HOW WE SCORE
Full disclosure of how MOGGED's facial-analysis tools work — the underlying AI model, the peer-reviewed research informing each scoring component, the accuracy ranges, the known demographic limitations, and what these tools are explicitly not for.
Questions or corrections: support@metellusproductions.com.
- →Approach: structured AI prompts per tool, calibrated against published anatomical norms.
- →Scoring approach: 5-component composite scoring per tool (where applicable), grounded in peer-reviewed attractiveness research and orthodontic anatomical literature.
- →Accuracy: ±1-2mm for soft-tissue distances, ±3-5° for angles. Photo-based estimation, NOT equivalent to a cephalometric radiograph.
- →Photo handling: Photos are processed in real time, not stored, not used to train models, not shared. Discarded after the response is returned.
- →Not medical advice. Not a surgical consultation. Not for use on minors. Not a clinical diagnostic instrument.
Scientific Foundation
The peer-reviewed research that informs how we score. Each citation links to the original paper.
11 meta-analyses across thousands of raters: cross-cultural agreement on attractiveness is r ≈ 0.65; within-culture agreement is r ≈ 0.90. Establishes that attractiveness ratings are not purely subjective.
Averageness, symmetry, and sexual dimorphism are independently attractive across cultures. Maps directly onto the components our tools score.
Of 21 measured facial proportions, 12 correlated significantly with attractiveness ratings. Four features (eye height, nose area, cheek width, smile width) explained over 50% of variance. Justifies multi-component scoring.
Sexual dimorphism (masculine features in male faces, feminine in female) independently predicts attractiveness ratings. Supports our tool's tier framing of masculine traits.
Survey of aesthetic preferences. Ideal male gonial angle: 130°. Source of the specific numeric benchmark in our Jawline Test.
Foundational definition of the E-line (esthetic plane) — line from nose tip to soft-tissue chin. Standard reference for chin projection assessment in our Recessed Chin analysis.
CNN attractiveness scoring achieves Pearson r ≈ 0.96 with human raters under controlled conditions. Justifies our photo-capture guidance (neutral expression, controlled lighting). Photo expression and lighting significantly affect AI scores.
How Each Tool Works
Specific scoring approach, output format, and research grounding for every tool we ship.
Mogger Test
Run the toolSingle-output PSL tier classification (Chad / Chadlite / Normie / LTN). The model evaluates the face holistically against community-defined PSL tiers and returns a tier + 2-4 markers explaining its read.
Tier framing follows the PSL scale (PUAHate / Sluthate / Lookism), a community convention emerged from looksmax forums in the early 2010s. Cross-cultural consensus on attractiveness (Langlois 2000) supports the validity of subjective tier classification.
Canthal Tilt Test
Run the toolEstimates the angle (in degrees) between the medial canthus (inner eye corner) and lateral canthus (outer eye corner) relative to horizontal. Returns a degree value + classification (positive / neutral / negative).
Reference range: ideal positive male tilt ~+4° to +6° per orthodontic norms. Classification thresholds: positive ≥ +3°, neutral -1 to +2°, negative ≤ -2°.
Hunter Eyes Test
Run the tool5-component scoring on a 0-10 scale per component, summed × 2 to produce a 0-100 composite score. Components: canthal tilt, upper-lid exposure, orbital socket depth, almond-ratio shape, brow-to-eye distance. Verdict tiers: Hunter (80+) / Hunter-Leaning (65-79) / Mixed (45-64) / Prey-Leaning (30-44) / Prey (≤29).
5-component framework adapted from looksmax community conventions (looksmax.org, huntereyes.net). Multi-feature aggregation outperforms single-score ratings (Cunningham 1986; Obwegeser 2024). The component selection corresponds to the structural traits the looksmaxxing community has codified as defining 'hunter eyes.'
Jawline Test
Run the tool5-component scoring on a 0-10 scale per component, summed × 2 to produce a 0-100 composite score. Components: gonial angle, jawline sharpness, chin projection, mandibular width, jaw symmetry. Verdict tiers: Chiseled (80+) / Defined (65-79) / Average (45-64) / Soft (30-44) / Recessed (≤29).
Gonial angle ideal of ~120-130° per Mommaerts 2016. Chin projection assessed via Ricketts E-line (1957). Symmetry contribution validated by Perrett 1999 ("Symmetry and human facial attractiveness"). Mandibular-to-bizygomatic width ratio target ~92-94% (cosmetic surgery convention).
PSL Score
Run the toolReturns a 1-10 PSL rating with one of 7 tier labels. The model maps holistic facial attractiveness to the PSL scale, calibrated so 5 = population average.
PSL scale origins documented at Wikipedia / KnowYourMeme. The community convention is to interpret 1-10 as a normal distribution where 5 is the mean. Cross-cultural consensus on attractiveness (Langlois 2000) supports the validity of holistic ratings; cf. Rhodes 2006 for the underlying psychological mechanisms.
