Everything we can say about Korean casting rests on a median of five recorded names per title

We have published several things this week that come from one join: which Korean actor appears in which Korean title that reached a Netflix Top 10. Before any of it is used, here is what the join is actually made of.

A sixteen-episode drama is represented here by a median of five names. Seventy-five titles have exactly one person recorded. And 270 of the 906 Korean titles we matched carry no cast statement at all, so they are not in the join in any form.

That is the whole foundation: 3,415 casting slots, 1,355 people, 636 titles.

What thin records do to each figure

The useful thing about this particular gap is that its direction is knowable even where its size is not.

“48.3% of people appear in exactly one title” is a ceiling, not an estimate. Every missing credit turns a repeat worker into a one-timer. It cannot work the other way. The true share of one-timers is lower, by an amount we cannot compute.

“61.6% of multi-title actors work in both film and series” is a floor. A missing credit can only make a crosser look like a specialist. The true crossing rate is higher.

“96.8% of titles form one connected body” is also a floor. Adding a cast member can create a link between two titles; it can never remove one. More complete records would make the catalogue look more connected, not less.

“The busiest actors are film actors” is the one that could genuinely reverse. Series casts are long and recorded thinly — that is exactly where the missing names should be. If they were filled in, the series slot count would rise and the gap would narrow. We flagged that in the piece itself rather than here, because a limit that could overturn a finding belongs beside the finding.

Why we use it anyway

Because the alternative is not a better dataset. Netflix publishes no cast. No open source lists who appeared in Korean films and series with identifiers you can join on. Wikidata is thin, and it is the only thing there is.

So the rule we work to is: publish the figure as it comes out, state which way the error runs, and never apply a correction factor we cannot defend. A number with a known direction of bias is usable. A number silently adjusted towards what we expected is not.

What we changed to make the join usable at all

Until 7 August 2026 we had stored, for each actor, how many charting titles they appeared in — and not which ones. Every question of the form did this show move its cast was therefore unanswerable, and rebuilding the link took a full day of re-querying.

The join is now keyed on Wikidata Q-numbers for both the person and the title, not on names, because names change and identifiers do not. Matching these two sources on title text attaches cast to only 317 of 1,005 titles — chart names and article titles disagree constantly. On Q-numbers it reaches 636 of 906.

The file is in our data bundle as cast-title-join.csv, with a column dictionary that says what a blank cell means in each column, and a coverage file that carries every number above.