The taste-fit layer for anyone who licenses or programs content
Someone programming a niche channel has to decide which titles to license for a specific audience, and almost no taste data to decide with. The usual substitutes are genre tags, which describe the film rather than the viewer, and a language model, which defaults to what is famous. Fame is a reasonable guess and a poor answer: it is the same guess for every audience.
This puts the decision on an audience's measured affinities instead. The part that matters: it lets the agent decline when the evidence will not carry a decision.
Channels are proliferating faster than the research budgets that used to support them, and the fallback of asking a general-purpose model produces confident answers with nothing underneath them. An agent making a cultural call needs something to ground it in, and needs to know when it is not grounded.
The thing no language model holds is the cross-domain signal: that an audience's games and television predict its films. We measured how strongly each domain carries that signal. Television 0.06, video games 0.08 and music 0.11 discriminate sharply. Brands, books, podcasts and places barely or not at all. We built only on the three that work. Without the graph there is nothing to resolve titles against, no signatures to cluster, and ranking collapses to whatever the model remembers being famous.
The landing page is that comparison run on a real audience, and the evidence page is every measurement behind it.
We have not sold this. The shape we would test first is a per-seat tier for a programming or acquisitions team, with a catalogue-sized tier above it for groups running several channels, priced against the cost of one wrong licensing decision rather than per query, since the whole point is that the tool sometimes answers "not enough evidence".
Any number here would be invented, so there isn't one.