Fact-based findings on the constants, equations, and values behind this site — including the ones that turned out to be wrong.
The Same Number Twice
This site could always tell you how likely a prospect was to fail, never how likely he was to be great. The star-odds number built to fix that is well calibrated and survives every test that matters — and it ranks players in almost exactly the order the failure number already did, at a rank correlation of −0.999. Covers how a boom was defined, why hitters and pitchers kept different rates on purpose, and what a real second opinion would actually require.
Where the Data Begins
For a few hours, the best prospect in the history of this database was Clint Hurdle — a 19-year-old whose only record is one hot 1977 season, because his real career started before this database's coverage does. The model wasn't wrong about what it could see; it just didn't know what it couldn't. Covers the fix, why it took three separate pieces, and the badge you may now see on a player's page.
What the Comps Already Knew
This site picks a prospect's twenty comparables using his power and his speed, then throws both numbers away before predicting anything — which looked like the biggest unclaimed improvement in the system. Handing them back made the model worse. Covers why a statistic can be genuinely predictive and still useless, the one version that did work and the −3.25-WAR career it promoted into the top twenty, and a result that reverses depending on which half of history you train on.
The Survivor's Median
Seventy percent of Wade Boggs's closest minor-league comparables reached the majors, and the ones who did were scored as if a real, unglamorous career were a disaster — because the model only ever asked the players who made it. Fixing that made Boggs's own number much better. It also made his rank fall, and the reason why is the actual subject of this page.
Worse Than Guessing
Every player page prints the odds a prospect reaches the majors. For the older half of the database those odds had decayed until they were less accurate than ignoring the player and saying “everyone is at 58%” — and nobody caught it, because the tool built to check the model rebuilt the model before checking it. Covers settings that expire, why 45% of modern careers are invisible to a fair test, and a fix that changed nothing but the description.
Punished for Making Contact
Wade Boggs hit .325 for eighteen years and this site ranked him twenty-one-thousandth — not underrated but a whole type the index read backwards, because it tracked walks and strikeouts but never looked at them together and never looked at batting average at all. Covers the two minor-league stats that were missing, the ones that were measured and cut, and an honest account of a separate fix that later moved his rank again.
The Glove That Kept Disappearing
A shortstop who can play shortstop is worth more than a first baseman who hits the same; the site did not know it, and four straight attempts to teach it came back empty. The fifth found it — because the signal only appears once value is split into whether a player reaches the majors and how good he is once there, and position belongs only to the second. Covers how position finally entered the Spectral Index.
Ranked on Someone Else's Career
The site once called a 301-batter arm the best pitching prospect in its history — ahead of Clemens, Verlander, and Gooden — because his single best comp was Roger Clemens and the model was quietly borrowing Clemens's career to rank him. Covers the pitcher profile-quality term behind the Spectral Index, and why fewer minor-league innings can't be allowed to look like more talent.
Grading the Model's Own Misses
The model rated Wade Boggs below average and loved a pitcher who never threw a big league inning. Measuring that requires putting a prospect score and a finished career on one scale — five attempts at which failed, one by ranking 21.7% of all major leaguers below players who never reached at all. Covers the Eclipse Index on the Blind Spots page.
What Counts as an Average League?
Every stat here is measured against its league, and that league average was carrying a three-year memory. In 2026 the Southern League hit for 52% more power than the year before and the site did not notice. Covers a constant that turned out to be answering three different questions at once — and a fix that did not improve the Spectral Index.
Ranking Farm Systems
How thirty individual prospect scores become one number per organization. Three metrics were built and thrown away first — including one that matched the published rankings better than the version that shipped. Covers the System Value column on the Teams page.
When the Obvious Number Is Wrong
A prospect whose comps all reached the majors is not a 100% bet, a 28.7% strikeout rate can be better than league average, and the 20th-closest comp is sometimes a Hall of Famer who does not belong there. Three cases where the obvious reading was misleading — plus the ideas that were tested with the same rigor and thrown out.