What 5,000 films say about themselves

00 · How to read this page

What this page is for

Explore only works if its tags lead somewhere. Two tags that never meet give an empty page; two that always meet give hundreds of films and no decision. This page measures both.

Part one describes the catalog and how Explore feels, in words. Part two puts numbers and charts to it, each with a note on choosing a film tonight. Part three is the technical detail.

Every number comes from the same data file the site is built from.

01 · The library

What kind of library this is

Built around directors, not franchises. More than half the catalog comes from directors with at least three films here, so if one Hitchcock or Herzog lands, more are waiting. Sequels are rare and earn their place on their own.

Deep in older film. Nearly a quarter of the films predate 1970, back to the Lumière brothers in 1895, shelved alongside this year's releases.

International by default. About half were made outside the United States, so subtitles are part of an ordinary evening.

A large documentary shelf. Documentaries are the biggest theme, close to three films in ten.

The quiet themes are the most international. Reflective themes, like identity, comfort films and documentaries, are mostly made outside the US; loud ones, like big comedy and AI and aliens, are mostly American. A quiet film tonight will often come with subtitles.

02 · Using Explore

What Explore feels like to use

Big moods combine easily. Comfort Watch, Feel-Good or Makes You Think pair with almost anything and still leave a healthy list.

Specific tags dead-end. Archive Footage, War and Film Noir make good first taps but meet few other tags; War with a comedy mood leaves two or three films. Faded tags show this before you tap.

Near-twins barely narrow. Nearly every documentary is also Learn Something, and nearly every noir is Black & White. A useful second tag pulls in a different direction.

Three tags is where it settles. Two broad tags often leave more than a screen; a third usually brings it to one.

03 · Strengths and gaps

Where it shines, and where it's thin

Where it shines. American, European and British film of every era, Japanese and French cinema, documentaries, crime and noir, and films about the inner life. Picks here always have company.

Where it's thin. South Asia and Oceania hold about a hundred films each, many of them documentaries, and the smallest themes, like time loops and dystopias, about a hundred each. Those are the best places for nominations, through the green + in the header.

04 · At a glance

The catalog and Explore in four numbers

The catalog is wide, Explore's 60 tags make 1,770 possible pairs, and in simulated sessions about two in three three-tag searches land on one screen.

05 · The catalog

What's in the catalog

5,000 films from 1895 to 2026, chosen from critics' polls, festival records and canon lists: 1,414 subtitled, 1,462 documentaries, each in one of 16 themes.

The US is the largest source and Europe the second; South Asia and Oceania are the thinnest, which is where picks run out first.

Each film has one theme. Documentaries, identity and comfort films lean international; comedy and the high-concept themes lean American.

Choosing tonight. The deepest shelves are American, European and British film since 1990. For older or further-afield films, start broad with Classic or Subtitled and add one mood.

06 · Two tags

Pairs: from too narrow to too broad

Explore offers 60 tags, so there are 1,770 possible pairs. 395 of them (22%) return fewer than 3 films and 158 return none. The median pair returns 11 films.

The empty pairs are contradictions, not gaps, such as:

The broadest pairs return several pages each: Documentary + Learn Something (1,376), Documentary + Makes You Think (511) and Classic + Black & White (497).

Most are near-synonyms. The specific tags have the opposite problem, as the chart shows.

Choosing tonight. Make the second tap change direction: Comfort Watch alone returns over four hundred films; add British and it's about sixty.

07 · Try it

Combination explorer

Pick two or three tags to see the exact count and the films. After two, every possible third tag is listed with its count; those under 3 are dimmed.

Choosing tonight. Pick the two things you agree on, then see which third tags still have films.

08 · Every pair at once

Pair heatmap

Each square is one pair of Explore's 60 tags, most common first. Red is fewer than 3 films; deeper green is more. Hover or tap for the count; a tap loads the pair into the explorer in section 07.

Big tags are green almost all the way across. The red clusters are the specific tags, and the documentary rows rarely meet fiction moods.

09 · Three tags

Three tags, and why not four

There are 34,220 three-tag combinations among the 60, and most are empty because the tags never meet.

People don't pick at random, so 20,000 simulated sessions start with a popular tag and add tags in proportion to the counts on the chips.

4,329 three-tag combinations return 3 or more films, and 692 return 12 or more. With four tags, only 2,290 return 3 or more.

Simulated sessions

A fourth tag isn't the fix. 72% of three-tag searches that stay broad include Documentary, Learn Something or Archive Footage, which nearly always appear together. Folding those twins and sorting by score would bring most onto one screen.

10 · The vocabulary

Explore's 60 tags

The catalog uses 2,074 distinct tags. Films average 7.4 each, and Explore's 60 cover every film. The 60 are curated, not simply the most common, and synonyms are merged first, so every count is honest.

11 · All films

120 tags, two at a time

All films adds 60 specific tags (subgenres, settings, eras, techniques) to Explore's 60 and takes two at a time, for questions like "a Cold War spy thriller". A third pick replaces the older one.

Specific pairs return a focused handful or nothing: Spy Thriller + Cold War returns 28, Heist + Caper Comedy 18, Cold War + Journalism 1. Empty pairs fade out before you tap.

Of the 7,140 possible pairs, 38% return one screen of 3 to 48 films and only 3% return more, against 11% of Explore's pairs, so All films doesn't need a third tag.

The six groups on All films:

• Genre & subgenre: 48 tags, 23 of them new
• Mood & feel: 16 tags, 6 of them new
• Style & craft: 21 tags, 8 of them new
• Themes & subjects: 22 tags, 15 of them new
• Setting & era: 4 tags, 4 of them new
• Language & region: 9 tags, 4 of them new

Decades, people and one-off tags are left out, and synonyms such as Biography and Biopic are merged.

12 · Tag hygiene

Clean the vocabulary next

1. Near-twins in the cloud

2. A long tail nobody can reach

1,196 tags are used once, 513 of them director names, and only 69 reach 100 films. People belong in their own facet, where search can still find "Kurosawa".

3. Language tags should be a facet of their own

With 56 languages, language works better as a filter, like Region, than as tags in the cloud.

13 · The writing

What the descriptions say

4,079 films have a written description (about 254,851 words, 377 characters on average); the other 921 still show a placeholder. The written "why it matters" notes average 259 characters. The house voice holds: 2,303 descriptions introduce the cast with "plays".

Each description is three sentences: the premise and leads, one true detail, and why it suits the viewer. "Why it matters" covers awards, craft and influence.

2,760 written descriptions (68%) use "For your love of …". The hooks are varied (1,482 different ones), but the sentence shape is not. The next editorial pass should vary it.

Read twenty cards in a row and that closing line starts to sound like a template. Varying it is the biggest improvement the writing could get.

851 black-and-white films, 308 tagged Oscar Winner and 27 Palme d'Or winners.

14 · Plan

Recommended changes, in order

Not recommended yet: a fourth Explore tag (see section 09).

Ordered by impact: placeholders are the most visible problem, the next two fix broad results cheaply, and the editorial pass can run alongside.

15 · Method

How this was measured

Tags are compared lower-cased after the site's synonym merge, and a combination matches a film that carries every tag, as on Explore; section 11 uses All films' 120 tags. Word counts use written descriptions only, without common stop words. Every number comes from films.csv.

The simulation uses a fixed random seed, so it repeats exactly.