It started with a tweet about physiognomy — the claim that the Greeks could tell whether you’d look good bald from a few archetypes of how a face is built. The idea was too good to leave as a tweet. One session later — first commit ten to five in the afternoon, live before one in the morning — it was software.

Upload a portrait and the Oracle consults. A vision model reads the structure of the face — jaw definition, cheekbone prominence, the Vitruvian thirds, hairline stage — and maps it to one of the four classical temperaments the Greeks actually used: Sanguine, Choleric, Melancholic, Phlegmatic. Each verdict arrives as an editorial report: your archetype and its history, your statue twin from the classical record, a structural reading tied to visible cues, and the question the tweet asked in the first place — the Bald Verdict, grounded in real perception research rather than vibes. Every piece of jargon carries its own inline explainer; nothing is left as mystique.

The engineering under the toga is serious. The read runs as an adaptive self-consistency vote — one pass when the face is decisive, fanning out to three or five when it isn’t, failing closed on anything that shouldn’t be analysed. Structured output, schema-validated, with confidence carried per observation. Your photo is downscaled and stripped in the browser, analysed in the moment, and never stored — no database, nothing to leak.

And the honesty is architectural, not a footnote. Physiognomy is historically real and scientifically discredited, and the product says so on its front page. Every historical claim in the content is labelled at the data level — attested in the sources, defensible synthesis, or the product’s own licence — and Kallos never reads character from a face, never ranks people, never touches ethnicity. It promises what the Greeks would have said, not what they could have measured. Erudite entertainment, played completely straight.

→ kallos