Slow horror follows a rule. Fast horror has none.

We scored 105 horror works — films, television, novels and records — on six continuous axes of fear, by the same rubric, on the same scale. Then we spent two rounds trying to break our own finding, and broke it twice. What survived is narrower than what we started with, and much stranger: in the patient half of horror, how fast the fear arrives predicts how much you will be shown. In the fast half, that rule is gone.

What happened to the first two versions of this page

Version one looked like a law of the form. The index started as the canon: the films, novels and records that get written about. Scored that way, pace of fear and viscera correlated at r = +0.67 across the 86 works in the index at the time — fast horror shows you things, slow horror withholds — and the fast-but-restrained corner held 2 of them.

It was a law of our sample. The corner was empty because we had not scored a single jump-scare film. The Conjuring, Insidious, Sinister, Lights Out, Annabelle, The Nun — the Conjuring Universe is the only horror franchise ever to pass two billion dollars at the box office, and it is built precisely to frighten you fast while showing you almost nothing. We scored 10 of those under the same rubric. The corner filled and the correlation fell to +0.46.

Version two died to one word: Saw. Having found one blind spot we went looking for its mirror image and found it immediately. We had scored the multiplex that shows nothing and skipped the multiplex that shows everything: Saw, Final Destination, Evil Dead (2013), It, Terrifier 2 — another franchise past a billion dollars, and four more films that made a great deal of money pointing the camera exactly where the canon will not. So we scored those too, and the correlation went back up — it sits at r = +0.50 today, after every addition described below. Two rounds of work to arrive roughly where we started is the point at which the number stops being the finding.

The shape was never a line. It is two populations. Cut the index at the pace midpoint and the single correlation comes apart:

• The slow half — 71 works — r = +0.60 (95% CI +0.43 to +0.73). Here the original law holds and holds firmly: the more patient the work, the less it shows you.
• The fast half — 34 works — r = +0.05 (95% CI -0.29 to +0.38). The fast half still runs the length of the viscera scale, from 18 to 98, but the climb is gone.

The confidence intervals do not overlap; the difference between the two correlations is z = 3.00 — about one chance in 366 of turning up by luck. Restricted to film and television only, it is the same story: +0.65 across 40 slow works, +0.13 across 28 fast ones.

Be precise about what that does and does not show. 34 works cannot establish that the relationship in fast horror is exactly zero; the interval reaches +0.38, so a weak version of the rule is still live there. What the interval does do is stop below +0.43 — below where the slow half’s interval begins — so the two cannot be the same number. The finding is the difference, not the absence. Whatever fast horror is doing, it is demonstrably not doing what slow horror does, and the headline is shorthand for that.

The uncomfortable part: it was there from the beginning. Take the version-one index — before the jump-scare films, before the gore, before any of this — and split it the same way, and it gives +0.67 across 68 slow works against +0.04 across 18 fast ones. The two populations were visible in the first 86 works we ever scored. We spent two rounds arguing about a corner because we were looking at a corner. The lesson we are taking from that is about instruments rather than about horror: a single correlation across a whole index is a bad instrument, and we used one for two rounds without asking what it was averaging together.

The obvious objection, and why it doesn’t hold. Splitting on pace narrows pace, and a narrowed predictor mechanically deflates a correlation — so the fast half should look weaker whatever is true. Except that the deflation runs the wrong way here. Cut at the median instead of at 50 and the fast group is the wider of the two in pace (SD 12.8 vs 8.9) and the relationship is still gone: r = +0.02 against r = +0.61 for the slow group. A wider predictor producing a weaker relationship is not range restriction. Something else is going on.

Our guess, offered as a guess. Speed doesn’t make the choice; somebody made it upstream. A film committed to a fright every few minutes still has to build each one out of something, but how much that production is willing to show was settled before pace was — by a rating, a budget, and whatever audience the title already has. A patient film has room to modulate, so its restraint reads as a property of the film itself. A fast film spends whatever it was handed, quickly. That is what an r of +0.05 would look like from the inside.

The tidier version of that guess is wrong, and we checked. The tidy version says speed forces a binary — restraint at speed becomes the jump scare, explicitness at speed becomes the set piece — so fast horror should pile up at both ends of the viscera scale and hollow out in the middle. It doesn’t. 11 of 34 fast works sit in the middle band of viscera (40–60), against 21 of 71 slow ones: 32% against 30%, effectively the same share. Fast horror is not polarised into two camps. It is spread across the whole range and simply does not care how fast it is going. That is a weaker and less quotable claim than the one we wanted, and it is the one the numbers support.

