what was actually knowable about him at the time — then check the grade against the career that followed. A model that peeks at the future aces every backtest and fails every real draft; ours is never allowed to peek.

The same discipline governs what else stays out. Expert prospect rankings and devy consensus boards are useful — we use them as a benchmark to audit our own outliers — but they are never a model input and never a target. A model trained to agree with the consensus can only ever be the consensus, minus a lag.

What goes into the grade

Every prospect is graded on what's currently known about him, and only that. The main ingredients:

• Production, adjusted for age and competition. Raw college stat lines lie constantly, in both directions. The same thousand-yard season means something completely different from a 19-year-old sophomore versus a 23-year-old fifth-year senior, and against SEC defenses versus a light mid-major schedule — so every season is scored on a continuous competition spectrum with the player's age attached. Monster numbers against weak competition get deliberately damped; the history of gaudy small-school stat lines demanded it. And the profile the model loves most — dominant young production against strong competition, with the pedigree to match — earns extra conviction, because that combination almost never lies. Think of what made Ja'Marr Chase's age-19 LSU season different from a career year by a 23-year-old: same box score category, completely different signal.
• Breakout timing and class context. When a player arrived matters nearly as much as what he did. Breaking out as a true freshman is one of the strongest signals in the sport; declaring early is information; needing a fifth year to produce is information too, in the other direction.
• Recruiting pedigree. Where the recruiting industry ranked a player out of high school, aggregated across multiple services. Here's the finding that surprised us: we assumed pedigree should fade once real college production accumulates — a senior with three years of tape shouldn't need his high-school stars anymore. The backtest said otherwise: pedigree keeps predicting NFL outcomes even for upperclassmen, so it stays in the grade for everyone (more in the reject pile).
• Athletic testing — where it actually predicts. Measured testing feeds the grade for quarterbacks, receivers, and tight ends, where it demonstrably sharpens the forecast. Running backs are the deliberate exception: we tested it, and testing numbers simply didn't predict RB outcomes well enough to earn a seat. A model that includes an input at one position and excludes it at another isn't being inconsistent — it's reporting the evidence.
• Size, against position-specific thresholds — a lens, not a verdict.
• Charted college data. The licensed, professionally charted dataset from our other guides extends to the college game, and it does the same job here: separating what a player created from what his scheme handed him, which is precisely the distinction a projection to the NFL needs.
• Draft capital — projected, then real. Where NFL teams spend picks is the single most durable prior about a young player's opportunity, so the grade uses projected draft capital for players who haven't been drafted yet, and swaps in the realized pick the moment they are. And capital leads: a prospect with real first-round capital outranks a recruit with an identical college profile and no draft verdict yet, on purpose. Ashton Jeanty is a clean illustration of the ingredients negotiating: monster production, but against lighter competition — the competition lens tempers it, and then the NFL settled the argument by spending a top-10 pick, which the grade respects.
• Context and confidence. Team offensive context, transfer-portal moves, and returning-production pressure round out the picture. And because college data goes missing far more often than NFL data, every grade carries an explicit confidence reading based on how complete the player's file actually is — shown to you on the board, and priced into his value rather than hidden.

How the grades are validated

A prospect grade is a prediction, so it gets judged like one. Every version of the grading model replays against years of historical prospect classes — under the no-hindsight rule — and is scored on how well it anticipated real NFL careers.

And here's the part that shapes everything: we grade the model on difference-makers first. A prospect model tuned to maximize overall "hit rate" of becoming a WR2 learns a cheap trick — favor safe, high-floor profiles, collect partial credit, and miss every superstar. We believe that kind of model is suboptimal for dynasty, because leagues aren't won by safe players who manage to put together one good season of RB1 or RB2 scoring; they're won by the picks that become elite. So our promotion gates treat elite outcomes (top-six positional seasons) and generational outcomes (top-three) as first-class targets. If a proposed model change improves the average but regresses the model's ability to identify true difference-makers, it is rejected, full stop, no matter how good the aggregate looks. We'd rather be right about the next Amon-Ra St. Brown than slightly less wrong about everyone else.

