How we score

Star ratings compress everything worth knowing into one number. VHSCREAM scores every work on six continuous axes instead, against fixed anchor works so the numbers still mean the same thing in six months. Scores are drafted with machine assistance and confirmed by a human against this rubric before they count as settled. This page is the whole method — including how many have actually been confirmed so far, which is currently none.

The six axes

Each axis is anchored by three works pinned at a low, middle and high value. Scoring is placing a work between those anchors, not inventing a number.

What we do not claim

Of 105 works in the index, 0 have had their vector explicitly confirmed by a human reviewer against this rubric. The rest are drafts, and every page carrying one says so. We would rather publish the number than round it up.

That date is deliberately here instead of a promised schedule. A cadence we announce is a promise whose breach is silent; a date that stops moving is one you can see from the outside, which is the only kind of commitment this page has any business making.

You can move that number. Every work page now carries a control on each of the six axes: mark the score right, too high, or too low. It takes one click, needs no account, and it is the only way a second human gets into this process. A vote never changes a published score — scores change when a person re-scores the work against the rubric above. Votes decide which work that person picks up next, which is a job the two of us cannot do alone across 105 titles.

Scores drift, so the policy is that a fixed control set of twenty titles gets re-scored every quarter without reference to the previous numbers; drift over ten points on any axis means that axis is ambiguously defined and gets rewritten.

What the era axis turned out to be

The rubric above says era bias measures texture and craft sensibility rather than release date. Checked against the index, it does not behave that way. Correlated against release year the axis sits at r = +0.966 across the 61 films we have scored, +0.996 for television and +0.972 for records — close enough to a straight line that in those three media the score is very nearly the release date wearing a different name. Only books come apart from it, at +0.729, which is the one place the axis is measuring something the calendar does not already tell you.

Two explanations fit that and we cannot yet separate them. Either the catalogue holds almost nothing whose texture diverges from its period — only six of the 61 films sit more than fifteen points from where their year predicts — or the axis has been scored from the date rather than from the grain, which would make it the weakest of the six and the one doing the least independent work. We do not know which, and saying so is cheaper than defending a claim we have not tested.

What would settle it. A deliberate run of works whose texture fights their date: contemporary films built to look recovered, and older films shot clean. If the correlation survives twenty of those, the axis is a date proxy and should be renamed or retired rather than quietly kept. The films furthest off the line today are the ones to score against: The Blair Witch Project (1999, scored 030 where the line says 049), Scream (1996, scored 026 where the line says 044), Final Destination (2000, scored 068 where the line says 051).