By decade

The index compares each decade's share of single-film picks with its share of the collection. Box sets are excluded from it: taking the Bergman set is a decision about Bergman, not about the 1950s.

By country

Every country with at least 10 films

Country	In collection	Picks	Reached for	Films reached
United States	582	1497	1.50	81%
France	277	314	0.66	69%
Japan	216	166	0.45	62%
United Kingdom	142	274	1.13	71%
Italy	88	180	1.20	76%
Sweden	66	39	0.35	83%
Germany	40	56	0.82	60%
Hong Kong	38	13	0.20	79%
Iran	30	16	0.31	37%
Canada	19	38	1.17	79%
Cuba	18	9	0.29	17%
India	18	11	0.36	56%
Czechoslovakia	17	26	0.90	94%
Mexico	16	43	1.57	94%
Soviet Union	16	55	2.01	75%
Spain	15	37	1.44	80%
Denmark	14	23	0.96	100%
Belgium	11	12	0.64	91%
Poland	11	30	1.59	64%

Criterion's listing gives one country per film, so a co-production counts only under whichever country Criterion credits first. Countries with fewer than 10 films in the collection are left out.

By trade

Criterion says what each guest does for a living, and it turns out to predict the one thing you would expect it to least: whether they walk out with a film in another language. Guests who direct and do not act reach for subtitles 19 points more often than guests who act and do not direct. The guests who do both sit between them, which is the part that makes it an ordering rather than a coincidence.

Share of picks in a language other than English, by trade

the shelf 56%

acts, does not direct 29%

acts and directs 37%

directs, does not act 48%

0%20%40%60%80%

single-film picks films reached, box sets opened

Table view

Share of picks in a language other than English, by trade

	single-film picks	95% interval	Picks	films reached, box sets opened
acts, does not direct	29%	26% to 32%	864	45%
acts and directs	37%	32% to 43%	313	59%
directs, does not act	48%	44% to 51%	760	68%
the shelf	56%	·	·	·

Two things worth reading off the second lens. Everyone reaches for subtitles less often than the shelf offers when they choose a single film, and directors overtake the shelf once box sets are opened. The box sets are where the subtitles arrive.

The same share, year by year

acts, does not direct

22%

2022 2026

acts and directs

38%

2022 2026

directs, does not act

33%

2022 2026

Every panel is drawn to the same scale, topping out at 80%. The rule across all three is the shelf, 55.6%.

Table view

The same share, year by year

Filmed	acts, does not direct	acts and directs	directs, does not act
2022	33% of 36	·	35% of 40
2023	46% of 85	46% of 24	50% of 156
2024	30% of 205	33% of 57	49% of 163
2025	28% of 256	35% of 103	45% of 181
2026	22% of 257	38% of 86	33% of 51

Guests who act are a recent arrival: there were seven visits by directors and none by an actor in 2012, and the two are near parity now. So the pooled comparison above is also a comparison of eras, and this is the control. What survives it is the gap between the two ends: in every one of these years, guests who direct reached for subtitles more often than guests who act. The three-way ordering does not survive. The middle group is the smallest in every column and it sits a whisker under the actors in 2023 and over the directors in 2026, so read it as corroboration and not as a third result. Years need at least 25 single-film picks from both ends of the divide to earn a column, and the last one is a part-year, so its final point is the least settled figure on the chart.

Every trade, counted as many times as it applies

The three groups above are exclusive. This table is not: a guest described as an actor and a director is both, the same way a film is several genres at once. Trades are read off 307 of the guests and 129 carry no description to read.

A trade is read from Criterion's own description of the guest and only on a solo visit, so a pair filmed together carries neither person's job. The three groups in the divide are exclusive; the table above them is not, because a guest described as an actor and a director is both.

By genre

Genres are TMDB's, over the 1720 of 1736 collection films that carry one (99%). A film carries several at once, so these shares sum past 100%.

The canon or the new releases

A spine number is roughly release order, so low numbers are the films Criterion built its name on. Guests land almost exactly in the middle: the mean pick sits at the 49% mark, against the 50% a blindfolded guest would manage.

Over the 1263 films that carry a spine number. The other 473 are non-spine releases and are excluded rather than treated as spine zero. That excluded quarter of the collection is described here in full, because a chart silently leaving out 27% of the shelf should be able to say what it left out.

By language

The trade comparison above reads language as a yes or no. The field is not: 41 languages, and the shape inside it is sharper than the binary. Guests do not avoid subtitles in general. They under-reach for the two largest subtitled shelves Criterion has, French and Japanese, and over-reach for Spanish and Russian.

The index compares each language's share of single-film picks with its share of the collection. Both shares are taken over the 1719 films that carry a usable original language: TMDB's `xx` means no language, used for silent films and undetermined records, and it is treated as unknown here exactly as trades.py treats it, so neither side of the ratio can absorb missing data. Box sets are excluded from the index, as they are from every other index on this site, because taking the Varda set is a decision about Varda rather than about French. Languages with fewer than 15 films in the collection are left out. The index measures selection against the collection as it stands today, not against what was on the shelf when a given visit was filmed.

How far back

The median film chosen on its own is 46 years old, and that has not moved: across the years carrying enough picks to compare, the rank correlation between the year of a visit and the median age of its films is -0.047 with a p-value of 0.8796 over 2,000 permutations. What does move is the median year: 1961 in the first visits, 1981 in the most recent. The series keeps reaching about the same distance back, and the distance travels with the calendar.

Single-film picks only, and an age is the visit's year minus the film's year, because Criterion's listing carries a year rather than a release date. Quartiles are shown only for years with at least 25 picks.