Heracles

Evidence-aware resistance training

Know what your training data actually means.

Log every set in the gym. See exactly what it says about your strength and which muscles it trained — with the working shown.

Heracles combines a persistent training ledger with equipment-aware history, RIR-aware strength estimation, body evidence, and explicit movement and muscle models. The observations stay separate from the interpretations built on top of them.

See it in action
  • 116catalogued exercises
  • 4preserved e1RM estimates
  • 40detailed muscle entities
  • 116 exercises
  • 40 movement patterns
  • 40 muscle entities
  • 13 readable muscle groups
  • 4,640 authored relevance cells
  • 1,600 functional-anatomy cells
  • 4 e1RM estimates per set
  • 0 tonnage metrics

The app

One place for the session and everything it proves.

Heracles is built for the gym floor first: fast set entry, previous performance on screen, autosave that survives bad signal. Afterwards, the same records become history, trends and model-backed analysis.

Home

Your training at a glance.

Start or resume a workout in one tap. A year-long heatmap shows every day you logged working sets, and your latest sessions sit right below with the muscle groups they trained.

  • Resume an interrupted workout exactly where it stopped
  • Heatmap shades use fixed set-count bands, so old days never re-scale
  • Jump straight into any recent session

Screens show illustrative example data.

How it works

Follow one set all the way through.

Train, record, interpret, track. Heracles keeps what you did separate from what it infers, so every number further down the chain can be traced back to the set that produced it.

01

Train

Start from a preset or from scratch. Heracles lays out your exercises with blank sets and puts last session's performance on the same exercise and equipment right next to them.

From preset · Push AOrder rotated · blank sets
  1. Chest Press (Incline 30)3 sets
  2. Pushdown (V-Bar)3 sets
  3. Chest Press (Incline)3 sets
  4. Lateral Raise (Standing)3 sets
  5. Fly3 sets
Last time on this machine — 23 Sep Set 1: 30 kg per dumbbell × 8 · 2 RIR Set 2: 30 kg per dumbbell × 8 · 1 RIR Set 3: 30 kg per dumbbell × 7 · 1 RIR
02

Record

Log load, reps and reps in reserve. Every set is classified the moment it is complete, and the edit is saved in the background — even when the gym has no signal.

Weight32.5kg per dumbbell
Reps7completed
RIR1reported

1 RIR · Working set

  • Warm-upMarked explicitly. No RIR. Excluded from analysis.
  • Working setReported RIR 0–3. Counts toward analysis.
  • 4+ RIRKept as history. Not counted as a working set.
  • DraftPartly filled. Blocks concluding until finished.
Saved ✓Saving…OfflineFailed
03

Interpret

Two equations, run on completed reps and on reps plus reported RIR, give four strength estimates. They are kept apart — never averaged into one false certainty. The exercise's authored relevance row turns the set into weighted sets per muscle.

Estimated 1RM · kg per dumbbell
Brzyckicompleted reps32.5 × 36 ÷ (37 − 7)39.00
Epleycompleted reps32.5 × (1 + 7 ÷ 30)40.08
Brzyckireps + RIR32.5 × 36 ÷ (37 − 8)40.34
Epleyreps + RIR32.5 × (1 + 8 ÷ 30)41.17

A model spread — not a confidence interval, not a measured max.

Chest Press (Incline 30) · relevance row
  • Pectoralis Major — Clavicular1.00
  • Pectoralis Major — Sternocostal0.75
  • Anterior Deltoid0.75
  • Triceps Brachii — Lateral Head0.75
  • Triceps Brachii — Medial Head0.75
  • Triceps Brachii — Long Head0.50

Group value = highest child-muscle coefficient per set: Chest 1.00 · Shoulders 0.75 · Triceps 0.75 weighted sets.

04

Track

One representative working set per session and equipment becomes a point in your progression, drawn as the full range of its four estimates. Body weight, tape measurements and muscle exposure sit alongside as separate evidence.

Chest Press (Incline 30)Estimated 1RM range · kg per dumbbell

Illustrative example data. Each band spans the four estimates for that session's representative set.

Try the math

Four estimates. Zero black boxes.

This calculator runs the same estimation code as the app. Change a set and watch the four perspectives move — and see exactly when Heracles refuses to produce a number.

Reported RIR
Equipment

2 RIR · Working set

Brzycki completed reps
124.14100 × 36 ÷ (37 − 8)
Epley completed reps
126.67100 × (1 + 8 ÷ 30)
Brzycki reps + RIR
133.33100 × 36 ÷ (37 − 10)
Epley reps + RIR
133.33100 × (1 + 10 ÷ 30)

Estimated 1RM in kg

Calculated estimates from equations, not a measured one-rep max. The spread between them is a model range, not a confidence interval.

The model, in full

4,640 decisions. All of them public.

Every one of the 116 exercises is rated against every one of 40 muscles. This is the complete hypertrophic-relevance matrix the app uses — loaded live from the published source file. Pick an exercise and inspect its row.

Compare
0 · not counted 0.25 · small but worthwhile 0.50 · substantial secondary 0.75 · major target 1.00 · principal target How every cell was decided

An authored model of worthwhile involvement for one hard set — not measured force, stimulus or growth. Columns are ordered by each exercise's principal muscle group.

Current product

Built for the training session and the evidence it creates.

