Omen
Platinum
- Joined
- Sep 5, 2023
- Posts
- 1,002
- Reputation
- 707
I inputted cluade sonnet 4.6(cuz fable 5 is banned) with every psl based rating system that I can find wikis for + give advice how to fruad + good peptides stacks for your face + situation (if you explain) and the rating is super accurate compared real ratings on the forum and compared to how the system should be l, gives also tips on what surgeries are needed to achieve certain psl’s.
The rater is still not complete at all made it in my free time at work will update and maybe post it here when Im done tweaking it to perfection.
————————————————————-
Rating Basis
The PSL (Physical Subjectivity Level) scale — 0 to 8
Not a vibe-based rating. Every number is anchored to named celebrities so the model always calibrates against known references rather than drifting subjectively. The rating measures three things simultaneously:
Facial Harmony — how well proportioned the face is geometrically (neoclassical canons, golden ratio, facial thirds/fifths)
Sexual Dimorphism — how strongly testosterone markers read on the face (fWHR, brow ridge, jaw angle, chin projection — backed by Carré & McCormick 2008, Gangestad & Thornhill research)
Feature Quality — individual feature assessment across all 14 markers in a fixed order so nothing gets missed
It also applies a BF% penalty in real time — if you’re above 30% it gives both a current rating AND a lean estimate, rather than rating the fat as your face.
What Was Implemented From My Suggestions
Structured hierarchy — the system prompt is divided into clear labeled sections (rating scale → assessment order → BF protocol → phenotype → contacts → fraud tiers → surgery → peptides → output format) instead of dumped wiki text. This forces consistent behavior.
Mandatory 14-marker assessment order — always assessed in the same sequence so no feature gets skipped or weighted arbitrarily.
BF% adjustment protocol — five specific ranges (under 15% / 15-20% / 20-25% / 25-30% / 30%+) each with a defined accuracy level and lean estimate instruction.
Phenotype classification first — identifies MENA/Mediterranean, Northwestern European, East Asian, Sub-Saharan before assessing anything else, then applies phenotype-specific criteria.
Contact lens guide by phenotype — specific brand recommendations (Solotica, Anesthesia, Desio) with exact color names for each skin type. Explicitly bans pure blue on olive skin.
Three-tier fraud system — 3.5→4.5 softmaxxing, 4.5→5.5 halo stacking, 5.5→6+ medmaxxing, each with specific interventions for that PSL range rather than generic advice.
Surgery table by deficit — each structural weakness mapped to a specific procedure with PSL gain estimate and risk level (Low/Med/High/Very High).
Peptide stacks by goal — four separate stacks (collagen/skin, jaw/connective tissue, hair preservation, GH optimization) with danger ratings and cycling protocols.
Consistency anchoring — instructed to compare against 3 celebrity PSL anchors before finalizing any rating to prevent drift.
Mandatory 12-point output format — every response must include phenotype, BF estimate, current PSL, lean PSL, tier classification, all 14 features, dimorphism score, harmony score, fraud stack, surgery recs, peptide stack, missing data, and ceiling analysis.
Photo protocol — if only frontal provided it flags which structural elements cannot be confirmed and requests profile + 45° angle before finalizing.
That’s all it can so currently I gave way more stuff to add but I’m still at work lol
The link for the jsx for anyone interested :
The rater is still not complete at all made it in my free time at work will update and maybe post it here when Im done tweaking it to perfection.
————————————————————-
Rating Basis
The PSL (Physical Subjectivity Level) scale — 0 to 8
Not a vibe-based rating. Every number is anchored to named celebrities so the model always calibrates against known references rather than drifting subjectively. The rating measures three things simultaneously:
Facial Harmony — how well proportioned the face is geometrically (neoclassical canons, golden ratio, facial thirds/fifths)
Sexual Dimorphism — how strongly testosterone markers read on the face (fWHR, brow ridge, jaw angle, chin projection — backed by Carré & McCormick 2008, Gangestad & Thornhill research)
Feature Quality — individual feature assessment across all 14 markers in a fixed order so nothing gets missed
It also applies a BF% penalty in real time — if you’re above 30% it gives both a current rating AND a lean estimate, rather than rating the fat as your face.
What Was Implemented From My Suggestions
Structured hierarchy — the system prompt is divided into clear labeled sections (rating scale → assessment order → BF protocol → phenotype → contacts → fraud tiers → surgery → peptides → output format) instead of dumped wiki text. This forces consistent behavior.
Mandatory 14-marker assessment order — always assessed in the same sequence so no feature gets skipped or weighted arbitrarily.
BF% adjustment protocol — five specific ranges (under 15% / 15-20% / 20-25% / 25-30% / 30%+) each with a defined accuracy level and lean estimate instruction.
Phenotype classification first — identifies MENA/Mediterranean, Northwestern European, East Asian, Sub-Saharan before assessing anything else, then applies phenotype-specific criteria.
Contact lens guide by phenotype — specific brand recommendations (Solotica, Anesthesia, Desio) with exact color names for each skin type. Explicitly bans pure blue on olive skin.
Three-tier fraud system — 3.5→4.5 softmaxxing, 4.5→5.5 halo stacking, 5.5→6+ medmaxxing, each with specific interventions for that PSL range rather than generic advice.
Surgery table by deficit — each structural weakness mapped to a specific procedure with PSL gain estimate and risk level (Low/Med/High/Very High).
Peptide stacks by goal — four separate stacks (collagen/skin, jaw/connective tissue, hair preservation, GH optimization) with danger ratings and cycling protocols.
Consistency anchoring — instructed to compare against 3 celebrity PSL anchors before finalizing any rating to prevent drift.
Mandatory 12-point output format — every response must include phenotype, BF estimate, current PSL, lean PSL, tier classification, all 14 features, dimorphism score, harmony score, fraud stack, surgery recs, peptide stack, missing data, and ceiling analysis.
Photo protocol — if only frontal provided it flags which structural elements cannot be confirmed and requests profile + 45° angle before finalizing.
That’s all it can so currently I gave way more stuff to add but I’m still at work lol
The link for the jsx for anyone interested :
