I Used 'Coherence' as Bait: Which 'More Like This' Tool Actually Understood the Assignment?
There's a specific kind of streaming despair that hits around 9:40 p.m. on a Tuesday. You've scrolled past the same three rows of "Because You Watched" tiles four times. You know what you want — something small, tense, a little weird — and the algorithm keeps offering you a $200 million movie with a Hans Zimmer score.
So I ran an experiment. I picked one deliberately awkward seed film and fed it to six different recommendation tools to see which ones could handle it.
The Bait: A Movie That Breaks Lazy Algorithms
My seed was Coherence (2013), the micro-budget sci-fi thriller about a dinner party that goes sideways when a comet passes overhead. It's the perfect stress test for a similarity engine because it refuses to sit in one bucket:
- No stars, no budget, no marketing. Nothing for a popularity-based model to latch onto.
- Genre is a trap. It's sci-fi, sure, but it's also a talky relationship drama shot in someone's living room. Call it "sci-fi" and you get spaceships. Call it "drama" and you get something completely wrong.
- The vibe is the whole point. Paranoia, escalating confusion, a tiny cast, one location. That's the thing a good engine should be able to match.
A tool that understands Coherence understands that "similar" is about texture, not genre tags. A tool that doesn't will hand you Interstellar and wish you luck.
How I Judged Them
Four criteria, scored 1–5 each:
- Taste accuracy — did the picks feel like they belonged in the same emotional zip code?
- Diversity — how many of the recommendations were basically the same movie in a different hat?
- Streaming availability — could I actually watch any of this tonight without renting four things?
- Blockbuster resistance — did it default to the biggest, most obvious title in the genre?
What Each Tool Actually Did
TasteDive — the scattershot overachiever
TasteDive came back with Primer, The One I Love, and Time Lapse, which is a genuinely good read on the seed. It also, because it indexes everything, tossed in a podcast episode and a video game. Charming, but not helpful when you're trying to pick a movie in the next four minutes. No streaming data at all — I had to check availability myself, which killed the momentum.
Accuracy 4 · Diversity 4 · Availability 1 · Blockbuster resistance 4
Taste.io — great taste, annoying onboarding
This one makes you rate a stack of titles before it'll say anything useful. Twenty ratings later, it produced Primer, Triangle, and The Endless — legitimately sharp. It also slipped in Interstellar near the bottom, which felt like a nervous tic rather than a real suggestion. If you're willing to do the homework, the payoff is there.
Accuracy 4 · Diversity 3 · Availability 2 · Blockbuster resistance 3
Letterboxd's "Similar Films" — the crowd knows
Letterboxd leans on what real people actually watched together, and it shows: +1, The One I Love, Safety Not Guaranteed, The Infinite Man. Weird, specific, and almost none of them were things I'd already seen. Best discovery engine of the bunch by a mile. The catch is that it doesn't tell you where any of it is streaming, so it's a research tool, not a decision tool.
Accuracy 5 · Diversity 5 · Availability 1 · Blockbuster resistance 5
IMDb's "More Like This" — the safe answer
Interstellar. Arrival. Moon. Predestination. All fine movies. All movies I could have named myself in six seconds without a computer. IMDb is doing popularity math, not taste math, and it shows. Zero risk, zero surprise.
Accuracy 3 · Diversity 2 · Availability 2 · Blockbuster resistance 1
JustWatch — asked a different question entirely
JustWatch is built around what's available on your subscriptions, which is a genuinely useful filter and a terrible similarity engine. Because it prioritizes "on your services right now," the genre boundaries got mushy fast — at one point it offered me a cooking show. Not a movie. A cooking show. The logic, as far as I can reconstruct it, was roughly "you like watching things, here is a thing."
Accuracy 1 · Diversity 3 · Availability 5 · Blockbuster resistance 2
Criticker — the quiet winner
Criticker predicts what you specifically will rate something, based on users whose taste matches yours. For Coherence it surfaced Primer, Upstream Color, The One I Love, and — delightfully — My Dinner with Andre, which is the kind of left-field pull that makes you sit up. It understood that the movie is about conversation and dread, not spaceships.
Accuracy 5 · Diversity 4 · Availability 2 · Blockbuster resistance 5
The Scoreboard
| Tool | Accuracy | Diversity | Availability | Blockbuster Resistance | Total |
|---|---|---|---|---|---|
| Criticker | 5 | 4 | 2 | 5 | 16 |
| Letterboxd | 5 | 5 | 1 | 5 | 16 |
| Taste.io | 4 | 3 | 2 | 3 | 12 |
| TasteDive | 4 | 4 | 1 | 4 | 13 |
| IMDb | 3 | 2 | 2 | 1 | 8 |
| JustWatch | 1 | 3 | 5 | 2 | 11 |
So What Do You Actually Do With This?
Here's the workflow that came out of this mess, and it's the only part of the article I'd tattoo on my forearm:
Use two tools, not one. Pair a crowd-sourced engine (Letterboxd) with a prediction engine (Criticker). One finds things you'd never hear about; the other tells you whether you specifically will like them. Neither one is enough alone.
Then check availability somewhere else. JustWatch is genuinely excellent at the one job it's good at — tell it what you subscribe to and it'll tell you what's free tonight. It just shouldn't be your taste consultant.
Ignore the first result on principle. On four of the six tools, the top pick was the most famous movie in the general vicinity of the genre. That's not a recommendation, that's a shrug with a poster attached.
The Takeaway
Nobody won outright, and that's sort of the finding. Letterboxd and Criticker tied on points but for opposite reasons — one is a room full of strangers with good taste, the other is a math problem that knows your embarrassing four-star ratings. If you want the single best pick tonight, start with Criticker. If you want the most interesting list to work through over a month, start with Letterboxd.
And if your tool's answer to Coherence is Interstellar, congratulations: you've built a popularity contest, not a recommendation engine. Try again with a movie nobody's heard of and see how it does.