The old works in that corner, after we went and checked. A previous draft of this paragraph said only Halloween (1978) and The Texas Chain Saw Massacre (1974) sat in the fast-but-restrained corner from before 2000. That was true of the index and false of horror: we had never scored Jaws, Poltergeist, The Omen or Psycho, which is precisely the mistake that killed the last two versions of this page. So we scored them. Jaws and Poltergeist landed in that corner; The Omen and Psycho did not. The corner now holds The Texas Chain Saw Massacre (1974), Jaws (1975), Halloween (1978), Poltergeist (1982) — four works, not two. Restraint at speed is not a modern invention, and 1975 is not a fluke.

Of the 20 fast works we have scored from this century, 8 take that restrained lane and 12 take the explicit one. Studio horror did not choose a new lane. It widened one that was already there.

Horror is still not having fun. Median tone across the index is 18 / 100, where 0 is unrelieved and 100 is gleeful. The genre’s reputation for camp rests on a handful of outliers. Pace and tone correlate at r = +0.50: the fast stuff is the funny stuff.

Where this index actually lives

A byproduct of wiring up availability, and the only figure on this page that is not about fear. Of the 68 films and television series here, 50 are on a subscription service somewhere, and the services are not evenly spread: Philo carries 18, AMC+ carries 12, Shudder carries 12, fuboTV carries 11. Collapsing the resale duplicates — “Shudder”, “Shudder Amazon Channel” and “Shudder Apple TV channel” are one service — this is disproportionately a Shudder and AMC+ catalogue, which says something about where the kind of horror we score has ended up.

As of the 2026-10-05 availability pass, US region. That date is doing real work: streaming rights move constantly, so this figure rots in a way the fear scores do not. It is computed in the same build, from the same snapshot, as the “where to find it” line on each of the 68 pages — so the total and the pages cannot disagree, and you can check it by tallying them yourself the day it is published.

Check this yourself

Every number on this page comes out of one table, and that table is published as CSV and JSON under CC BY 4.0 — 105 works, 630 scores, the rubric linked. The correlation above rests on roughly fifty works per group, which is fragile enough that a handful of additions could move it. We would rather you tested it than believed it.

Method, and what would falsify this

Every work is scored 0–100 on each axis against fixed anchor works, so numbers stay comparable across media and months. The rubric is published. The halves use a strict cut at 50 — above 50 is fast, 50 and below is slow. Every figure describing the index as it stands is computed from the table at the bottom, so anyone can reproduce or contradict it; the version-one figures are computed the same way from the archived catalogue, over the 86 real works in it, which is why the history is quoted against 86 and not 105.

One scorer, zero confirmed. All 105 vectors were drafted by a single scorer and none has yet been confirmed by a second human. That matters more here than it did for the last version of the finding, because there is an obvious way to fake this result: if the scorer simply had a firmer mental model for patient horror and was guessing more freely about the fast stuff, the extra noise would flatten the fast half’s relationship and look exactly like what we are claiming.

That one is testable without a second human, so we tested it. Careless scoring is not selective — it would flatten every relationship in the fast half, not just the one we are interested in. It doesn’t. Across the other fourteen pairs of axes, the fast half is the more strongly structured of the two: mean |r| of 0.20 against 0.14 for the slow half, and 10 of 14 pairs stronger in the fast half. Viscera itself is spread wider there, not narrower (24.8 against 22.9). One pair collapses in the fast half and it is the pair the finding is about. That is what structure looks like; noise does not aim.

What that test does not rule out is the narrower version: a scorer whose theory of patient horror links pace and restraint, manufacturing the slow half’s correlation rather than the fast half’s absence. Confirming the slow half’s vectors against the published rubric is what settles that, and it is the specific thing the confirmation pass has to settle.

The sample is still not random — it is the canon, the corners, and now the box office. The ledger, so that a reader adding up the numbers above arrives at 105: 95 entries at version one, minus the 9 merchandise listings we deleted on finding they described products that did not exist, plus 10 jump-scare films, plus 5 from the gory multiplex, plus the 4 older films in the paragraph above. Those 9 are excluded from every figure on this page, the version-one ones included — which is why the history above is quoted against 86 works and not 95 entries. These correlations describe this index, not all of horror, and the index has 34 fast works to the slow half’s 71: the weaker result rests on the smaller pile. We have already been wrong twice here by not looking where we hadn’t looked. If you can name a work that should sit somewhere on this map and doesn’t, that is the most useful thing you can send us — both of the last two gaps of that kind rewrote this page.