How the rankings are computed

Canon Archive does not review anything. Every ranking on this site is computed from published, authoritative lists — and every step of that computation is documented here, including its weaknesses. If you disagree with a number, you should be able to see exactly where it came from.

1. Sources

Each domain aggregates a fixed roster of published lists: critics' and practitioners' polls, institutional selections, canonical anthologies, festival and award records, library data and large public votes. The full roster, with every weight, is in the tables below. Sources that are themselves aggregations of other lists (TSPDT, The Greatest Books, Newsweek's Meta-List) are excluded to avoid double counting. Regionally scoped lists are reserved for future regional canons rather than mixed into the global ranking.

2. Position score

Within a list of length N, the item ranked r earns s = (N − r + 1) / N — first place earns 1.0, last place approaches zero. Unranked lists (a common form: canonical anthologies, unordered polls) give every member the midpoint score s = (N + 1) / 2N ≈ 0.5. Award and curation records count as membership signals at a flat s = 0.5.

3. Eligibility — no penalty for absence you couldn't avoid

Every list has an explicit scope (era, and eventually language and genre) plus an implicit one: a work released after a list was published cannot appear on it. A work outside a list's scope is excluded from both the numerator and the denominator of its score. The Godfather is not penalized for missing a 21st-century list, and a 2019 film is not penalized for missing a poll from 2000.

4. Scope calibration

Being #1 of the 21st century is not being #1 of all time. Ranks from scoped lists are converted to global equivalents by fitting a monotone (isotonic) mapping between that list's ranks and the positions its works occupy in an anchor ranking built from unrestricted lists only, then re-scored against a reference pool of 1,000. The mapping is re-estimated twice, against the updated aggregate, until stable.

5. Time-tested damping

A canon is, by definition, what survives. An appearance on a list published shortly after a work's release carries less canonical information than one made decades later, so each appearance is damped by up to 50% when the gap between release and list publication is small — scaling linearly to full weight over a window sized to each medium's history: 50 years for books, 25 for film, 15 for TV. Games are exempt: published game canons already skew retro, and in a technology-driven medium age is not a guarantee of quality. Recent works still enter the ranking; they simply have to keep earning their place as the years pass.

6. Final score

Score = ( Σ w·s·t + m·C ) / ( Σ w + m )

w = source weight (published tier, see tables below)
s = position score after scope calibration
t = time-tested factor
C = mean score across the candidate pool (Bayesian prior)
m = prior weight — 6.0 for film, 4.0 for books/TV/games
Σ runs over the lists the work is eligible for; being absent
from an eligible list contributes s = 0 to the average.

The Bayesian term shrinks works with thin evidence toward the pool average, and a work must appear in at least three lists to be ranked at all.

7. Known limitations

• Entity resolution currently matches on normalized titles plus year and author signals. Cross-language titles and edition variants can still split or miss; migration to identifier-based matching (TMDB, Wikidata, Open Library) is in progress.
• Formal scope is corrected for; de facto cultural bias is not. A UK public vote is formally open to all works but skews British — this is mitigated only by source diversity and tier weights, and it is disclosed rather than hidden.
• Language-scoped award records (Booker, Pulitzer, Goncourt) are excluded until per-work language metadata allows fair eligibility handling.
• Some mirrors preserve only part of an original list (Time Out 1995: top 25 of 100); such sources are treated as the shorter list they actually are.
• Tier weights are editorial judgments. They are published, applied uniformly, and open to argument — which is the most any weighting scheme can honestly claim.

8. Every list we aggregate

Weights are editorial tiers, published in full: 3.0 for large international expert polls, 2.0 for professional selections and canonical anthologies, 1.0 for editorial picks, public votes and behavioral data, 0.5 for award records. List length, ranked-versus-unranked form, and era scope are handled separately by the formula — the weight expresses authority only.