Our Analysis Approach
All photo-based analysis is performed by a computer-vision model with structured prompts calibrated against published anatomical norms. Each tool sends a prompt that:
- →Specifies the exact components to score and the 0-10 scale per component
- →Provides anatomical reference values from published literature (e.g. ideal gonial angle 120-130° per Mommaerts 2016)
- →Constrains output to a strict structured format for consistency across runs
- →Includes calibration notes (e.g. ethnic variation in chin-projection norms, photo-angle sensitivity)
We do not train custom models on user data. We do not fine-tune on user photos. Our approach combines a calibrated prompt structure with anatomical norms from peer-reviewed research — that calibration is the proprietary part of our methodology.
Known Limitations & Biases
Honest disclosure of where these tools are weakest. Read this section before interpreting your results.
Photo angle, expression, and lighting
Even 10-15° of off-axis head rotation introduces measurable landmark error. Open mouth, smiling, or squinting alters lid exposure and canthal tilt readings. Backlighting or shadows alter perceived socket depth and contour sharpness. We strongly recommend front-facing, neutral-expression photos with even lighting.
Demographic bias in underlying model
The Buolamwini & Gebru 'Gender Shades' study (2018) and NIST FRVT demographic study (2019) document well-known biases in commercial face-vision systems — particularly for darker Fitzpatrick skin types, non-frontal poses, and certain age ranges. Our underlying computer-vision model inherits some of these biases. Scores may be less accurate for darker skin tones, faces outside Caucasian / East Asian / South Asian phenotypes most represented in academic benchmarks, and non-standard poses.
Reference-norm provenance (Caucasian-derived)
The Ricketts E-line (1957), Mommaerts gonial angle (2016), and several other anatomical norms cited above were originally derived from Caucasian samples. Documented ethnic variation exists — African, East Asian, and South Asian populations have naturally more protrusive lip and chin profiles relative to the Ricketts standard, for example. Adjust expectations based on ethnicity; the line is directionally useful, not universally applicable.
Photo measurement vs. cephalometric radiograph
Photo-based facial assessment is NOT clinically equivalent to a lateral cephalometric X-ray. 2D photogrammetry studies (Frontiers in Public Health, 2021) report ICC > 0.80 for most soft-tissue dimensions but ±1-2 mm for linear measurements and ±3-5° for angular features (canthal tilt, gonial angle). True bony landmarks require radiograph-based assessment by an orthodontist or oral surgeon.
Score variance between photos
Research (Obwegeser et al., 2024) confirms CNN attractiveness scoring varies meaningfully with photo conditions. Two photos of the same person under different lighting or angle can produce scores that differ by 5-10 points. The verdict tier (Chiseled / Defined / Average / Soft / Recessed, etc.) is much more stable than the exact 0-100 number. For most reliable results, take 2-3 photos at slightly different angles and average them.
Bias literature we anchor to
- Buolamwini & Gebru (2018) — Gender Shades — documented up to 34.7% error rate for darker-skinned female faces in commercial gender-classification systems.
- NIST FRVT Part 3 (2019) — Demographic Effects — false-positive rates 10-100x higher for Asian and African American faces vs. Caucasian across most tested algorithms.
- IBM Diversity in Faces (2019) — quantified over-representation of light-skinned subjects in public training datasets.
Photo Handling & Privacy
Photos are not stored
When you upload or capture a photo, it is processed for the duration of one analysis request only. Once the response is returned to your browser, the image data is discarded. We do not log, store, or retain your photo in any form.
No model training on user data
We do not use uploaded photos to train, fine-tune, or evaluate any model. Our analysis runs against calibrated scoring prompts grounded in published research — your photo is only ever an input, never training data.
Shareable results — what's shared
Our shareable result links contain only your numeric scores and verdict — encoded in the URL itself. Photos are never included in shared results. Anyone with the link sees your score and verdict, nothing else.
What These Tools Are NOT
Five explicit non-claims. If your situation matches any of these, do not use these tools as a substitute for the real thing.
Medical diagnosis or treatment
These tools do not diagnose or treat any medical condition. They are not a substitute for evaluation by an orthodontist, oral surgeon, plastic surgeon, sleep specialist, or any other licensed clinician.
Cosmetic-surgery consultation
Specific procedures (genioplasty, BSSO, mandibular implants, chin implants, masseter Botox, Kybella, etc.) require evaluation in person by a board-certified plastic surgeon or maxillofacial surgeon with appropriate imaging. Do not make surgical decisions based on these tools.
Sleep apnea / airway assessment
Severe retrognathia, hyperdivergence, and other patterns can be associated with obstructive sleep apnea risk. If you have symptoms (heavy snoring, daytime sleepiness, witnessed apneas), consult a sleep specialist for a formal evaluation. Our tools cannot assess sleep-disordered breathing.
Identity verification / forensic use
Our tools are not designed for, validated for, or appropriate for identity verification, surveillance, hiring, dating-app verification, or any other identity-bearing purpose. They evaluate aesthetic-relevant features only.
Use on minors
Our tools are intended for users 18+. Facial growth continues into the late teens / early 20s. Scoring an underage face produces results that are not biologically stable and can be psychologically harmful. Do not use on minors.
Questions, corrections, or research collaboration
Email support@metellusproductions.com. We respond to methodology questions, citation requests from journalists, and corrections from researchers and clinicians. We're building this in public; if you spot something wrong, tell us.
Run A Free Tool
Now that you know how the scoring works, try one. All free, no signup, 60-second results.