The two halves of the shelf are also different cinema, which is what makes the split worth drawing rather than just worth stating. Films inside a taken set are 73% subtitled against 49% outside one, with a median of 4,943 IMDb votes against 11,288. The box sets are where the collection keeps its older, more obscure, more foreign material.

Weight the same five measures by how often each film was actually taken and one row changes direction. Subtitled films are 58% of everything reached, above the shelf's own 56%, and only 38% of what guests chose alone. The box sets that were taken most often are the subtitled ones, so counting picks rather than films puts the middle column on the other side of the shelf. That reversal is a fact about how often those sets were taken, not about how many subtitled films they hold.

Two rows are worth the space precisely because they find nothing. A film over two and a half hours long is reached as often as a short one (-0.7 pt), and a film directed by somebody who has stood in the closet themselves is reached at the shelf's own rate (+0.2 pt). Length does not put guests off, and being one of them does not help.

Everything on that chart is one attribute at a time, and the attributes move together: an obscure film is likelier to be foreign, older and deeper in the collection all at once. The one claim this site publishes from that list is the language one, so it is worth reading inside cells of fame and era rather than on its own.

The penalty is not spread evenly across the shelf: it is concentrated in the films nobody has heard of. Among the least known films a subtitled one is reached far less often, and among the best known the difference changes sign. A famous subtitled film is reached about as readily as a famous English one, which means the shortest true version of the finding is not that guests avoid subtitles but that they avoid obscure subtitles. Neither reading escapes the larger problem, which is that the shelf as it stands today is not the shelf the early visits stood in front of.

The decomposition counts films: chosen on their own, arrived only inside a box set a guest took, or never reached. Box-set membership is known only for the 102 sets somebody took, because the pipeline collects sets from the visit pages, so a film inside one is reached by definition and never_in_a_taken_set is zero by construction rather than by measurement. 1 taken set could not be opened, so its films are not counted as members. The ladder ships twice. In ladder.films one row is one film in every column, which is the comparison that holds across all three. In ladder.events reaches and direct picks count one row per visit-film edge, so a film chosen many times weighs many times, while the shelf column still counts each film once because a shelf has no picks to weight it by: that column answers what a pick looks like, not what a chosen film looks like. Each measure carries its own denominator, since ratings, runtimes and languages are missing on different films. Nothing here says packaging caused a choice: box-set contents differ from the rest of the shelf by era, language, director and fame at once, and these are those differences, not an explanation of them.

One attribute at a time, over the films where that attribute can be observed: 'without' means observed and absent, never absent-or-unknown, so each row carries its own eligible count and the collection-wide baseline is context rather than the comparator. The interval on each difference is Newcombe's, because the mark is a difference; the two Wilson intervals are there to be read on their own. This is marginal association and not an explanation: fame, spine, era and language move together. The box-set row is flagged by_construction, because the pipeline learns of a set from the visit that took it, so a film in a known set is reached by definition. Every rate is over the collection as it stands today, which has had fewer visits to be picked in at the high spine numbers.