Training load modeling compares a slower average of recorded load with a faster one. The classic model calls the slower average CTL (Chronic Training Load), the faster average ATL (Acute Training Load), and their difference TSB (Training Stress Balance). The conventional labels "fitness," "fatigue," and "form" are historical shorthand. None of these values directly measures physiological fitness, fatigue, recovery, feelings, performance, or injury risk.

The Core Idea

The model runs the same daily recorded-load value through two weighted averages. One uses a longer time constant and changes gradually. The other uses a shorter time constant and responds more quickly. Subtracting the faster average from the slower one shows how recent recorded load relates to the longer recorded pattern.

This faster-versus-slower structure comes from the impulse-response model first described by Banister in 1975 and later adapted for practical training analysis by Dr. Andrew Coggan through TrainingPeaks. Its historical language describes modeled responses to training. The values remain mathematical summaries of the input load data.

Think of the two averages as traces drawn with different pens:

CTL is the slower trace. Each recorded day changes it a little.
ATL is the faster trace. Recent recorded days change it more.
TSB is the vertical gap between those traces. It describes their relationship, not how someone feels or will perform.

The Three Numbers

CTL — the slower average

CTL is an exponentially weighted average of daily training stress with a time constant of approximately 42 days. Training platforms conventionally call it "fitness," but the value is a slower summary of the recorded load supplied to the model.

Goes up when: You train consistently over weeks and months
Goes down when: You take extended time off or significantly reduce training
Changes slowly: A single hard workout barely moves CTL. A single rest day barely dents it.

A higher CTL means the slower average contains more recorded training stress. It does not prove that someone is fitter, can tolerate more training, or will perform better. Values also depend on the load method and data source, so comparisons between people are limited.

ATL — the faster average

ATL is an exponentially weighted average of daily training stress with a time constant of approximately 7 days. Training platforms conventionally call it "fatigue," but it is a faster summary of recorded load, not a measurement of tiredness or physiological fatigue.

Goes up when: You train hard in the past few days
Goes down when: You rest or do easy sessions
Changes quickly: One hard session noticeably raises ATL. Two rest days noticeably lower it.

ATL is not inherently bad. It simply gives recent recorded training more influence than older training. A high value does not by itself diagnose overreaching or overtraining.

TSB — the relationship between the averages

TSB is simply CTL minus ATL. Training platforms may label it "form" or "freshness," but it only describes the relationship between the two modeled averages. It cannot tell how you will feel or perform.

TSB = CTL - ATL (slower average minus faster average)

TSB Value	Model relationship	Cautious interpretation
Positive (above 0)	CTL exceeds ATL	Recent load is below the slower average
Slightly negative (-10 to 0)	ATL slightly exceeds CTL	Recent load is slightly above the slower average
Moderately negative (-10 to -30)	ATL exceeds CTL by more	Recent load is further above the slower average
Very negative (below -30)	ATL far exceeds CTL	A large modeled difference that needs context

Coaches often use this relationship as one input when reviewing training blocks and tapers. A negative TSB can occur when recent training rises; reducing recorded load lets ATL fall faster than CTL. The number still needs context from the athlete, the sport, and the quality of the source data.

A Simple Example

Let's follow a recreational runner named Alex over four weeks. We'll simplify the math and use arbitrary "training stress" units.

Week 1: Easy start

Alex runs three times: Tuesday (stress: 50), Thursday (stress: 40), Saturday (stress: 60).

ATL rises to roughly 21 (those three sessions averaged over 7 days)
CTL barely moves — three sessions in one week is a drop in a 42-day bucket
TSB is slightly negative: the faster average is a little higher than the slower one

Week 2: Higher recorded load

Alex adds a fourth run and increases intensity. Weekly stress: 220 (up from 150).

ATL jumps to roughly 31
CTL starts climbing — now there are two consistent weeks of data
TSB drops further — the faster average is moving above the slower one

Week 3: Highest recorded-load week

Alex's biggest week yet. Five runs, one long run, one interval session. Weekly stress: 300.

ATL peaks around 43
CTL continues climbing but more slowly — still reflecting the rolling 42-day average
TSB is solidly negative — the gap between the faster and slower averages is larger

Week 4: Lighter recorded load

Alex drops to three easy runs. Weekly stress: 100.

