The adjacent category and its lessons

Face-rating tools taught this category most of its defaults: the ten-point scale, the upward generosity drift and the share-card presentation all arrived here already settled. They ran the same kind of model at far larger volume for years longer, and left calibration disclosure as the main question they never answered.

The scale question, settled early

Ten points, not five, not a hundred - face raters converged on this early and stayed there, and the reasons a ten-point scale sticks once a category adopts it apply identically here. By the time this category needed to pick a scale, the choice had already been made for it by precedent rather than by anyone weighing the options fresh. Precedent had user preference behind it: when 149 respondents rated a recently visited store or restaurant on scales of 2 to 11 points, Preston and Colman (2000, Acta Psychologica) found preferences "highest for the 10-point scale". Research moved differently: the SCUT-FBP5500 benchmark from Liang and colleagues (2018) labels 5,500 faces with beauty scores "within [1, 5]", a sign the ten-point habit is a product convention more than a research one.

The generosity question, settled the same way

Face raters learned early that a tool which scores low loses users and a tool which scores high keeps them, and the centre of their output scales drifted upward accordingly over time. The same commercial pressure produces the same drift in this category, for the same reason: a user who feels judged harshly does not come back, regardless of what property is being judged. Nobody had to discover this twice.

The calibration question, still unsettled

Where face raters did not leave a clean answer is calibration disclosure. The larger category never converged on publishing how an internal value maps onto the visible number, and that gap carried over intact - most tools in this category do not document it either, for the same lack of external pressure to do so. Whether a given tool publishes anything about its own calibration at all is still checked tool by tool, because the adjacent category left no norm to inherit here.

Presentation, inherited wholesale

The share card, the reveal delay before a number appears, the word label sitting next to the digit - all of this is presentation vocabulary the face-rating category built and refined first, at a scale that let it iterate on what got shared and what got ignored. Tools in this category adopted the same elements largely unchanged, which is one reason a field guide to current presentation conventions reads like it could describe either category with only the subject swapped.

Cost model, inherited more loosely

One area where the adjacent category left a weaker precedent is how a tool charges. Face raters experimented across free-with-ads, per-result credits and subscriptions for years without any one model winning decisively, and that lack of a settled answer carried over as genuine variety rather than as convention - this category shows the same spread of cost models, and none of them reads as a copy of what came before in the way the scale or the generosity curve does. What did travel across is the more specific lesson about failure handling: a generation that errors and still charges damages trust regardless of category, and tools in both spaces converged separately on refunding failed results, more because the alternative is an obvious complaint magnet than because either category studied the other's approach to it.

Auditability as the newer lesson

The most recent thing this category has plausibly taken from the older one is less a settled convention and more a live argument still being had in both places: whether showing other users' results, breakdowns included, makes a tool more trustworthy or just more exposed. Face raters with public boards let users compare their own breakdown against a visible spread of others, which is the same feature this category's more transparent tools have started to offer. Whether it becomes as standard here as the scale and the generosity curve already are is not yet settled, which is one of the few open questions the adjacent category has not already closed on this category's behalf.

What did not carry over

The underlying computation is not identical between the two categories, even where the presentation is. What a given model is actually doing with an image differs by what it was trained to look at, and a face-rating model's judgement is not simply repurposed wholesale for a different subject. Nor did the adjacent category's tools ever touch physical measurement, which sits outside both categories' scope, or human review, which was never automated at scale in either one.

The lesson worth taking from the comparison is not that one category is more advanced than the other. It is that most of what looks like a considered design decision in a rating tool - the scale, the generosity, the share card - was decided once, in the older adjacent category, and simply arrived here already settled. Rate Cock reports six axes on the ten-point scale this lineage settled on, which is the one part of its design that departs from the adjacent category's more common single-total default.