方法THE METHOD
Deterministic and explainable. No machine learning, no black box: every film's position can be justified in a sentence, which is what makes it defensible as the ranking rather than an arbitrary one. About 120 lines of Python with no dependencies, and the whole substrate ships to your browser so you can rebuild it under your own weights.
手順THE PIPELINEsix steps, in order
STEP 1
正規化NORMALISEEvery score becomes a 0–1 fraction of its own scale, so a 5-point Letterboxd average and a 100-point Tomatometer stop being different units.
STEP 2
縮小SHRINKEach score is pulled toward its source's franchise mean in proportion to how few people produced it: (n·x + m·prior) / (n + m), where m is that source's median sample size. A film rated by 200 people cannot swing as hard as one rated by 200,000. This is what handles Godzilla 1985's 194 Filmarks ratings and Destoroyah's 6 critic reviews.
STEP 3
標準化Z-SCOREWithin each source, so the source's own scale and generosity drop out. What is left is where this film sits among Godzilla films, on this source. This is the atomic unit — every slice the site offers is a different average over exactly these numbers.
STEP 4
地域平均AVERAGE WITHIN A REGIONSources sharing a region are averaged, so IMDb and Letterboxd together are one voice, not two.
STEP 5
地域集計AVERAGE THE REGIONSOne region, one vote. Not weighted by sample size — that would hand the ranking straight back to the huge English-language samples and undo the point. Not weighted by population either: Douban is not a sample of 1.4 billion people.
STEP 6
合成COMBINE THE PILLARSAudience, critics and critic lists, at the chosen weights, renormalised over whichever pillars that film actually has. Missing is never imputed as zero.
地域THE SIX REGIONSand why there are six
A source must cover at least 30 of the canonical 38 films to be a pillar input. Below that bar, films would be scored on different recipes — so Metacritic, eiga.com and AlloCiné's press score are carried and displayed but do not feed the ranking.
| SOURCE | REGION | COVERAGE | ROLE |
|---|---|---|---|
| Douban | China | 38 / 38 | pillar input |
| FilmAffinity | Spanish-speaking | 38 / 38 | pillar input |
| Filmarks | Japan | 38 / 38 | pillar input |
| IMDb | International | 38 / 38 | pillar input |
| Letterboxd | International | 38 / 38 | pillar input |
| Rotten Tomatoes | United States | 38 / 38 | pillar input |
| AlloCine | France | 35 / 38 | pillar input |
| Metacritic | — | 10 / 38 | display only |
| eiga.com | — | 10 / 38 | display only |
| AlloCine Presse | — | 6 / 38 | display only |
This is the honest reason there are six and not twenty: they are the platforms with the coverage, not a designed sample of the world. Germany, Korea, India and Brazil are missing because Godzilla rating data at that scale does not exist there, not because they were judged less interesting. Presenting six equal columns as a world panel would claim a representative set that does not exist, which is why the split lives on its own page and the row carries only the best and worst.
重みTHE THREE PILLARSand why the house weights are 70/15/15
The lists pillar is critic-derived — all eight full-franchise lists are journalist-authored. So critics are counted twice, once as aggregate scores and once as ranked lists. The honest axis is not audience/critics/lists but audience voice versus critic voice, and a 1/3–1/3–1/3 split is not neutral: it is 33/67 in favour of critics.
| PRESET | AUDIENCE VOICE | CRITIC VOICE | |
|---|---|---|---|
| critic-led | 33% | 67% | a slice |
| balanced | 50% | 50% | a slice |
| audience-led — official | 70% | 30% | the house ranking |
The house ranking is audience-led: 70% audience voice, 30% critic voice. The argument is representativeness, not the shape of the output.
| BEHIND THE COMPOSITE | COUNT |
|---|---|
| Audience ratings behind the composite | 8,753,046 |
| Critic reviews | 2,176 |
| Journalist-list writers | 8 |
| Ratio | 4,023 : 1 |
A 50/50 split would hand 2,176 critics and 8 list writers the same weight as 8.75 million ratings. Even at 70/30 the critics remain enormously over-represented per head. This is deliberately not the argument first made for audience-led — that it said something the English-language web does not already say. That was an argument about wanting a particular result, and it was correctly rejected as a basis for a default. This one holds regardless of what the ranking produces.
What is given up, knowingly: the critic pillars carry real information the audience pillars do not, particularly about the older films. balanced and critic-led are first-class slices, one click away on the ranking.
除外WHAT IS DELIBERATELY NOT IN THE SCORE
| EXCLUDED | WHY | STILL AVAILABLE AS |
|---|---|---|
| Box office, attendance | Popularity, not quality | a separate axis, and on every film page |
| Screen time, height, body counts | Colour, not quality | film-page content |
| Awards | Only five films have any — too sparse to rank on | film-page content |
| Metacritic, eiga.com, AlloCiné presse | Under the ≥30-film coverage bar | display columns |
注意THE ONE GOTCHAand the checkbox that turns it into a control
Filtering the film set does not renormalise the z-scores. That is deliberate: a Showa-only view still answers how good is this among all Godzilla films, not among Showa films. Both are reasonable questions and they have different answers, so the site does not pick one for you — Rescore within this subset in the control panel switches between them. It is off by default, because the default should be the ranking the rest of the site is talking about.