Update · 9 August 2026. Since this paper was published, the Spectral Index has been redefined. It is no longer a single blended score of reach-probability and median outcome; it is now a prospect’s expected career value — the probability he reaches the majors multiplied by how valuable he projects to be if he does — shown as four numbers: Confidence, Impact, Bust risk, and Boom chance. Where this page treats the Spectral Index as the older composite, that describes how the model worked at the time; the pitcher profile-quality term covered here still feeds the new index. See Punished for Making Contact and The Glove That Kept Disappearing.
What this page is
The Spectral Index is the site’s one-number summary of a prospect — a blend of how often similar players reached the majors and how good the outcome was when they did. For pitchers, one half of that “how good the outcome was” number was quietly measuring the wrong thing, and it produced a result absurd enough to be worth walking through in full: the single highest-rated pitching prospect in the site’s entire history, ahead of Roger Clemens, Justin Verlander, and Dwight Gooden as prospects, was a college draftee who had faced 301 batters in the minor leagues. This page is the story of why that happened, the fix, and the one trap that would have quietly made the fix worse if it had been ignored.
The best pitcher ever, on 301 batters faced
Kade Anderson is a real, very good pitching prospect — around fifth overall on most 2026 industry lists. He is not the best pitching prospect who has ever lived, and no one thinks he is. The site did. Of every pitcher-season it has ever scored — more than thirty-four thousand of them, going back to the late 1970s — Anderson’s had the single highest Spectral Index. He outranked Gooden, Verlander, and Clemens’s own minor-league lines.
The reason was mechanical. The outcome half of a pitcher’s score leaned heavily on one number: the ceiling of his comparable players’ careers — specifically, the 90th-percentile career value among the handful of historical pitchers most similar to him. Anderson’s profile is genuinely elite, so his closest comps were themselves excellent, and the ceiling of that small group happened to be Roger Clemens’s actual career. So the number that was supposed to describe how good Anderson’s own future might be was, quite literally, Roger Clemens’s finished résumé — borrowed off a comp list and pinned onto a pitcher who had thrown about 82 professional innings at a single level.
This wasn’t a rounding problem or a display quirk. Measured against real outcomes, that borrowed-ceiling number turned out to carry essentially no information about whether a pitcher would even reach the majors, let alone how good he’d be — in a model predicting which prospects became genuinely good big-leaguers, its contribution was indistinguishable from zero. Yet it accounted for about a third of the whole Spectral Index. The all-time top of the pitcher list was a museum of this: Jeff Suppan, whose actual career was 2,543 innings of below-replacement pitching, sat inside the top ten as a prospect, because his comp pool’s ceiling was some other legend’s career too. Only about five of the top twenty-five had genuinely good big-league careers.
The important thing is what was right about the old number. Anderson’s rate stats really are elite — top-of-the-scale strikeouts, a genuine workhorse-starter workload. The model was not wrong to be excited about him. It was over-confident, and the over-confidence came from borrowing hundreds of innings of someone else’s proven career to stand in for a pitcher who hadn’t thrown them yet.
The fix: judge a pitcher on his own profile, and stay humble about thin evidence
The change re-sources that outcome number from the pitcher’s own minor-league profile instead of his comps’ careers — his strikeout rate, his starter workload, his walk and home-run rates — and then does something the old number never did: it shrinks that profile score toward the middle in proportion to how little evidence stands behind it. A pitcher who has faced 300 batters gets pulled toward average harder than one who has faced 1,500, because the 300-batter sample is simply a noisier estimate of who he really is.
The effect on Anderson is exactly what it should be. He falls from first of all time to the high thirties — still an excellent prospect, near the top of the board, but no longer improbably ahead of every pitcher in history. His score now says “elite rates, honest confidence interval” rather than “Roger Clemens.” And the pitchers who should have been near the top, but weren’t, rise: Pedro Martínez, whose own minor-league profile was outstanding but whose comp-pool ceiling was modest, climbs from outside the top 1,600 to inside the top 20. Clemens and Félix Hernández move into the top ten on the strength of their own lines rather than borrowed ones. Across the settled all-time top 25, the share of genuinely good big-league careers roughly doubled.
