The interpretation core is small and dense. Every figure is counted directly from the codebase.
221
intake signals
the read instrument · 8 domains
5
interpretive pillars
FMM primary + PTS/HABNS/RRS/BIRS
4 + 11
Fat Map zones
macro + extended
3
body states
the classification
CFFS
one read
refreshed weekly as the CFWS
39k
lines of canon
the doctrine behind it
This is the engine alone. The application built on it — the Performance Coaching environment, ~126k lines across 192 routes and 65 tables — is the far larger codebase, documented separately. The interpretation is small; the system that makes it usable is large.
What the engine reads. Mostly deterministic capture; the only model here transcribes blood panels. The capture surfaces are delivered by the application — the signal schema is the engine's.
Foundational Intake
deterministic8 domains · ~200 scale signals
The deep input. Eight scored signal domains map one-to-one onto eight JSONB buckets the engine reads: Fat Map, Injury, Training, Nutrition, Schedule, Sleep, Stress, Medications/Supplements.
→ intakes.*
Baseline
deterministicmeasurements + photos
Bodyweight, waist, hips, chest, and three photos (front / side / back). The photos feed the engine’s vision read; bodyweight floors the nutrition protein anchor.
→ baselines.*
Medications
deterministicfree text · longitudinal
A single free-text field soliciting hormonal and non-hormonal pharmacology. Read by the engine for signal class, never moralised, never propagated as a compound name to client prose.
→ clients.medications
Blood Panel
Claude Haiku 4.5optional · multimodal
Client uploads a lab PDF/photo. The engine transcribes every marker against the lab’s own printed range — transcription, not diagnosis. Only a coach-approved panel reaches the interpretation.
→ blood_panels.*
Coach-Facing Foundational Synthesis. The heart of the engine. Interpretation terminates here.
Body Recode™ Interpretation Engine
Claude Haiku 4.5 · multimodalReads
Does
Pre-summarises the scale domains, reads the photos as one signal stream feeding Spatial Patterning, and resolves the whole picture through the Fat Map’s four macro zones. Conservative resolution always overrides the optimistic read. It classifies the body state and rates four exposure-readiness signals — and prescribes nothing.
Produces
Vision
Front / side / back photos are base64-encoded and passed to the same Haiku call. Eight rules govern the read: photos are one signal, never the conclusion, and never an aesthetic verdict.
Blood markers
Only a coach-approved panel is read, against the lab’s own ranges. A single out-of-range marker is short-arc evidence, never a diagnosis.
One live read
Generating archives the previous CFFS. reassessment_flagged is always set server-side, never by the model — it is a temporal construct.
Coach-Facing Weekly Synthesis — the weekly re-read. Still interpretation, still coach-facing, still prescribes nothing.
Reads
The week’s check-in signal, anchored to the CFFS baseline. The capture itself is a Layer 2 surface; the engine consumes it the way it consumes the intake.
Rule
Readiness moves one notch at a time from the CFFS anchor. Red is the floor. Two-notch jumps only on a named safety event. Single-form answers can’t downgrade alone.
Produces
A weekly synthesis + the four readiness colours → cfws.*. The accumulated sequence becomes the input to the block-end arc.
The engine’s self-governance: it decides when the read has drifted enough to need refreshing. Deterministic. Prescribes nothing.
Drift
Any Red in the latest CFWS = high. The same signal Amber/Red across two consecutive weeks = sustained instability. A two-notch drop = high.
Reassessment routing
Drift routes to a lightweight, delta, or full re-read. Signal-based triggers are suppressed until the CFFS is ≥21 days old; a 12-week cap forces a full re-intake.
RPE creep as signal
Avg logged RPE ≥ prescribed +1.0 = creep; +2.0 or any set ≥9.5 = severe. The logging is a Layer 2 surface; the monitor consumes it as one more drift input.
Client readings
Generators
The weekly loop
Fat Map Method™PRIMARY
Primary pillar. Fat accumulation read as hormonal and metabolic signalling, not energy surplus. Where the body stores fat reflects its adaptive response to the current environment.
Performance Training System
Physical stress exposure, training load, fatigue and adaptation. Defines what training the body can tolerate now.
Hybrid Animal-Based Nutrition
Nutritional sufficiency and biological threat signalling. Is intake supporting regulation or reinforcing scarcity.
Recovery & Regulation System
Capacity to restore and return to baseline. Distinguishes genuine recovery from suppression. RRS always overrides training.
Behaviour, Identity & Rhythm
Emotional load and identity constraints as structural biological stressors. Compliance and identity threat are inputs, not psychology.
Remediation
public: DepletedUnder load it cannot resolve. Performance-level intervention here is counterproductive. Most who present believing they are elsewhere are actually here.
Optimisation
public: TransitioningStabilised. Capacity exists to pursue body composition and performance. Where most coaching begins, but few arrive here.
Post-Optimisation
public: ReadyPerforming. The goal shifts to identity-level performance and long-term maintenance. Built over time, never forced.
MZ1 · Stress Belt
Stomach / Waist
Cortisol and adrenaline dominance
Extended zones
MZ2 · Gut & Bloat
Digestive region
Insulin timing disruption
Extended zones
MZ3 · Hip & Thigh
Hips / Thighs
Reproductive hormone and metabolic conservation
Extended zones
MZ4 · Upper Body Stress
Upper body
Nervous system load, adrenaline, sleep retention
Extended zones
Model
Claude Haiku 4.5 for the read — one model, multimodal. Photos and lab files run in the same call as the text.
JSON
A brace-counting extractor, never a greedy regex. Every parse survives prose, code fences and stray commentary.
Em dashes
Recursively stripped from every generated string. A banned AI-writing signal across the whole platform.
One live read
Generating a CFFS or CFWS archives the prior one. reassessment_flagged is always server-set, never by the model.
Non-diagnostic
No interpretive surface names a disease, alters a lab range, or gives medical advice. A marker is a hypothesis, not a finding.
Governed, not assumed
Drift and reassessment are surfaced, not silent — the integrity of the read is monitored across the block.
It interprets, never prescribes
Layer 1 terminates at the CFFS. Prescription is Layer 2 and is always derived from the interpretation, never the other way round.
Non-diagnostic, absolutely
No stage names a disease, alters a lab range, or gives medical advice. Markedly abnormal values are referred to a GP in neutral language.
Pattern, not event
Interpretation is pattern-based and conservative resolution always wins. One reading is a hypothesis; the arc is the evidence.
No aesthetic verdict
Photos are signal, not judgment. Words like overweight, lean, soft or athletic are banned from the read.
The machinery stays hidden
No CFFS, RPE or cortisol language ever reaches the client. The reading family translates everything into one voice.
Brand boundary held
Structured compound tracking is Arete’s lane. Body Recode captures it only as free text and never propagates compound names to client prose.