How we rank books
Every number on this site is reproducible. Here is the whole method.
Where the data comes from
| Open Library | catalogue + ratings | CC0 / public domain | live |
|---|---|---|---|
| Wikidata | series, awards, translated titles, screen adaptations | CC0 | live |
| Wikipedia | summaries + popularity | CC BY-SA | live |
| Google Books | ratings + descriptions | API terms | live |
| Hardcover | ratings, new books, upcoming titles | API terms | live |
| New York Times | bestseller history (not scored) | API terms | live |
Weeks on the New York Times bestseller list are shown on a book's page but never enter the score. Sales measure marketing and distribution as much as quality, and mixing them in would quietly turn a quality ranking into a popularity ranking.
Screen adaptations come from Wikidata, where the film carries a "based on" property pointing at the book. They never touch the score — they say a book was filmed, not that it is better. Pornographic entries and unreleased projects are dropped, and the order is set by how many language editions wrote about the film, so the one a reader actually knows comes first.
How a book gets a category
A book enters a ranking only when two independent sources agree on its category — the Open Library subject index (corroborated by the work's own subject list), Wikidata genres and Google Books BISAC codes. One source is not enough: it keeps a mislabelled record from putting a novel in the psychology ranking.
Classics is the exception, because being a classic is not a genre. A genre is a claim about what a book is, and two catalogues can make it independently; being a classic is a claim about what happened to a book afterwards, and no catalogue records it — Google returns “Fiction”, “Readers” or “Foreign Language Study” for Alice in Wonderland, The Time Machine and Nineteen Eighty-Four. So the second signal here is the book itself: it is in Open Library's classics index and it was first published at least 50 years ago. Both are checkable and neither is an opinion.
The formula
A raw average rewards obscurity: five readers who loved a book would outrank a classic held by a hundred thousand. We use a Bayesian weighted average that pulls every title toward the category mean until enough readers have voted.
score = (v / (v + m)) · R
+ (m / (v + m)) · C- v = total ratings for the title
- R = weighted mean rating across sources, rescaled to 0–10
- m = 147 (prior weight, in ratings)
- C = mean score of the category (global fallback: 7.80)
Minimum 50 ratings to enter a ranking at all. The prior weight is the median rating count of the current snapshot, so it tracks how much evidence actually exists rather than being fixed in advance.
How series are scored
A series score is the rating-weighted mean of its volumes, with a penalty when only part of the series has enough ratings — so a saga cannot ride on one beloved first book.
Series plan: how we work out stages and breaks
Few people read a long series in one go. On a reading-order page we split it into stages and suggest a short book from outside the series between them. Everything comes from data shown on the page: the pages and ratings of each volume. The table shows what we compute, from what, and by which rule.
| What we compute | From which data | Rule |
|---|---|---|
| Reading and listening time | Page count of each volume | Reading: 1.1 min per page. Listening: about 28 pages per hour of audio, an estimate. The default pace is 45 min a day; the slider on the page changes it and everything is recomputed. A volume with no page count is counted at the series average and marked ≈; a side piece (such as volume 0.5) with no page count, and volumes not yet published, are left out. |
| Chance of finishing | Ratings of the first and the last volume | The number of ratings of the last volume published at least 2 years ago, divided by that of volume 1. Only volumes with a score count (enough ratings to pass our threshold), and volume 1 must have one; the result is capped at 100%. From 55% the chance is high, from 30% medium, below that low. |
| Cliff | Number of ratings of consecutive volumes | The largest drop in the number of ratings between two consecutive main volumes (whole-number places, so side pieces such as volume 2.5 are left out), if it is at least 25%. We look from the step between volumes 2 and 3 onward (the drop after volume 1 is normal and is covered by the chance of finishing), and only at volumes published at least 2 years ago that both have a score. The plan ends a stage before that volume when the stage before it already holds at least half a standard stage. |
| Dip | Score of each volume | A volume scored at least 0.25 points below the one before. We also name the volume where the scores recover. |
| Stage | Volume pages and your pace | A standard stage is about 3 weeks of reading at your pace, at least 15 h and at most 40 h. A real stage holds whole volumes, so it can be shorter (when the next volume does not fit) or longer: it may grow by 30% if the volume added has fewer than 450 pages, by 10% if it has 450 pages or more. A volume of 900 pages or more is a stage of its own. |
| Size of the break | Stage hours against a standard stage | A full break (a book of up to 340 pages and 2 days off): after a giant volume, before a cliff, or after a stage at least 20% longer than standard. A full break also always comes before a volume of 900 pages or more. A short break (a book of up to 180 pages, 1 day): after a stage of at least 85% of standard. After a shorter stage there is no break: it is merged with the next one, unless together they would run past 130% of a standard stage; then there is a short break. |
| Book for the break | Our ratings catalogue | A standalone book, not a volume of any series, from a different shelf than the series: 80 to 340 pages with a score of at least 8.0 and at least 500 ratings, or 80 to 180 pages with a score of at least 7.8 and at least 300 ratings. Never a picture book, a children’s book or a middle-grade book, and never the same one twice. After a stage longer than 25 h we pick a lighter one, e.g. humour, short stories or biography. We only show books that have a page in your language. |
| Shape of the series | Average scores of the beginning, middle and end | We compare the beginning, middle and end of the series and name one of eight shapes: “even”, “it gets better”, “the finale is the best part”, “a dip in the middle, then the form returns”, “peak in the middle”, “a drop at the end”, “best at the start” or “uneven”. Each shape comes with one sentence on what to expect. |
These are estimates from numbers, not a review. The plan appears when at least 3 volumes, and two thirds of all of them, have a page count. Page counts and ratings come from the sources described above.
How often it updates
Ratings are re-fetched every 3 months and stored as a dated snapshot, which is how the trend arrows and “biggest risers” are calculated.
Known limits
Open rating data skews English-language and skews recent. Translated editions are merged into one work, so a book's score reflects readers across every language, not just yours.