← Research · Above Replacement
For over a decade, Boris Chen's tier charts have been the closest thing fantasy football has to a shared language. Millions of drafters print them; whole leagues argue in his vocabulary ("he's a tier lower"). But almost nobody knows what the model actually does — so we read his published source code and rebuilt it, step by step.
It is simpler than its reputation. The pipeline:
We reran our replica on the same inputs as his published 2025 preseason board: 79% of players landed in exactly his tier, and 100% within one tier. Against his live 2026 board (published from slightly different snapshot timing), the replica agrees within one tier for 95% of players. For practical purposes, the model is fully recovered.
Two consequences follow from knowing the internals:
1. The tiers inherit everything from the consensus rank — including its blind spots. The model cannot know your league's scoring. A tier built on standard rankings is wrong in meaningful places for a league with unusual kicker, defense, or reception rules; the divergence is exactly where draft value hides.
2. Expert disagreement is not upside. A natural upgrade is to use the spread of expert opinion, not just the mean. We tested that on 1,106 rank-adjusted player-seasons: players experts disagreed about underperformed their rank expectation by 13–25 points with no fatter right tail. Disagreement flags bust risk, not breakout potential — so we surface it as a warning badge, never as a bonus.
| Check | Result |
|---|---|
| Same-input replay vs published 2025 board | 79% exact tier, 100% within one |
| Replica vs live 2026 board | 95% within one tier |
| Wide expert disagreement (top tercile), RBs | −25 pts vs rank expectation |
The replica powers the tier logic inside the Above Replacement draft tool, recomputed under each league's actual scoring rules. Method questions and corrections are welcome — this page exists to be checked.