The same gates check for position damage — a change that helps receivers isn't allowed to quietly break tight ends — and the same "seat isn't tenure" rule from our other models applies: inputs get re-tested as the model improves, and the ones an upgrade makes redundant are retired.

Draft weekend changes everything, and the model treats it that way

The single biggest information event in a prospect's life is the draft itself. Until that weekend, the model carries a projection of his capital; the moment his name is called, the projection is replaced with the truth, and the class re-grades around it.

This is where prospect evaluation gets humble, and we think that's a feature. NFL teams employ hundreds of full-time scouts, run private workouts and interviews we'll never see, and then stake real jobs on their conclusions in public. When that verdict disagrees with a college profile, the verdict carries serious weight. The grade doesn't surrender to the draft — a college resume still argues — but it updates hard, because pretending draft position is just another opinion is how you end up dying on hills the league already bulldozed.

The honest counterweight: capital is a prior, not a prophecy, and the outliers cut both ways. Puka Nacua was a fifth-round pick — no capital-respecting prospect model on earth had him where his career immediately went. That's exactly why the grade is a starting point that hands off to the dynasty model's leading indicators the moment real NFL usage exists: the prospect grade gets a player onto your radar honestly, and the weekly signals take over the moment there's something real to measure.

An honest word about distance

Prospect grading degrades with horizon, and we'd rather tell you how than pretend otherwise:

• The upcoming class is graded on the richest file — full college careers, testing, and mature draft-capital projections.
• Classes further out get progressively thinner: draft projections barely exist two-plus years ahead, underclassmen have shorter resumes, and the youngest devy stashes are graded mostly on pedigree, early production, and age — which is genuinely predictive, but wider-banded. The confidence reading exists precisely to keep that honesty visible on every row.
• The youngest prospects move the most. A freshman's grade updating sharply after his first real season isn't the model being fickle — it's the file finally having pages in it.

That's the deal with devy: the further from the NFL you shop, the bigger the discounts and the wider the error bars. The model prices both, and shows its confidence rather than hiding it.

How we keep ourselves honest

Same rules as every Signals model, tuned to this one's temptations:

• No hindsight, ever. Every backtest grades prospects on what was knowable at the time. The one carve-out (athletic testing) is a carve-out precisely because it isn't hindsight.
• Difference-maker capture is a gate, not a nice-to-have. No model change ships if it regresses elite or generational identification, regardless of what it does for the averages.
• Expert boards audit us; they never teach us. When our grade diverges hard from respected devy and prospect consensus, that player gets reviewed by hand — sometimes we find a data gap, sometimes we find conviction. Both outcomes improve the model; neither turns the consensus into an input.

From the cutting-room floor, prospect edition

The reject pile, as always — because the graveyard is where the credibility is:

• Phasing out recruiting pedigree for upperclassmen. The most intuitive idea we've killed: surely three years of college tape should retire a player's high-school stars. Tested, and rejected — it turns out almost all of the NFL's generational prospects had elite recruiting pedigrees. Pedigree kept predicting NFL outcomes even for seniors, and removing it made the grades worse. The recruiting industry's collective evaluation of a 17-year-old contains real information that college box scores never fully replace.
• The teammate-dilution boost. "His numbers would be huge if he weren't sharing the field with four other pros" — the classic prospect argument for the buried talent on a loaded roster. We built it and it didn't predict. Lower production while sharing a great room turns out to be nearly as often a warning as an excuse, and the boost got rejected.
• Athletic testing for running backs. Included everywhere it earned a seat, excluded at the one position where it didn't. RB testing numbers made headlines and not predictions.
• Leaning harder on RB receiving production. Weighting pass-catching more heavily made the average RB grade more accurate — and systematically demoted the generational pure rushers, the exact players the difference-maker gate exists to protect. The same lesson our redraft model learned with its own version of this lever, failed independently in both places. Averages lie; gates don't.

We keep this list for the same reason as the other guides: anyone can tell you why their rankings are right. We'd rather show you the ideas we killed to get here.