Transparency
How ranking works
“Higher score rises.” But a single five-star vote should not beat fifty people averaging 4.6. Here is the maths we use.
Star ratings: the Bayesian average
Articles, book chapters, questions, wiki pages and tour companies are rated with 1–5 stars. Each member gives one rating per item and may change it. Lists are ordered not by the plain average but by this score:
score = (C · m + Σ ratings) / (C + n)
- m = 3.5 — the rating an item is assumed to have before anyone votes (prior mean).
- C = 5 — how many “virtual votes” that assumption is worth.
- n — the number of real votes; Σ — their sum.
| Case | Plain average | Score |
|---|---|---|
| 1 vote of 5 stars | 5.00 | 3.75 |
| 50 votes averaging 4.6 | 4.60 | 4.50 |
| 10 votes averaging 4.0 | 4.00 | 3.83 |
| No votes yet | — | 3.50 |
As votes accumulate the score approaches the real average; with few votes it cannot jump to extremes.
Up / down votes: the Wilson lower bound
Comments and the positions posted under questions are voted ▲ / ▼. They are ordered by the lower bound of the 95% confidence interval of the share of upvotes: “at the very least, what share of readers approve of this?”
| ▲ | ▼ | Share | Wilson score |
|---|---|---|---|
| 1 | 0 | 100% | 0.207 |
| 10 | 1 | 91% | 0.623 |
| 100 | 20 | 83% | 0.757 |
| 3 | 3 | 50% | 0.188 |
For positions, ties are broken by the star score, then by recency.
Questions: score + activity
The Big Questions list adds a small, saturating activity bonus to the star score:
score + 0.25 · log₁₀(1 + comments + positions)
The “Trending” sort
This optional sort lets the score decay with time: it halves every half-life (30 days for articles and questions), so content that is both new and good comes forward.
Likes and honesty rules
- Social posts and tour companies can also be sorted by number of likes; tour companies additionally by cost.
- Membership is required to vote, rate, like or comment; one vote per member per item.
- There are no fabricated ratings or reviews on this site. Entries marked “Demo” are sample data.
- Rate limiting, honeypot fields and moderation limit abuse.