The Probability Rink
Each puck is a draft slot, 1 through 32. Its height shows the chosen metric. Move the spotlight slider to inspect a pick; toggle metrics to see two very different stories. Drag to orbit.
drag to rotate
What the historical data says
| Question | Answer |
|---|---|
| How often do mocks nail pick #1? | ~75–85% — consensus top prospects (Bedard, McDavid, Celebrini) are near-locks |
| Exact-match rate by pick #10? | ~15–25% — one trade or one surprise reach scrambles everything downstream |
| Exact-match by pick #20+? | Under 10% — essentially coin-flips among a 15-player pool |
| Whole first round exactly right? | Effectively 0% — it has never happened for any major outlet |
| Do the players even work out? | Studies of NHL drafts find only ~50–65% of first-rounders play 200+ NHL games; by round 3 it's under 25% |
So mocks are unstable predictions of unstable predictions: even the actual GMs — with scouts, combine data, and interviews — miss on a third of first-rounders. Expecting a blogger's mock to be "close" misunderstands the difficulty of the problem.
How to judge a forecaster fairly (in any sport)
- Score against chance, not perfection. With ~50 plausible first-round names, random guessing gets ~2% exact matches per pick. A mock averaging 20% is genuinely skilled — even though it "looks wrong" 80% of the time.
- Use distance metrics. Serious analysts score mocks by average slot error (e.g., "players were picked 4.2 slots from prediction") — much more informative than hit/miss.
- Keep score over time. The tweet's real complaint — "I wish I could look back at all these old mocks" — is the correct instinct! Forecast accountability (Brier scores, calibration tracking) is exactly how meteorology and election forecasting improved. Archive predictions, grade them annually.
- Beware hindsight bias. After the draft, the result feels like it was obvious. Before it, 32 front offices with millions in scouting budgets disagreed with each other. If it were predictable, they wouldn't.