Built on Sports Science. Not on AI Hype.

1. The Problem with Black-Box "AI Workouts"

In recent years, the fitness software industry rushed to replace training spreadsheets with Large Language Model (LLM) prompts. Users enter a prompt like *"build me a 4-day workout"* and receive a text output generated by probabilistic token matching.

From a physiological perspective, this is irresponsible. LLMs possess no internal mathematical model of fatigue, joint stress, or weekly volume accumulation. They frequently hallucinate:

• Erratic Volume Spikes: Prescribing 30 sets for chest in a single week followed by 4 sets for legs.
• Dangerous Beginner Loading: Prescribing high-complexity barbell movements to total novices with zero Repetitions in Reserve (RIR 0 / failure).
• Contradictory Joint Loading: Recommending heavy overhead presses to trainees who explicitly flagged shoulder impingement.

Grada was engineered from the ground up to solve this failure mode through a strict separation of concerns.

2. Deterministic Mathematics Core

In Grada, an artificial intelligence never decides your exercises, your sets, your reps, or your progression increments. Every structural training decision is computed deterministically in pure TypeScript by our plan engine.

Where does AI fit in? Terra, our AI coach persona, acts strictly as a natural-language communicator. Terra receives the mathematical output that the engine already computed and explains the reasoning in calm, encouraging prose. Terra has read-only access to your plan and cannot alter workout numbers.

3. The 12-Rule Hard Safety Gate

Before any 4-week training cycle is rendered on your device, the generated schedule must pass through an automated validator gate. If a plan violates even a single physiological rule, generation fails loudly and the engine triggers a conservative fallback or throws an explicit error:

1. Experience Gate: Advanced exercises (complexity rating > 3) are never assigned to beginners.
2. Joint Load Exclusion: Flagged joints (e.g. knee, shoulder, lower back) automatically exclude aggravating exercise categories from every candidate pool.
3. RIR Safety Floor: Beginners and senior lifters are locked to a minimum of 2 Repetitions in Reserve (RIR ≥ 2) to ensure safety and motor learning.
4. No Empty Blocks: Every scheduled workout session contains valid primary exercises.
5. Modality Integrity: Strength, steady Zone 2 cardio, and mobility exercises are never mixed incorrectly within specialized slots.
6. Zone 2 Minimum Duration: Aerobic base sessions are scheduled for a minimum of 15 continuous minutes.
7. Mobility Duration Ceiling: Mobility routines are capped at 20 minutes to maintain high adherence.
8. Catalogue Resolution: Every exercise ID must resolve against our verified 400+ canonical exercise catalogue.
9. Conditioning Exclusion: Conditioning exercises (e.g. burpees) are never placed in hypertrophy primary strength slots.
10. Activation Drill Filtering: Warm-up activation drills are barred from intermediate/advanced primary compound blocks.
11. Frequency Bounds: Strength routines require a verified weekly schedule of 3 to 7 days per week.
12. Hypertrophy Volume Landmarks: Every targeted muscle group receives at least 10 sets per week for muscle-building goals.

4. Volume Landmarks & Progression Models

Muscle hypertrophy and strength adaptations are governed by the volume dose-response curve established in sports science literature (Schoenfeld et al., 2017). Grada implements calibrated volume landmarks across all 4 product goals:

For beginner trainees, linear load progression coupled with exercise familiarity is the established gold standard (Ratamess et al., 2009). For intermediate and advanced lifters, the engine provides progressive volume overload without exceeding Maximum Recoverable Volume (MRV).

5. Daily Readiness Autoregulation

A training plan that cannot adapt to real life is a plan designed to fail. Grada captures subjective readiness through a 30-second pre-workout check-in evaluating sleep quality, general fatigue, muscle soreness, and available time.

When readiness is low, the engine adjusts session parameters in real time:

• Time Compression: If you only have 20 minutes instead of 45, the engine prioritizes primary compound movements and trims secondary accessories.
• Load Modulation: Working loads and target RIR are eased back to protect recovery while maintaining movement practice.
• Layoff Recovery Ramp: When returning after a break of 7+ days, volume is scaled back according to detraining kinetics (Mujika & Padilla, 2000) so you never restart from scratch or risk acute overreaching.

6. The 60-Persona Testing Harness

Before shipping any algorithm update, Grada runs an automated quality evaluation suite across 60 synthetic control profiles (40 standard lifecycle personas + 20 adversarial edge cases):

• Octogenarian with Joint Constraints: Tested for low-complexity, knee-friendly bodyweight movement selection.
• Time-Compressed Executive: Tested for 15-minute micro-session integrity without compound lift truncation.
• Returning Athlete after Layoff: Tested for detraining decay curves and conservative reload pacing.

Every single synthetic profile must generate a 100% valid, validator-passing plan before code is merged.

How the Engine Approaches Your Goal