Heracles connects live training, personal history, strength interpretation and physical measurements in one resistance-training system — and it installs to your home screen like a native app.

Live training

Persistent workout sessions

Resume an in-progress workout, reference previous performance, preserve equipment context and recover durable edits when connectivity fails.

Programming

Reusable workout presets

Create reusable exercise pools with set counts, then start a session without copying stale load, RIR or equipment assumptions forward.

Strength

Four e1RM perspectives

Brzycki and Epley estimates are retained for observed repetitions and for repetitions adjusted by reported RIR. Heracles does not average them into false certainty.

Body weight

Strength in body-weight context

Import body-weight evidence or synchronize Apple Health data, then optionally view each strength estimate relative to measured or interpolated body weight without extrapolation.

Measurements

Standardized circumference history

Record repeatable body measurements by site and compare like with like over time, including separate relaxed and flexed upper-arm series.

Hypertrophy model

Explicit muscle relevance

Versioned movement and muscle matrices connect exercises to forty detailed muscle entities while thirteen broader groups keep the interface readable.

Product principles

What Heracles refuses to calculate matters too.

Many apps multiply weight × reps and call it volume. Twenty-five easy reps with a light weight can out-score eight hard ones with a heavy weight, and a leg press out-scores a leg extension by sheer stack size. Tonnage measures mass moved — not effort, mechanics or which muscles did the work — so Heracles never uses it as a training, progress or hypertrophy metric.

Preserve the observation. Label the interpretation. Expose the assumptions. Do not manufacture certainty where the evidence does not support it.
Why tonnage is excluded
↗

Performance stays exercise-specific

Load and repetitions are interpreted within the same movement and equipment context instead of being compared across mechanically different exercises.

≠

Tonnage is not called volume

Weight × repetitions measures external mass moved repeatedly. It is not treated as hypertrophic stimulus, workload quality or progression.

◎

Estimates stay plural

Observed and RIR-adjusted strength estimates remain separate, making disagreement between models visible instead of hiding it behind an average.

§

Uncertainty remains visible

Recorded observations, calculated estimates, authored models, scientific evidence and product decisions are presented as different kinds of knowledge.

Longitudinal evidence

Different signals stay different.

The system does not force training performance, body weight, circumference and modelled muscle relevance into a universal progress score. Each answers a different question.

EvidenceWhat it helps answerBoundary
EnteredWorkout historyWhat was trained, with which exercise, equipment, load, repetitions and effort?A completed set is an observation, not a direct measure of adaptation.
CalculatedStrength estimatesHow is exercise-specific performance changing across comparable sessions?e1RM values are equation-based estimates, not measured one-repetition maxima or confidence intervals.
MeasuredBody weightHow does strength compare with body mass on the workout date?Relative strength is shown only where measured or interpolated body-weight evidence exists; Heracles does not extrapolate.
MeasuredBody measurementsHow are standardized circumference observations changing over time?Tape measurements remain raw observations and are not converted into body-fat or body-composition estimates.
AuthoredMuscle relevanceWhich muscles are worth considering for a sufficiently hard set of a given exercise?This is an authored relevance model, not measured force, stimulus or hypertrophy.

Fabbro Systems

Evidence → state → action.

Heracles goes deep on one domain. It remains useful as a standalone training system while also producing higher-quality physical-performance evidence for the wider Fabbro Systems ecosystem.

Heracles observes and interprets a domain. Kleos models current state. Ariadne decides and organizes desired movement.

Scientific foundations

The reasoning is part of the product.

These documents expose the evidence standard, data model, authored assumptions, limitations and product decisions behind the interface. Read them before you sign in.

Evaluation protocol

How studies are judged

Evidence is assessed claim by claim, with attention to directness, trained populations, comparators, adherence and uncertainty.

Study-selection protocol
Authored taxonomy

Muscle-group taxonomy

Forty muscle entities balance anatomical resolution with what can be modelled and defended for hypertrophy applications.

Muscle taxonomy
Authored model

Movement-pattern coefficients

Independent contribution ratings organize exercises biomechanically without pretending to be measured joint torque or stimulus.

Functional anatomy

Movement to muscle function

A 40 × 40 matrix records anatomical capability as a functional prior, while leaving exercise context to downstream layers.

Functional-anatomy model
Derived model

Exercise to muscle composition

Ordinary matrix composition provides a traceable bridge from exercises through patterns to muscles without becoming a growth score.

Composition method
Exercise-specific model

Hypertrophic relevance

An authored 116 × 40 layer filters anatomical possibility through exercise context for one sufficiently hard, competently performed set.

Relevance method
Model limitations

Current mapping limitations

The project documents missing context—including force-sharing, muscle length and resistance profiles—rather than hiding it.

Current limitations
Product decision

Why tonnage is excluded

External mass × repetitions is rejected as a hypertrophy, progression, workload or personal-statistics metric.

Tonnage decision
Product standard

Design rules

Transparent calculations, defensible metrics, owner-scoped data and visible limitations govern what reaches the interface.

Design rules

Start with your next session

Every set is evidence. Start keeping it.

Sign in with Google and log your next workout. Your training data belongs to your account alone, enforced by row-level security in the database — not just hidden in the interface.

Read the science first
  • Installs to your home screen
  • Autosaves through bad signal
  • Every calculation documented