Reps: A Recovery Score That Rewrites The Workout, Not Just The Dashboard

Most runners reading this already own a recovery score. WHOOP has one, Garmin ships Body Battery and Training Readiness, Apple Watch added Vitals, and Oura has been scoring sleep for years. Another 0 to 100 number is not, on its own, interesting.

The unsolved part is what happens after the number appears. The score lives on one screen, and the training plan lives on another, and reconciling them is a judgement call made at 7 am, half-awake, with no rule for how much to back off. Most athletes either ignore a bad score and train as planned, or overreact and bin a session that only needed its intensity trimmed.

Reps, an iPhone and Apple Watch app from Toronto-based solo developer Siddharth Natamai, is built on the bet that this last step is the product. The readiness score feeds a decision engine that edits the day’s planned session before the athlete opens it.

How it works

TL;DR: Each morning Reps reads the athlete’s Apple Health data, computes a 0 to 100 readiness score, converts it into a maximum training intensity, and applies that ceiling to the workout already scheduled for today, downgrading, shortening or swapping the session before it is generated.

The chain runs in four stages, and the interesting part is stages two and three:

• Score. A weighted composite produces a number and one of five bands: Optimal (85 to 100), Good (70 to 84), Moderate (55 to 69), Low (40 to 54), Poor (0 to 39).
• Decide. A separate layer grades four domains (autonomic recovery, sleep quality, training-load safety, and injury risk) Green, Amber, or Red, then derives one maximum intensity ceiling: Unrestricted, Moderate, Light, or Rest Only.
• Rewrite. The ceiling is applied to today’s entry in the weekly plan. A high-intensity day under a Light ceiling becomes moderate; a day under Rest Only becomes active recovery at half the planned duration.
• Generate. The ceiling then travels with the workout-generation request, so the session produced is already constrained rather than written and then apologised for.

The difference from a conventional readiness screen is that stage three writes to the plan. The athlete never has to translate “62 today” into a training decision. The translation has already happened, and arrives as a specific proposed change with its reason attached.

What goes into the readiness score

Sleep carries the largest single share at 19.4%, then subjective wellness at 18.9%, HRV at 15% and training load at 12.3%. The rest splits between previous-day protein (7.6%), resting heart rate (7.2%), hydration (6.6%), alcohol in the previous 24 hours (5.7%), respiratory rate against baseline (4.7%) and walking symmetry and steadiness (2.8%).

The HRV share is worth explaining, because until late July 2026 it was zero. The original reasoning was specific to Apple Watch: it samples HRV opportunistically through the day rather than producing one clean overnight figure the way a chest strap does, and the app was reading the single most recent sample, which could be a stressed mid-afternoon reading. Scoring that would have been worse than scoring nothing.

What changed is the input, not the appetite for HRV. Reps now takes the median of the SDNN samples inside each qualifying night (three hours or more of actual sleep, naps excluded) and scores that. The 15% is split 8:7 between last night alone and a seven-night rolling trend, each compared against the athlete’s own baseline rather than a population curve. That split is deliberate: a rolling average is the right instrument for detecting chronic maladaptation and the wrong one for answering “should I train today”, because it lags an acutely bad night by seven to one. Garmin and Oura both score the trend; WHOOP scores the night just ended. Reps does both and weights them separately.

There are two details worth flagging: missing factors are redistributed, not zeroed, so an athlete who logs no alcohol is not silently penalised; and until enough history exists for a personal baseline, the score is labelled as calibrating rather than presented as a verdict.

A worked example: the same Tuesday at 52 and at 88

Take an athlete whose weekly plan has Tuesday down as a 55-minute upper-body session at high intensity, written on Sunday. What Reps does with it depends entirely on Tuesday morning.

At 7 am with a readiness of 52 (the Low band), the score is dragged down by a sleep quality of 47%, below the 50 threshold that flags the sleep domain Red, and by a resting heart rate elevated against its three-day rolling average combined with an HRV reading roughly half a standard deviation below baseline, which together flag the autonomic domain Red.

A Red autonomic domain caps the ceiling at Light. The consequences are concrete rather than advisory:

• The readiness card reads “Poor sleep is severely limiting recovery. Prioritise rest and recovery today.”
• Tuesday’s session is rewritten from high intensity to moderate, tagged with the reason “Intensity reduced — CDS ceiling is light”, and presented as a proposed change the athlete can accept or decline.
• The generation request carries the Light ceiling, which becomes a hard instruction (“LIGHT only — recovery training”) so the session written is a lighter one from the outset. The 55 minutes are not cut; the intensity is.

At 7 am with a readiness of 88 (the Optimal band), all four domains are Green, the ceiling stays Unrestricted, and because the score is at or above 85, the decision layer escalates from “ready” to peak. No adjustment is proposed at all: the daily check returns nothing, Tuesday’s session stands exactly as planned, and the generator is told: “No restrictions — train at any intensity.”

The same plan, the same athlete, the same weekday: two different sessions, decided before the app was opened. Between those poles, an Amber in the autonomic, training-load or injury domains caps the day at Moderate, and a Red sleep domain alone caps it at Moderate rather than Light. If the ceiling drops to Rest Only (an irregular rhythm notification, a critical blood-oxygen reading, or rest mode enabled manually), the session is replaced rather than trimmed.

