Forget Netflix's Algorithm: These Fans Built Smarter Recommendation Systems With a Spreadsheet and Too Much Free Time

Somewhere in a Discord server you've probably never heard of, a person who goes by celluloid_cryptid is updating a color-coded spreadsheet that tracks 847 films by mood, cinematography style, runtime, pacing, and something they've labeled "emotional damage index." It took them three years to build. It has never once recommended Bird Box.

That might be the most compelling endorsement of DIY recommendation culture you'll ever hear.

Why the Algorithm Keeps Getting It Wrong

Here's the thing about Netflix, Spotify, YouTube, and basically every major platform's recommendation engine: they're not actually trying to find you something you'll love. They're trying to find you something you'll click on. Those are very different goals, and the gap between them is where a lot of smart, obsessive people have decided to plant a flag.

Algorithms are optimized for engagement metrics — watch time, click-through rate, session length. They learn from what the masses do, which means they're inherently biased toward the popular and the familiar. If you're the kind of person who wants to find slow-burn Taiwanese new wave cinema from the early '90s, or a horror podcast that only covers regional American folklore, or an R&B artist who sounds like what would happen if D'Angelo grew up in New Orleans instead of Richmond — the algorithm is not your friend. It's not even trying to be.

That vacuum is exactly what's driving a quiet but remarkably sophisticated grassroots movement of fans building their own recommendation infrastructure from scratch.

The Spreadsheet as Love Language

The tools these people use are almost aggressively unglamorous. Google Sheets. Airtable. Notion. The occasional custom-built web app that one person in the group coded over a long weekend. What they lack in polish they make up for in specificity — and specificity is the whole game.

Take the film community that's grown up around a shared Airtable base called (informally) "The Vault" — a collaborative database that about 200 contributors add to, tagging movies across dozens of custom fields. We're talking stuff like narrative structure (nonlinear, episodic, circular), color palette (desaturated, high contrast, warm), emotional register (melancholic, quietly devastating, weirdly funny), and a field called would recommend at 2am that is exactly what it sounds like.

The result is a recommendation engine that can surface films based on genuinely granular taste preferences. Want something that feels like Lost in Translation but has more plot? The Vault has a filter for that. Looking for a horror film that's scary but not gory, set in a rural location, made after 2010? Done. This isn't something Netflix can do. Not because they lack the data — they have infinitely more of it — but because they lack the intent. Nobody at a billion-dollar company is going to manually tag 3,000 movies with a would recommend at 2am field because it doesn't map to any monetizable metric.

Collaborative Filtering, But Make It Human

Some of the most impressive DIY systems borrow from actual data science — specifically, the concept of collaborative filtering, which is the backbone of most algorithmic recommendations. The idea is simple: find people with similar taste profiles and use their preferences to predict what you might like.

Big platforms do this computationally. These communities do it manually, and somehow it works better.

In several niche music Discord servers, members have built shared rating systems where you score albums across multiple dimensions — not just overall quality, but things like replay value, accessibility to newcomers, and how much you need to be in the right headspace. The community then uses those multidimensional ratings to match listeners to records based on their own stated preferences. It's slow, it requires real participation, and it produces recommendations that feel like they came from a friend who actually knows your taste — because functionally, they did.

One server focused on experimental and ambient music has been running a system like this for about two years. Members describe it as "the first time I've consistently gotten good music recommendations since I had a college roommate with good taste."

The Mood-Board Library Problem

One of the more fascinating developments in this space is the rise of what some people are calling "mood-based libraries" — personal or collaborative collections organized not by genre or release date or any traditional metadata, but by feeling.

This sounds fuzzy, but in practice it's surprisingly functional. One Letterboxd power user (who also maintains a public Notion database) has organized their entire film library around emotional states: need to feel something real, want to feel smart, want to be comforted, want to be unsettled in a productive way. Each category contains films that have been personally vetted and annotated. Friends and followers use it constantly. It gets updated. It has a changelog.

This is essentially what streaming platforms have tried to do with their vibe-based browsing features — Netflix's "Feel-Good Movies" rows, Spotify's mood playlists — but those always feel like they were designed by a committee that was afraid to commit to anything interesting. The DIY versions are opinionated in a way that's actually useful. They're built by someone who has a point of view, and that's the ingredient that no amount of machine learning has successfully replicated.

What This Says About Where We're Headed

The spreadsheet revolution isn't going to replace streaming platforms. That's not the point. What it represents is a fundamental shift in how a certain kind of audience thinks about discovery — less passive consumption, more active curation. Less trusting the feed, more building the infrastructure yourself.

There's also something genuinely communal happening here. These systems only work when people participate, contribute, and maintain them. They create accountability and shared ownership in a way that clicking "thumbs up" on Netflix never will. When your recommendation carries your name and your reputation, you take it seriously.

And maybe that's the real unlock: the best recommendation engine isn't a recommendation engine at all. It's a community of people who care enough to do the work.

Netflix has 13,000 employees and a research division dedicated to this problem. celluloid_cryptid has a spreadsheet and strong opinions. For a growing number of niche audiences across the US, the spreadsheet is winning.

Welcome to the vault.