How Sigil works

Sigil is the engine behind every score in Mavqen. It runs in four stages, and it is worth knowing which is which, because they fail in different ways and only one of them is something you can fix.

A detection model places 77 anatomical points — 49 on the front photo, 28 on the side. Pupil centres, the corners of each eye, the widest point of the cheekbone, the gonion where the jaw turns, the tragus in front of the ear canal.

Then you correct them. Every point, with its name, a definition diagram and a reference photograph showing where it belongs. This is not a formality: a gonion placed 4% too low moves four separate measurements at once, and a point that drifts on one side only reads as asymmetry you do not have.

The model is a starting point, not a finished mapping — and correcting it never costs you a scan.

From those points, pure geometry. Distances, ratios and angles — 44 of them in the Harmony pillar alone, built on a system of reference lines rather than measured ad hoc.

On the side view nearly everything is measured against the Frankfort plane, the horizontal from the ear canal to the lowest point of the eye socket. It is the standard the clinical literature uses, and it is why the tragus has to be visible in your side photo.

Nothing at this stage is a judgement call. Give it the same points twice and you get the same numbers twice.

The five pillars combine into your overall score out of 10, weighted — and the weights differ between men and women, because the same feature does not carry the same weight on both. Harmony dominates the female model; the male model spreads more evenly across structure and definition. The exact numbers live in Sigil's own configuration rather than in this page, so that what you read here can never drift from what actually ran.

Two things worth knowing about how it behaves:

A pillar we could not read properly does not get to vote. If too little of a pillar was scored — a missing side photo, a set of landmarks that never got corrected — it is reported with its coverage shown and left out of the total rather than being quietly counted as a bad result.

Appeal is a composite, not a single reading. It is roughly three-fifths the colouring-and-appeal pool, three-tenths Physiognomy, and a tenth a deviation term. That structure is why the Appeal Report is its own surface with three chapters in it.

5 is the population average. The scale is deliberately non-linear: moving from 5 to 6 is common, moving from 7 to 8 is rare, and the top of the scale is essentially theoretical — Sigil does not hand out 9s. Your overall score is shown with its label and its rarity; each pillar is shown as a percentile off the same curve. Either way you can read it without memorising it.

Rarity is symmetrical: a 3 is as uncommon as a 7. The percentiles are the same curve read the other way round — a 6.5 is roughly the 95th percentile, a 7.0 the 99th, and an 8.0 leaves about one in twenty thousand above it.

The geometry is objective. The ideal ranges those measurements are compared against, and the way the pillars are weighted together, encode aesthetic research and a set of decisions we made. They are not a law of nature, and we are not going to pretend otherwise.

What that makes a Sigil score genuinely good for: a structured, repeatable reading of your proportions, stable enough to compare against itself two months later. What it is not: a ranking of you against other people. There are no leaderboards here and there never will be.