Training load runs alongside as an acute: chronic workload ratio on conventional bands (below 0.8 detraining, 0.8 to 1.3 optimal, 1.3 to 1.5 caution, above 1.5 elevated risk), fed automatically from Apple Health workouts and active energy.

Where the loop stops

There are three limits, because a closed loop that overrides the athlete is a worse product than one that does not.

The athlete’s own request wins: ask for a specific session, and that instruction explicitly outranks the recovery recommendation. Most constraints are individually overridable. The exceptions are safety-critical ones, where an irregular rhythm or critical blood-oxygen reading cannot be dismissed.

The AI coach, Rex, is cloud-assisted rather than on-device. Some lighter queries are routed to Apple’s on-device model, and the app labels which one answered, but the main coaching conversation sends context to a server.

And Reps is not a medical device. It contextualises an Apple Watch ECG against that day’s training, sleep and resting-HR trend, but will not diagnose or re-interpret the Watch’s own classification.

Data and integrations: what Reps can and cannot see

Reps reads all of its health data through Apple Health. There is no Garmin Connect integration, no WHOOP API, no Oura API, no account linking and no hardware pairing. The one direct vendor connection runs outbound: Strava is write-only, an opt-in link, off by default, that posts finished Reps workouts to a Strava feed and imports nothing.

The practical consequence for anyone wearing a non-Apple device:

• What generally does come across: workouts, steps, active energy and body-weight entries, because Garmin, WHOOP and Oura write those into Apple Health.
• What generally does not: HRV, resting heart rate, stress and Body Battery. Those vendors do not write them to Apple Health, so Reps cannot see them, and the readiness score is built largely from exactly those signals.

This leads to the requirement worth stating plainly rather than burying: Reps need an Apple Watch for the recovery half of the app to mean anything. Without one, there is no overnight sleep staging, no dense resting-heart-rate series, and no HRV. Strength logging, nutrition and the coach still work; the closed loop above does not.

Within Apple Health, Reps reads sleep stages, heart rate, resting heart rate, HRV, respiratory rate, blood oxygen, ECG, VO2 max, workouts and routes, active energy, steps, walking steadiness and asymmetry, running dynamics, body composition and nutrition. It writes back its own workouts and GPS routes, logged nutrition and water, and body measurements.

Connectors for Garmin and WHOOP exist in the shipping app but are inert, with no credentials and no live server-side, because Garmin’s developer programme is currently closed to new applications with no reopening date. A Fitbit row appears in settings marked “Coming Soon” and is not implemented. Readers should treat all three as absent rather than pending.

Requirements and pricing

• Platform: iPhone on iOS 26 or later, with a native Apple Watch app on watchOS 26, plus iPad support, Home and Lock Screen widgets, Live Activities and Siri Shortcuts.
• Free: tracking, strength logging, manual nutrition entry and the readiness score itself.
• Reps Pro: $9.99/month or $69.99/year, with a one-week free trial on both, plus a $99.99 one-time lifetime option. Those are US-storefront prices; other regions differ. Pro unlocks the AI features: the coach, generated workouts and plans, photo-based nutrition analysis and health reports.

Pro also covers a caffeine model that derives a daily cut-off time from a roughly five-hour half-life and the athlete’s circadian profile, macro logging from a photo of a plate, and a body-battery-style energy metric.

Summary

Reps first shipped on 10 October 2025 and has put out 28 released versions since, roughly weekly, from one developer. It holds 4.3 stars from 18 ratings on the App Store, 13 of them five-star, a small enough sample that reading the reviews beats reading the average.

It is not competing with a Garmin, and it cannot see one properly. The narrower pitch is the interesting one: for an athlete who already owns an Apple Watch, Reps turns what that watch measures into a morning verdict that edits the day’s session before it is opened, with thresholds, reasons, and proposed change all shown rather than hidden behind a score. No extra strap, no second hardware subscription.

Reps is on the App Store and at reps.siddharthnatamai.com.

Quick answers

Will it work with a Garmin, WHOOP or Oura?
Only via Apple Health, and only for what those devices write there: workouts, steps and active energy. HRV, resting heart rate, stress and Body Battery do not come across, and there is no direct API integration with any of them. The only direct vendor link is an optional outbound one to Strava.

Is an Apple Watch required?
For readiness, sleep and the training-adjustment loop, effectively yes. Strength logging, nutrition and the coach work without one.

Does it actually change the workout, or just advise?
It changes it. The intensity ceiling is applied to today’s entry in the weekly plan and travels with the generation request, so the session produced is already constrained. The athlete sees the proposed change with its reason and can decline it, and an explicit request for a specific session overrides the recommendation.

Related reading on the5krunner

• IntervalCoach: AI Training That Adapts to Your Recovery Each Day: the closest conceptual sibling. Another training app that adapts today’s workout from recovery signals, though it uses Intervals.icu data rather than Apple Health.
• Vitara: Apple Health Tells You What Happened, Vitara Shows What Changed: another privacy-first iPhone app operating on Apple Health data locally on the device.
• BeanFit: Strength and Cardio Tracking for Runners: another indie iPhone and Apple Watch app combining strength and cardio for endurance athletes.

Author: Siddharth Natamai, developer of Reps, edited by the5krunner.

Last Updated on 27 July 2026 by the5krunner