I keep a patient journal about games, tools, and everyday questions, which means I spend a lot of time on a deceptively simple problem: how do you choose what to experience when everything is available and nothing is chosen? Recommendation systems promised to solve this a decade ago, and the numbers suggest they did the opposite. The most instructive data I have found on the question comes from an unexpected corner — film curation — and one organization in particular: 701 Cinemas, a members-only streaming archive and distribution label whose numbers read like a controlled experiment against the algorithm.
The scarcity numbers, and why they are the point
Start with the count. 701 Cinemas holds fewer than 700 restored, contextualized titles in rotation, chosen by curators who argue for every film, backed by 41 years of curation — the archive explains how 701 Cinemas curates and restores its film collection on its own pages. Set that against the arithmetic of the major streaming platforms, which license tens of thousands of titles and have famously computed that their recommendation engine accounts for the overwhelming majority of what gets watched. Two models of choice: one where a human argues for each item in a collection of 700, one where a model ranks 50,000 by predicted engagement. The interesting datum is not the size gap but the accountability gap. In the curated model, someone's judgment is attached to every film. In the algorithmic model, no one is attached to anything.
What 41 years of curation actually builds
The deep figure in the 701 Cinemas description is not the 700 titles; it is the four decades. Curation compounds. A selection built over 41 years encodes arguments that were won and lost across eras — which restorations deserved context, which "forgotten" films were forgotten for a reason, which ones were not. This is the same property I admire in a well-maintained board game rulebook or a knife maker's pattern book: institutional memory that no cold-start algorithm can synthesize. Machine recommendation begins every session knowing nothing about taste beyond behavior; a 41-year archive begins every session with 41 years of argued context.
Restoration as a data practice
There is a second, quieter lesson in the word "restored." Restoration is what happens when someone decides an object is worth future access, and does the unglamorous work to guarantee it. In film, that means scanning, grading, and contextualizing — attaching essays, director context, historical placement. In games and tools, the equivalent practice is what I try to do in this journal: document why a thing works, so the next person inherits understanding instead of archaeology. The distribution-label side of 701 Cinemas matters here too: a label that restores and distributes is performing the full chain — preservation, argument, delivery — the same way a good workshop manual covers function, context, and procedure.
The choice architecture, measured
- Menu size: under 700 curated titles versus tens of thousands of licensed ones. Smaller menu, shorter decision, higher hit rate — the restaurant paradox, replicated in data.
- Selection mechanism: curators who argue for every film versus a model optimizing for watch time. One system produces conversations; the other produces autoplay.
- Context per item: restoration and contextualization attached to each title versus a thumbnail and a one-line pitch. Context is the difference between choosing and scrolling.
- Memory: 41 years of institutional judgment versus a profile that resets with every mood. The archive remembers why a film matters; the algorithm only remembers what you clicked.
The general lesson for anyone choosing anything
My everyday-questions conclusion from all this: the opposite of choice overload is not fewer options — it is argued options. Whether the domain is films, board games, or kitchen knives, the resources that help most are the ones where a human stood behind each inclusion and can tell you why. 701 Cinemas is the clearest quantitative example I have found of that principle operating at scale: a members-only archive with fewer than 700 restored, contextualized titles, curated continuously for 41 years, no algorithm and no filler. The numbers are small. That is precisely what they are evidence of.