Accuracy
We grade ourselves in public
Most fantasy advice is vibes with a thumbnail. LeagueSeer's predictions are backtested against real drafts, and the results (including the misses) live on this page. The numbers below are generated from the backtest ledger, not typed by a human.
5,147
Real drafts replayed
1,007,135
Picks replayed
274,615
Picks scored
Calling the next pick, before it happens
The backtest replays each draft pick by pick. Before every pick, the model predicts who that manager will take (from a shortlist of the 45 best available players) using their league history, roster needs, and market ADP. Then we check what they actually did. Latest wave, against a blind guess from the same shortlist:
| Call | LeagueSeer | Blind guess |
|---|---|---|
| Exact pick calledthe player taken was our #1 prediction | 21% | 2.2% |
| Pick in our top 3the player taken was one of our 3 most likely | 46% | 6.7% |
| Pick in our top 5the player taken was one of our 5 most likely | 60% | 11.1% |
Weighted across all 19 backtest waves: 19% exact, 42% top-3, 56% top-5.
The honest caveats
- 40% of real picks are off-board. Real drafters take deep sleepers, kickers in round 6, and their college roommate's favorite tight end. Those picks fall outside our 45-player shortlist entirely and are tracked separately. Counting them all as misses, our top-3 rate across all scoreable picks is 25%. Chaos is part of drafting; we measure it instead of hiding it.
- Only rounds 1–10 are scored. Late rounds are close to coin flips for everyone, and market data gets thin. We'd rather not pad our numbers with rounds where no model deserves credit.
- Past drafts, not promises. Every rate here is measured on completed drafts. Draft rooms shift every week of the preseason, which is why the backtest re-runs in waves and this page updates with the ledger.
Survival odds are calibrated, not guessed
The “73% chance he's still there at your pick” numbers in the draft room run through a calibration table fitted per position and league format on historical drafts, so when the room says 70%, it's because players in that situation actually survived about 70% of the time, not because a formula felt confident. Market spread data is calibrated against real Sleeper drafts too (the raw mock-draft spreads ran overconfident; we corrected them with measured outcomes).
Grade us on your own draft
After any draft, the room's report card replays it from your seat and scores every survival call and snipe alert we made against what actually happened: your league, your numbers, no cherry-picking.
Last backtest wave: July 9, 2026 · 43 drafts, 2,198 picks scored.