How THDB rates horror movies

Five magnitude axes run 0–5, higher meaning more of the thing:

Psychological (0–5)

How much the fear works on your mind — dread, paranoia, a sense of wrongness. 0 means it never gets under your skin; 5 means the film is an anxiety machine.

Visceral (0–5)

How much the fear attacks the body — physical threat, shock, assault on the senses. Scored independently of psychological: a film can be high on both.

Gore (0–5)

On-screen blood and viscera, nothing else. 0 is bloodless; 5 is wall-to-wall. Gore is deliberately separate from how scary a film is.

Lingering (0–5)

How long the dread hangs around after the credits. 0 fades immediately; 5 follows you around for days.

Intensity (0–5)

The blunt total: how hard the experience hits, all sources combined. 0 is gentle by horror standards; 5 is one of the most punishing watches in the catalog.

Three spectrum axes run 0–100 between two poles — they describe a position, not an amount:

Slow burn ↔ Jump scares (0–100)

0 is pure creeping dread that never startles you; 100 is a jump-scare rollercoaster. The middle means the film balances both.

Supernatural ↔ Real-world (0–100)

0 is fully otherworldly — ghosts, demons, curses; 100 is fear that could actually happen. This is about the source of the fear, not the plot.

Fun-scary ↔ Disturbing (0–100)

0 is horror as a good time — scares that leave no bruise; 100 is genuinely upsetting, the kind that troubles rather than thrills.

Profiles are generated by a large language model against a calibrated rubric with fixed anchor films per axis, in one-time batch runs — never at page view. The model and prompt-rubric version that produced each profile are recorded on the row, and every profile carries the model’s own confidence score: obscure films it likely guessed at are flagged low-confidence right on the page.

Regeneration is versioned and conservative: community-corrected values are pinned and can never be overwritten by a newer generation run.

The generator is contractually forbidden from describing plot events, endings, twists, deaths, or character fates. Profiles describe texture — pacing, register, aftertaste — so you can read everything on a movie page before watching. The intensity-over-runtime curve goes further: it is hidden behind an explicit reveal, because even a pacing shape is a spoiler to some viewers.

Every profile starts as AI-generated, unconfirmed — and is labeled that way on the page. Members who have seen the movie grade whether the profile fits; at a net +3 the profile flips to community-reviewed. Members can also propose per-axis corrections: when 5 distinct members agree within a tolerance band, the corrected value applies automatically and the change is logged publicly on the profile’s revision history.

Anti-abuse: consensus contributions only count from accounts at least 7days old, so a batch of fresh accounts can’t flip statuses the day it’s created.

Movie metadata (titles, posters, release years, credits) is synced from TMDB and never hand-entered. Streaming availability is provided by JustWatch via TMDB. Scare profiles, intensity curves, and texture summaries are THDB’s own data and exist nowhere else.

Scare perception is personal — a 3/5 for one viewer is a 5/5 for another, and the axes measure the film, not you. AI-generated profiles can be wrong, especially for obscure films with thin source material; that’s exactly what the confidence flag and the community ladder exist for. Treat unconfirmed profiles as a good first estimate, not a verdict.

Why THDB exists →