Scoring and statistics
These are this site's own rules. They are not the rules of the Majority game, whose server-side details are not public.
What the numbers are
| Name on the page | What it is | What it is not |
|---|---|---|
| Valid site answers (n) | Answers this site accepted for a question and has not removed | A count of verified people — one person can use several browsers |
| Reference data | Answer counts from the question's source, frozen when imported, turned into shares | Votes on this site; they never enter the vote table |
| Score amount (B) | Site answers in a group plus the group's remaining reference weight | A number of players — it can be fractional |
| Score share | A group's score amount divided by the question's total | “The percentage of players” while reference data is mixed in |
How answers are grouped
- Your original text is stored as typed. It is never rewritten.
- A match key is made with safe changes only: Unicode form, full-width letters, extra spaces, curly apostrophes and letter case. Short all-caps words like US keep their capitals, so US and us stay apart.
- The key is looked up in this question only. If it equals a known answer or a maintainer-confirmed alias, it is recorded immediately. The result tells you how it was grouped.
- Otherwise we may suggest up to two existing groups. Suggestions come from looser comparisons (punctuation, accents, “the”, simple plurals, small typos). They show no counts and are not ordered by popularity.
- You choose. Pick a suggestion, or keep your wording as its own answer. Picking a suggestion affects only your vote.
We also keep a small list of confirmed spelling and word-form variants for specific questions. These can direct an imported misspelling to its established answer. They are our own rules, not aliases supplied by the source. We do not guess synonyms or choose matches by vote count.
Answers that differ by a negation (sugar / no sugar), an opposite (on / off), a number (2 / 20, 1.5 / 15) or a protected acronym (US / us) are never suggested as the same answer. A maintainer can only merge them with a second reviewer's confirmation, and it is logged.
Maintainers can merge two whole groups of the same question or undo a merge. A merge never moves a vote: every vote stays on the answer it was cast for, and a group's total is the sum of its members. Undoing restores the previous totals exactly.
How reference data fades out
For each question, with n valid site answers and h the highest number of valid site answers it has ever had:
w = max(0, 1 − h / 30)
B(group) = site answers in the group + 20 × w × reference share of the group
score share = B / (n + 20 × w)
| h = n | Weight w | Reference amount | Denominator |
|---|---|---|---|
| 0 | 1 | 20 | 20 (reference only) |
| 1 | 29/30 | 19.33 | 20.33 |
| 15 | 1/2 | 10 | 25 |
| 29 | 1/30 | 0.67 | 29.67 |
| 30 or more | 0 | 0 | n |
Worked example Reference: McDonald's 50%, Wendy's 30%, Subway 20%. After 10 site answers (Wendy's 6, McDonald's 3, KFC 1), w = 2/3 and the reference amount is 13.33. Score amounts: Wendy's 6 + 4 = 10, McDonald's 3 + 6.67 = 9.67, Subway 2.67, KFC 1; total 23.33. Wendy's leads with 42.9%, and a McDonald's answer earns 9.67 ÷ 10 = 96.7 sync on this question.
Because h never decreases, a question that reached 30 valid answers stays on site answers only, even if some answers are later removed. If a question with no reference data has no valid answers, it cannot be scored and we say so instead of showing 0%.
Stars and sync
- Star: your group's unrounded score amount equals the highest one. Ties for first all earn the star. Two groups that both display 20.0% are not tied unless their exact amounts are equal.
- Question sync:
min(1, your group's B ÷ top group's B), computed from unrounded amounts. - Day sync: the average of the five question values × 100, rounded only for display. A removed answer counts as 0 for that question.
| Question | Your group / top group | Shares | Sync | Star |
|---|---|---|---|---|
| Fast food chain | Wendy's 84 / McDonald's 152 | 21.0% / 38.0% | 55.3 | No |
| Before bed | Brush teeth 130 / same | 43.3% / 43.3% | 100 | Yes |
| Common car colour | White 108 / Black 110 | 36.0% / 36.7% | 98.2 | No |
| Yellow fruit | Banana 300 / same | 60.0% / 60.0% | 100 | Yes |
| Where keys go | Hook 95 / Hook and Pocket 95 each | 38.0% / 38.0% | 100 | Yes (tied) |
Average: (55.26 + 100 + 98.18 + 100 + 100) ÷ 5 = 90.69 → sync 91, 3 stars.
Group comparison
“Higher sync than X% of valid complete runs” appears only on a settled day that meets all of these:
- At least 30 valid complete runs: five answers accepted before the cutoff, none removed, browser not blocked.
- All five questions have retired their reference data and each has at least 30 valid site answers.
- The day is not a practice day.
X counts runs with a strictly lower unrounded sync, divided by all valid complete runs, rounded down. Equal scores are not counted as beaten, and 0% can appear. We never estimate this from single-question numbers or from the source's counts.
Cutoff and settlement
| Time (UTC) | What happens |
|---|---|
| Day D, 00:00 | Questions open. Your score is provisional and can change as others answer. |
| Day D+1, 00:00 | Cutoff. Only answers our server had accepted before this moment count. No grace period. |
| Day D+1, about 00:10 | Settlement freezes the groups, counts, reference weights and results as data version v1. |
If you had not finished before the cutoff, your accepted answers still count for that day, but the run is not a complete official run and does not extend your streak. You can replay the day as practice once it is settled. If settlement is delayed, the page says “Settling” and keeps provisional numbers labelled as provisional.
History and corrections
- Before the cutoff, a maintainer's merge, undo or removal of abusive answers updates provisional scores.
- Between the cutoff and v1, nothing that affects counts can change.
- After settlement, a real correction publishes a new version (v2, v3 …) with a public reason. Earlier versions are kept, and your record and the day's page both move to the new version.
- If a question turns out to be seriously broken, the whole day becomes a practice day: no official scores, and streaks skip it rather than break.
- Scores you copied earlier keep the numbers they had; the day's page shows the current version.
- Replays keep the frozen counts and grouping of the version they started with. New guesses can also use our current confirmed variants to find an existing group. This does not rewrite old answers, completed scores or the frozen distribution.
Limits of the data
- A browser identity is not a verified person; clearing storage or using several devices creates new identities.
- The source's public answer lists include no aliases, and we cannot prove they are complete.
- Suggestions use simple text rules. They can miss a real match or offer an unhelpful one, which is why a suggestion is applied only when you choose it.
- Reporting a grouping problem does not change any vote by itself. A maintainer reviews it.
All parameters
| Questions per day | 5 |
|---|---|
| Initial reference amount | 20 weight units per question (not 20 players) |
| Reference retires at | h = 30 valid site answers, per question, permanently |
| Day boundary | 00:00 UTC (08:00 Beijing), by server acceptance time |
| Settlement | About 10 minutes after the cutoff; not extra answering time |
| Group comparison | ≥ 30 valid complete runs, all questions retired with ≥ 30 valid answers |
| Suggestions | At most 2, plus “Keep my wording” |
| Answer length | 1–60 characters |
| Text normalization version | N1 |