One number can’t describe a scare

Every review site collapses quality into one number. That’s mostly fine — a restaurant is good or it isn’t. But horror breaks the model, and Horrorometer exists because of how it breaks.

The problem in one comparison: a slow-burn atmospheric ghost story and a wall-to-wall splatter film can both score 3.5 stars. They have almost nothing in common — different pleasures, different audiences, different reasons you’d recommend them. The star rating says "equally good" when what the user needs to know is "completely different experiences". A horror fan who hates gore but loves dread gets actively misled by the flattened score.

Three axes, one personality

Horrorometer rates every title on three axes:

• Creepy — dread, atmosphere, the thing you feel before anything happens.
• Gory — the explicit stuff, what’s shown rather than implied.
• Jumpy — the physical shock, the cat-in-the-fridge school of filmmaking.

The axes are independent, deliberately. High Creepy / low Jumpy is — arguably — the genre’s high art. High Jumpy / low Creepy is a theme park ride, and a perfectly good one. The profile is the recommendation: a user who learns "my sweet spot is Creepy 8, Gory 2, Jumpy 5" has a search query no star system can express.

What the design week actually involved

The rating system took a week of design, and most of it was spent saying no:

• No composite score. The tempting middle ground — average the three into one headline number — reintroduces the exact flattening the axes exist to avoid. The profile is the answer; there is no summary.
• Three axes, not ten. — I tested lists like "slow burn, found footage, folk horror, slasher…" — each axis added makes rating more accurate and less likely to actually happen. Three is the number where users reliably finish the job. Coverage beats precision.
• No half-stars. A 1–5 range per axis keeps the mental load down. Users aren’t calibrating instruments; they’re expressing impressions.
• Community over critic. Ratings are averages of user profiles, with moderation — the taste graph is the product, not my opinion.

The taxonomy is the moat

Anyone can clone a review site. Cloning a taste model requires users to have already populated it — every rating makes the next recommendation better, and none of it is portable. The lesson I keep from this applies well beyond horror: the structure of your ratings encodes your understanding of your domain. A star system says "I understand quality". A three-axis profile says "I understand this genre’s quality" — and the second one earns loyalty the first can’t.

It also changed how I watch horror, which I did not expect from a data model. Once you split dread from shock, you notice how many “scary” films are only loud.