—

Every measurement rendered where the engine placed it. Four layers over either capture: Geometry — the construction chords, rays and midlines of the registry metrics; Contour — Angularity's five region hulls, tinted by where their slots sit against the reference centre; Features — the exact pixel regions and marks the Misc-Features pillar read (skin ROIs, sclera rings, silhouette rows, the hairline chord); Dimorphism — the real landmark sets each trained sex-axis is fitted on, tinted by within-sex percentile. Hover a row to isolate; click to pin.

The one number in this readout that is a rating. Five terms, explicit weights, no renormalisation when a term is withheld (it contributes zero and is named). Front-/Side-Harmony and Features are scored against their currently-buildable budget, not their nominal 100 — dividing by a ceiling nobody can reach would cap every face regardless of its geometry. Weights are evidence-informed priors from a literature pass, not a fitted model.

Four weighted decision matrices on one engine. Each row is a component: the bar fills to the points earned, the tick marks the component's budget, and the label counts how many of its inputs actually scored. Dormant components stay visible so coverage is never overstated.

Contour typicality, not "how sharp": every slot is read against its own p10–p90 reference band and scored by distance from the band centre in either direction. Zero is the centre; ±1 is the band edge; beyond ±1 is outside the reference interval. Sex-specific bands are used for the eight slots with a real measured sex effect. Four contour constructs (zygomatic-mandibular angle, ogee curve, buccal thickness, periorbital hollowing) cannot be measured from 2D landmarks at all and are excluded rather than estimated.

Descriptive only, never scored on its own. One front global axis and seven regional axes, each an independent Procrustes → PCA → LDA fit on its own landmark set (the hulls in the overlay's Dimorphism layer, Frontal view). The dot is where this face sits among the reference males on that axis; the shaded middle is the 25th–75th percentile. Regions do not sum to the global read — a wide spread across them is the expected pattern in real sex-difference studies, not a contradiction. A separate profile axis (◆ diamond, dashed divider below) reads the side capture's own WFLW landmark shape — a different capture, a different Procrustes fit, never combined with the front numbers above it, and never used to score Side-Harmony (see its own card below for why).

For every scored input, how much of its reference interval this face uses. 100% sits exactly at the edge; below is inside, above is beyond. One-sided bands (asymmetries, feature flaws) measure the distance from the ideal end; two-sided bands measure the distance from the centre. Sorted by distance.