MyoNoesis

Evidence-based hypertrophy mesocycles. A deterministic, transparent training engine. Mesociclos de hipertrofia basados en evidencia. Un motor de entrenamiento determinista y transparente.

▶ Live app · myo.noesisargentina.com

English · Español

---

English

MyoNoesis is an offline-first PWA that plans and runs hypertrophy training mesocycles. You define one base week (days, exercises, sets, rep ranges, starting loads); the engine generates the full cycle — volume ramp, weekly RIR targets, final deload — and then autoregulates every next prescription from your actual logged performance.

It is intentionally the opposite of black-box training apps: every number is explained, every scientific threshold lives in a versioned config file, and the UI communicates uncertainty instead of faking precision.

Why this project is interesting (engineering-wise)

• Pure domain engine, 77 unit tests. All training logic (src/domain/) is pure TypeScript functions — no React, no DB, no Date.now(). Deterministic and fully testable: progression models, load rounding against real plate inventories, mesocycle generation, deload triggers, cross-cycle correlations.
• Policy as data. Every scientific constant (volume landmarks, RIR schedules, progression increments, deload triggers) lives in rules.config.json, validated with Zod at load. Recalibrating the engine never touches code. The reasoning behind every number is documented in SCIENCE.md, with primary sources.
• Offline-first for the real world. Gyms have bad signal. IndexedDB (Dexie) as the single source of truth, service worker precache (Serwist), protected storage, screen wake-lock during sessions, and full JSON backup/restore. No backend, no account, no tracking: your data never leaves your device.
• Epistemic honesty as a product feature. MEV/MAV/MRV landmarks are labeled as heuristics (they are), deloads are suggested from multi-signal fatigue evidence (never "you exceeded your MRV"), and single-cycle stats carry an explicit small-n warning.

Features

• 📋 Base week editor with live fractional-set volume per muscle vs. MEV/MAV/MRV landmarks, and a preloaded 6-day routine template to start from.
• 🔄 Four progression models (linear, double progression, DUP, block) that equalize volume — per the evidence that periodization models are equivalent for hypertrophy.
• 🏋️ Session runner: per-set logging (kg × reps × RIR), rest timer, wake lock, instant persistence.
• 🧮 Autoregulation: next load/reps computed from achieved reps at target RIR, rounded to your plate inventory; if the minimum jump exceeds the target %, it adds reps instead.
• ⏸️ Multi-mesocycle management: pause a cycle, interleave another, resume without losing progress (progression is session-driven, not calendar-driven), and "continue progression" — a follow-up block inheriting loads from real performance (deload week excluded).
• 📏 Body measurements with an interactive SVG tape-placement guide, re-prompted after each cycle and every ~6 months.
• 🏃 Extra activities (running, swimming…) logged as optional covariates.
• 📊 Cross-cycle correlation heatmap: Δe1RM, adherence, sleep, protein, energy balance, extra cardio and measurement deltas, per progression model — with an explicit "this is noise below ~4-6 cycles" disclaimer.
• 📤 Excel export (results per set, sessions, volume, activities) and one-tap JSON backup with a 7-day reminder.
• 🌗 Light/dark theme, installable on iOS/Android, Spanish UI.

Architecture

Pure domain + adapters (repository pattern). The UI never touches persistence directly.

UI (Next.js App Router · React) ── useLiveQuery (reactive render)
 └─ store/ (Zustand: UI/session state)
     └─ db/ repositories → Dexie → IndexedDB      ← only layer touching persistence
         └─ domain/  PURE FUNCTIONS (read rules.config.json)
            progression · load rounding · meso generation · deload triggers
            swap suggestions · stats · 1RM · correlations — 77 unit tests

Stack: Next.js (static export) · TypeScript strict · Tailwind 4 · Dexie.js · Zustand · Serwist · Recharts · Zod · Vitest · SheetJS. Deployed to GitHub Pages via Actions under a custom domain.

Getting started

npm install
npm run dev        # dev server (service worker disabled on purpose)
npm run test       # 77 engine unit tests (Vitest)
npm run typecheck  # tsc --noEmit
npm run build      # static export → out/

Open http://localhost:3000. All data lives in your browser's IndexedDB.

Roadmap

Post-meso AI advisor (LLM interprets cycle stats — advises, never prescribes) · intra-subject landmark recalibration after 3+ cycles · plate calculator · superset support · English UI.

---

Español

MyoNoesis es una PWA offline-first para planificar y ejecutar mesociclos de hipertrofia. Definís una semana base (días, ejercicios, series, rangos de reps, cargas iniciales); el motor genera el ciclo completo — rampa de volumen, RIR objetivo por semana, deload final — y después autorregula cada prescripción siguiente a partir de tu rendimiento real registrado.

Es deliberadamente lo opuesto a las apps de entrenamiento "caja negra": cada número se explica, cada umbral científico vive en un archivo de configuración versionado, y la interfaz comunica incertidumbre en lugar de fingir precisión.

Por qué es interesante (técnicamente)

• Motor de dominio puro, 77 tests unitarios. Toda la lógica de entrenamiento (src/domain/) son funciones puras de TypeScript — sin React, sin base de datos, sin Date.now(). Determinista y testeable: modelos de progresión, redondeo de cargas contra inventario real de discos, generación de mesociclos, disparadores de deload, correlaciones entre ciclos.
• Políticas como datos. Toda constante científica (landmarks de volumen, calendarios de RIR, incrementos, disparadores de deload) vive en rules.config.json, validado con Zod. Recalibrar el motor no toca código. El porqué de cada número está documentado en SCIENCE.md con fuentes primarias.
• Offline-first para el mundo real. En el gimnasio no hay señal. IndexedDB (Dexie) como única fuente de verdad, precache con service worker (Serwist), almacenamiento protegido, pantalla siempre encendida durante la sesión, y respaldo/restauración completa en JSON. Sin backend, sin cuenta, sin tracking: tus datos nunca salen de tu dispositivo.
• Honestidad epistémica como feature. Los landmarks MEV/MAV/MRV se etiquetan como heurísticos (lo son), el deload se sugiere a partir de señales multi-fuente de fatiga (nunca "superaste tu MRV"), y las estadísticas de un solo ciclo llevan aviso explícito de n pequeño.

Funcionalidades

Editor de semana base con volumen fraccionado en vivo por músculo · cuatro modelos de progresión con volumen igualado · ejecución de sesión con registro por serie, timer y wake lock · autorregulación con redondeo a tus discos · gestión multi-meso (pausar, intercalar, reanudar, continuar progresión heredando cargas del rendimiento real) · medidas corporales con guía visual de dónde pasar la cinta · actividades extra · heatmap de correlaciones entre ciclos (Δ1RM, sueño, proteína, cardio, medidas, por tipo de meso) · export a Excel · respaldo JSON con recordatorio · tema claro/oscuro · instalable en iOS/Android.

Correr el proyecto

npm install && npm run dev   # abrir http://localhost:3000
npm run test                 # tests del motor

En Windows: doble clic en Abrir MyoNoesis.bat.

---

Idea y desarrollo de Rodrigo Skerlak · myo.noesisargentina.com

Parte de la familia Noesis (psicometría · entrenamiento).