Is facial analysis science or pseudoscience?

The honest answer is "some of both", and any tool that tells you otherwise is overselling. Facial metrics run from genuinely research-backed to outright folklore, and lumping them together as "science" or "pseudoscience" is how people get misled in both directions.

The part that holds up

A few findings replicate well. Symmetry has a real if modest link to rated attractiveness (Rhodes). Averageness — being close to the population mean — is one of the more robust effects in the literature. Some sexually dimorphic features track with perception. None of these are certainties, but they are a long way from guesswork.

The part that does not

The golden ratio, Marquardt's mask and strict neoclassical canons are weak predictors; attractive faces break them constantly (Zwahlen). Single "ideal numbers" presented to four decimal places imply a precision that the evidence simply does not have. When a tool scores you harshly against phi and frames it as objective truth, that is the moment it tips into pseudoscience.

Where we try to sit

Our headline score is not a canon average — it is a model trained on thousands of human ratings and then calibrated so your number reads as an approximate percentile within your sex. In testing it agrees with human raters around 88% of the time for women and 87% for men, and it holds up fairly across ethnicities on faces it never trained on. That is meaningful, and it is also bounded: it predicts how groups of people on average tend to rate proportions, from one neutral photo.

What no score can do

It cannot tell you whether a particular person will be drawn to you — attraction is personal and contextual. It cannot see expression, voice, humour, style, confidence or the way you move, all of which move the needle hard in real life. And it cannot weigh your worth. Used well, a facial analysis is a research-informed mirror: useful context if you are curious or weighing a cosmetic decision, and nothing to organise your self-image around.