What it does · what it doesn't · where its limits sit
Triangula reads one of sixteen archetypes from two inputs: a photograph of your face and twelve answers you give about how you move through a day. The two are scored independently and then reconciled. This page is a plain-language description of what happens between those inputs and the label you see at the end.
The geometry side
The face-side is a measurement, not an interpretation. The current implementation uses face-api.js: Tiny Face Detector locates the face, and the faceLandmark68Net model estimates 68 landmark positions around the jaw, eyebrows, eyes, nose and mouth. The landmark count describes the model output; it is not evidence of personality or compatibility accuracy. Photo processing runs on your device: neither the image file nor its base64 representation is sent to our server or AI text providers. The browser sends derived face measurements and your twelve answers to our server for scoring. Raw landmark coordinates are not included in these requests. To generate the written reading, the server may send your archetype, dimension scores and face-shape, nose-shape and mouth-width categories to DeepInfra or DeepSeek. If you request an email, your email address and the generated report are sent to Resend for delivery. See our Privacy Policy for data storage details. From the landmark coordinates the engine computes three structural axes:
Face shape and proportion. How long, wide, and angular the face is along three independent dimensions.
Nose orientation. Whether the nose reads as wide-set and grounded, narrow and rising, or somewhere between.
Mouth-to-face ratio. How the mouth's width and lift sit against the face as a whole.
These three axes are reduced to a three-dimensional vector. The vector is the geometry contribution to your reading. It is deterministic: the same photograph produces the same vector every time.
The answer side
The twelve questions are not personality items in the psychometric sense. They are short, situational, and chosen to load cleanly onto one of three orientations — drive (how you push toward outcomes), bond (how you orient to people), and structure (how you handle ambiguity). Each answer nudges a position along those three axes. The result is a second three-dimensional vector, this one behavioural.
Where the two meet
The two vectors are compared. If they agree, the engine returns a confident archetype with a short list of natural allies and natural friction. If they diverge, the engine flags the disagreement and returns the closest archetype with the disagreement made visible. The reading never averages a mismatch into a confident label.
The same photograph paired with different answer sets can move the label across adjacent archetypes. That movement is the feature, not the bug: a face that reads as one thing and a life that reads as another is not a contradiction to be averaged away, it is information.
Triangula does not diagnose, predict, or recommend. It names a pattern. What you do with the name is yours.
What the method is honest about
Triangula is built on twelve years of paired observation of faces and behaviour, systematised into a finite library of sixteen archetypes and a strict scoring formula. It is reproducible on the geometry side. It is interpretive on the archetype side — closer to a well-built horoscope than to a clinical instrument. It does not claim diagnostic validity, and it does not present itself as peer-reviewed psychology.
The old claim that facial features directly determine character — that a wide jaw means dominance, that a small forehead means weakness — is not what the method does and not what it claims. Triangula measures structural geometry and pairs it with what you say about yourself; the answers carry as much weight as the face.
Frequently asked questions
Is this based on science?
Triangula is a pattern-based classification, not a peer-reviewed psychological instrument. The geometry side is reproducible; the archetype side is interpretive. It is closer to a well-built horoscope than to a clinical test — honest about its limits, useful as a mirror, not a verdict.
What does it predict?
Nothing specific about your future. The printable report offers an archetype profile and dimension scores, strengths and blind spots, career suggestions, selected archetype comparisons for business, romantic and family contexts, and a brief reflection on personal growth. These are interpretive prompts, not predictions about you or your relationships.
What does it not predict?
It does not predict intelligence, moral character, mental health, success in any specific field, romantic compatibility with a named individual, or truthfulness. Reading one face and one set of answers cannot tell you whether to hire someone, marry them, or trust them. Use it for self-reflection, not for decisions about other people.
Isn't physiognomy debunked?
The 19th-century claim that facial features directly determine character is not supported by modern evidence, and Triangula does not rely on it. Triangula measures three structural axes and pairs them with your self-reported answers. The face side produces a vector, not a verdict; the answer side does most of the interpretive work.
How is this different from MBTI, the Big Five, or horoscopes?
MBTI and the Big Five are self-report questionnaires only; they ignore your face entirely. Horoscopes project a fixed narrative onto a date. Triangula takes a photograph and twelve answers from you and produces one of sixteen labels. It sits between the two: more grounded than a horoscope, less clinical than MBTI, and intentionally narrower than a personality survey.
Will two readings of the same photo give the same result?
Yes, for the geometry vector — the measurements are deterministic. The final archetype label blends the geometry vector with your twelve answers, so different answer sets can move the label across adjacent archetypes. Where geometry and behaviour diverge, the reading stays honest about it instead of forcing one or the other.
Does the reading change as I get older?
Facial proportions and estimated landmark coordinates are not assumed to remain unchanged with age. The model uses a shared landmark layout, but that does not establish equal accuracy across age groups. Your reading combines measurements from the photograph with your answers; it is an interpretation, not a fixed diagnosis.
Who built it?
The method was built by Sergius G., engineer-researcher in non-verbal archetypes. It is maintained with a small team of non-verbal archetype researchers.