Fermi Estimation Game
Answer Fermi questions with calibrated confidence intervals, tight when you know, wide when you don’t, and score on a log scale.
Settings
Statistics
Track your best scores and watch your accuracy climb across every session.
Sign up →Leaderboard
Last 30 days| 1 | TactfulAlpha | 1150 pts |
| 2 | Kahota | 1146 pts |
| 3 | PropHedger | 1110 pts |
| 4 | zero | 1092 pts |
| 5 | HedgedRaven | 1070 pts |
| 6 | Priyansh | 1068 pts |
| 7 | Potatoman67 | 1050 pts |
| 8 | AstuteArb | 1020 pts |
| 9 | 7vik | 1000 pts |
| 10 | ArbTrader | 942 pts |
Sign in to join the leaderboard and compete with others.
How to play
Objective
Estimate the answer to Fermi questions as accurately as possible. You start with 500 points and earn or lose points based on how close your guess is on a logarithmic scale.
How to play
Each question, you’ll see a Fermi estimation question with a unit of measurement shown above the answer inputs. Enter a low and high estimate to form a confidence interval. Scoring is based on the geometric mean of your interval. Overestimating and underestimating by the same factor are penalized equally.
Scoring
Your score depends on how close your estimate is and how wide your interval is:
- Hit (answer in range): base points × spread multiplier. Tighter intervals earn a bigger bonus (up to ×1.5), but very wide intervals (>10x) reduce your score (×0.8)
- Miss (answer outside range): you lose points, and tight misses are punished harder (overconfidence penalty). A wide miss is more forgiving
- Skip / timeout: −20 points
Base points (log₁₀ distance of geometric mean)
| Log₁₀ dist. | ≈ Ratio | Hit | Miss | Label |
|---|---|---|---|---|
| ≤ 0.05 | ~1.1x | +100 | −50 | Near perfect |
| ≤ 0.10 | ~1.3x | +80 | −40 | Excellent estimate |
| ≤ 0.20 | ~1.6x | +60 | −30 | Very close |
| ≤ 0.30 | ~2x | +40 | −20 | Good estimate |
| ≤ 0.50 | ~3x | +20 | −10 | In the ballpark |
| ≤ 1.00 | ~10x | +10 | −10 | Right order of magnitude |
| > 1.00 | > 10x | 0 | −10 | Too far off |
| — | — | — | −20 | No answer / Skipped |
Spread multiplier (interval width)
The spread multiplier depends on whether you hit or miss. Narrow intervals are high risk, high reward. On a miss, the overconfidence floor sets a minimum penalty based on how tight your interval was, even if your estimate was way off.
| Spread | ≈ Width | Hit mult. | Miss floor | Example |
|---|---|---|---|---|
| ≤ 0.30 | ≤ 2x | ×1.5 | −50 | 50 – 100 |
| ≤ 0.50 | ≤ 3x | ×1.2 | −30 | 40 – 120 |
| ≤ 1.00 | ≤ 10x | ×1.0 | −15 | 10 – 100 |
| > 1.00 | > 10x | ×0.8 | −10 | 1 – 1,000 |
The miss floor is the minimum penalty for that spread, you lose at least that much for missing, even if your estimate was way off. Being confident and wrong is always costly.
Strategy
This is a calibration game. Be narrow when you’re confident, be wide when you’re unsure. A tight hit (≤2x spread) can earn up to +150 pts, but a tight miss costs you dearly due to the overconfidence penalty. A very wide interval (>10x) is safer but reduces your score even on a hit (×0.8 multiplier). The optimal play is honest confidence, match your interval width to your actual uncertainty. Skipping costs −20, so always submit something.