Trusted Longest, Turned Out Shortest
PlayerGem's analytics shop traces how a corpus entry with apparent depth collapsed under inspection — and what that r...
Read itLatest Articles
The Corpus Entry We Trusted for Its Age
PlayerGem traces a specific failure: a corpus entry kept active because it was the oldest record we had, and what tha...
Read itWhen More Data Made the Rating Smaller
A PlayerGem corpus entry that looked strong at 40 games shrank to mediocrity at 120 — and the shrinkage was the corre...
Read itThe Corpus Entry That Aged Well and Taught Us Nothing
A PlayerGem corpus entry stayed accurate for two full seasons, survived every review, and still produced no usable si...
Read itThe Corpus Entry We Built Twice
PlayerGem's account of how a duplicate corpus entry went undetected long enough to skew shrinkage on a basketball pla...
Read itA Corpus Entry We Stopped Trusting
PlayerGem traces how a well-populated corpus entry lost the shop's confidence without a clear cause — and what that s...
Read itWhen the Corpus Grew and the Ratings Got Worse
PlayerGem added two seasons of historical data to its corpus and watched calibration scores drop — a case study in ho...
Read itWhen the Feed Quietly Changed Shape
How a silent upstream format change corrupted months of PlayerGem's corpus before the calibration logs finally caught...
Read itCleaning Names Is Most of Building a Corpus
How PlayerGem found that name-matching failures — not missing data — were quietly corrupting the corpus everything el...
Read itThe Corpus Entry That Was Correct and Useless
PlayerGem's corpus desk on how a statistically accurate player record can be built from the wrong unit of measurement...
Read itThe Corpus Entry We Kept Updating
PlayerGem's account of a corpus entry that kept receiving updates long after it should have been closed — and what th...
Read itThe Boring Half of the Work Is All of the Work
PlayerGem on how the corpus gets built, why it breaks, and what the shop learned from trusting a thin one too early.
Read articleThe Season We Trusted a Nine-Game Sample
A player with almost no history looked like our strongest candidate for three weeks. The corpus was working exactly as designed, which was the problem.
Read articleWhat Counts as Enough History
How much prior performance we require before we will rate a player at all, and the two years we spent getting that number wrong.
Read articleWhy a Proven Performer Outranks a Hot Streak
The oldest belief in the shop, the one piece of it that has survived every rebuild, and the two places where it is simply wrong.
Read articleAll The Articles
- Trusted Longest, Turned Out Shortest
- The Corpus Entry We Trusted for Its Age
- When More Data Made the Rating Smaller
- The Corpus Entry That Aged Well and Taught Us Nothing
- The Corpus Entry We Built Twice
- A Corpus Entry We Stopped Trusting
- When the Corpus Grew and the Ratings Got Worse
- When the Feed Quietly Changed Shape
- Cleaning Names Is Most of Building a Corpus
- The Corpus Entry That Was Correct and Useless
- The Corpus Entry We Kept Updating
- The Boring Half of the Work Is All of the Work
- The Season We Trusted a Nine-Game Sample
- What Counts as Enough History
- Why a Proven Performer Outranks a Hot Streak