How scoring works
Points reward obscurity. This is the full derivation — the fame model, the logarithmic points curve, the six bands, and why the game never shows you a percentage.
The short version
Every valid answer to a cell is worth between 5 and 200 points. The more obvious the answer, the fewer points it pays. Your board score is the sum of the nine cells, minus 25 for each hint you took. Higher is better.
The six bands
Point values are grouped into bands, which drive the colour of a filled cell and the emoji in the grid you copy when you share a result.
| Band | Points | What it means | |
|---|---|---|---|
| 🦄 | Mythical | 100+ | A genuinely obscure name that fits. Most boards never see one. |
| 💎 | Legendary | 75–99 | A deep cut — a short spell most people have forgotten. |
| 🌈 | Epic | 55–74 | A good player who is not the obvious answer for this pairing. |
| 🟪 | Rare | 40–54 | Off the beaten track without being obscure. |
| 🟦 | Uncommon | 25–39 | A recognisable name, but not the first one that comes to mind. |
| 🟩 | Common | 5–24 | The answer most people reach for first. Still correct — just cheap. |
These bands describe the point value of an answer. They are not a claim about how many real people would pick it — see below.
Where the number comes from
Each player in the dataset carries a fame figure between 1 and 100, derived at build time from how substantial their public record is. Within a single cell, the game models how likely each candidate is to be the answer somebody reaches for, in proportion to that fame raised to the eighth power.
The exponent is doing real work. A gentler curve was fine when the dataset held a couple of hundred players, but it collapses at this scale: shares have to sum to 100% across every candidate in the cell, so across a few hundred answers even a household name ends up statistically indistinguishable from a complete unknown. Sharpening the curve keeps the difference between "the obvious pick" and "a deep cut" meaningful.
That modelled share is then converted to points on a logarithmic scale — roughly ten points for every halving of the share. Logarithmic rather than linear because the interesting range spans four orders of magnitude, and a linear mapping would flatten every rare answer into the same handful of points.
Finally a small deterministic wobble of up to 8% is applied, seeded from the board and the player, so the same pairing does not feel numerically identical every time it appears. Being seeded means it is reproducible: the same board always awards the same points to the same answer, for everyone.
Why you never see a percentage
Earlier versions of this game displayed rarity as something like "only 4% of players chose this". That figure was a simulation. The site does not collect per-answer statistics from real players, so it had no way of knowing what percentage of anybody picked anything — and presenting a modelled number in the visual language of a measured one is simply a false claim about data.
So the percentage was removed everywhere it appeared and replaced with a point value. The model still runs underneath, because it is a reasonable way to rank obviousness. It just no longer pretends to be a survey. A point value is honest about being a scoring rule.
Hints, and what they cost
A hint sketches a career — clubs, era, honours, the shape of the thing — without naming the player and without repeating anything the cell already tells you. It costs 25 points off the board total, taken whether or not you then find the answer. A board score can never go below zero.
Streaks
Completing the day's daily board extends your streak by one. Missing a day resets it. Archive replays and practice boards deliberately do not count: if they did, a streak would measure free time rather than consistency, and anybody could backfill a perfect year in an afternoon.