The trap: fewer innings can’t be allowed to look like more talent
Here is the part that is easy to get exactly backwards, and it is the whole reason the fix is built the way it is.
Elite pitchers have fewer minor-league innings than ordinary ones — because they were too good to keep down and got promoted fast. Clemens cleared the minors on about 500 batters faced; Verlander on 463. Anderson has faced 301. If you build a “quality” number that treats low minor-league workload as a positive signal — a plausible-sounding move, since the greats do move quickly — you would rank Anderson above Verlander for having thrown fewer innings, which is precisely the over-confidence you were trying to fix, dressed up as a feature.
So innings are allowed to do exactly one job here, and never the other: they can only reduce confidence in a thin sample, never add to a pitcher’s quality score. “He climbed fast” is real information, but it already lives in the reach half of the Spectral Index and in the level-aware way comps are chosen. A pitcher with 301 innings and a pitcher with 1,500 innings can have the same elite rate profile; the difference between them is not that one is better, it’s that we are more sure about the other. That is the only thing the innings count is permitted to express.
What didn’t survive contact with real outcomes
Three ideas were tried on the way to this and discarded, kept here because an honest accounting should include the wrong turns.
Tuning the shrinkage to make Anderson’s rank look right. The strength of the shrinkage is set by a single constant — how much minor-league evidence it takes to be believed at close to face value. An early version of this constant was chosen because it left Anderson looking “strong but not buried,” which is a polite way of saying it was reverse-engineered from the answer we wanted for one player. That is exactly the failure the site’s own working rules warn against. The shipped constant is instead derived directly from the data’s own signal-to-noise structure, with no free knob to tune — it happens to land close to where the eyeball version did, but now it comes from a measurement rather than a preference.
A square root that was hiding an assumption. The shrinkage was first written with a square-root form that looked dimensionally innocent and turned out to be the wrong equation entirely — the statistics of a noisy average shrink linearly with sample size, not as its square root. The two forms disagree in a way that matters, and “both versions gave a similar constant” was a false reassurance, because they were measuring different things. It was caught not by any accuracy test — both versions ran fine and produced believable numbers — but by stopping to ask what the exponent actually meant. The same kind of check caught a second one: a raw coefficient in the quality formula looked six times larger than the strikeout term, which would have implied minor-league home-run rate mattered more than missing bats — until it was put on a common scale and turned out to be an artifact of the raw units, with strikeouts the strongest signal after all, as everyone had assumed.
Age as a quality signal. Being young for your level is a real, strong indicator — it earns its own adjustment elsewhere on the site. But added directly into this pitcher-quality number, on top of the rate stats, it contributed nothing measurable and was dropped. A term no one can defend on its own is the exact thing that caused this whole problem, so it didn’t get to stay just because it sounded reasonable.
Two honest limitations remain, and they’re worth naming. A command-first pitcher like Greg Maddux, whose minor-league game was contact management rather than strikeouts, still rates modestly under a strikeout-weighted quality number — a real blind spot, not a fix. And Verlander himself takes a real haircut: at 463 minor-league batters faced he genuinely was a thin sample as a prospect, and the shrinkage treats him accordingly. That is the fix working as designed, applied evenhandedly — the same humility that pulls Anderson down pulls Verlander down too, and it should.
The general standard
The change was checked the three ways every scoring change here is. It was walked through real, named players — Anderson, Clemens, Pedro, Suppan — start to finish. It was fit on one slice of history and tested on a different one it hadn’t seen, and held up in both directions. And, because any change that reshuffles a ranking can improve a metric just by shuffling, it was run against eighty randomized versions of itself, in which the quality profiles were scrambled across players. A random quality number didn’t just fail to help — it actively hurt, diluting the reach signal. The real one beat all eighty shuffles. None of that makes the number perfect; Maddux and Verlander above are proof it isn’t. It makes it honest about how much it actually knows.