ATL drops quickly — from 43 down to roughly 20 in one week
CTL barely drops — it accumulated over 3 hard weeks and only lost 1 easy week
TSB goes positive — the faster average has fallen below the slower one

This is the mathematical pattern often seen during a taper. The faster average responds more quickly to the lighter week, while the slower average changes less. Coaches may interpret that pattern alongside the athlete's goals and direct feedback, but the model alone does not establish someone's physical condition or predict performance.

The Math (Optional but Useful)

If you want to understand the actual calculations:

Daily Training Stress varies by sport. In cycling, it's TSS (Training Stress Score) based on power output. In running, it's rTSS based on pace. For general heart rate-based training, it's TRIMP (Training Impulse) based on heart rate duration and intensity.

CTL = Yesterday's CTL + (Today's Stress - Yesterday's CTL) / 42

ATL = Yesterday's ATL + (Today's Stress - Yesterday's ATL) / 7

These are exponentially weighted moving averages (EWMA). The "42" and "7" are time constants — they determine how quickly each metric responds to new recorded load. A longer time constant produces a slower, steadier average. A shorter time constant produces a faster, more responsive average.

On rest days, training stress is 0, so:

CTL decays slightly toward 0
ATL decays more aggressively toward 0
TSB rises because ATL usually falls faster than CTL

On hard training days, training stress is high, so:

CTL nudges up slightly
ATL jumps up
TSB drops because ATL usually rises faster than CTL

The difference in response speed is what makes the two recorded-load patterns useful to compare.

Where CTL/ATL Came From

The model historically known as the fitness-fatigue model originated in exercise science research by Banister et al. in 1975. It was an academic concept for decades until Dr. Andrew Coggan adapted it for practical use in cycling through TrainingPeaks in the early 2000s. Power meter data made it possible to calculate a consistent training-stress input from recorded rides rather than laboratory measurements.

TrainingPeaks popularized the Performance Management Chart (PMC), which graphs CTL, ATL, and TSB over time. It became the standard tool for endurance coaches planning training blocks and tapers. If you've heard a cycling or triathlon coach talk about "building CTL" or "managing TSB," this is what they mean.

For over two decades, CTL/ATL modeling has been almost exclusively an endurance sports tool. Cyclists, runners, triathletes, and swimmers use it routinely. Strength athletes and hybrid athletes do not — because the standard model has no way to quantify a squat session in the same "training stress" units as a bike ride.

The Problem: Strength Training Doesn't Fit the Classic Model

Traditional CTL/ATL modeling uses heart rate or power output to estimate training stress. Those continuous signals are well suited to recorded endurance sessions.

Strength training breaks this model:

Heart rate does not describe the lifting work. During a set of squats, heart rate may rise, then fall during the rest between sets. A session average does not contain the exercise, weight, repetitions, set type, or effort recorded in a strength log.

Power output isn't measured. Cyclists have power meters. Runners have pace-based estimates. There's no standard "power meter" for a barbell squat. You can calculate work (weight x distance x reps), but translating that into a stress score comparable to cycling TSS requires a different approach.

One aggregate cannot identify muscle-group or tissue recovery. CTL and ATL each reduce the supplied load history to one value. That aggregate does not show which exercises were performed, which muscle groups were involved, or whether any tissue has recovered.

This is why strength work needs its own recorded-load estimate before it can join an aggregate model. A generic session value can enter the faster and slower averages, but the resulting values still cannot locate muscle-group load or determine tissue recovery.

How Incredible Uses Recorded Workout Load

Incredible uses one shared source of recorded workout load across its training metrics. Training Load combines recorded cardio and strength workout load. Movement outside a recorded workout remains separate.

Here's how the pieces connect:

1. Training Load starts with recorded workouts. Cardio load uses workout duration and heart-rate intensity when that data is available. Strength load uses logged set details such as weight, repetitions, set type, and effort, then compares the session with recent non-zero strength sessions. The two estimates are added as raw daily workout load. Incredible converts that raw total through a curve to display Load from 0 to 100. The displayed number is not a percentage or a physiological maximum.

2. Recent training is the faster average. Incredible keeps a weighted average of raw recorded workout load that changes at a seven-day rate. Recent days have more influence, but it is not a simple seven-calendar-day window. The app displays this as 7-Day Load.

3. Fitness is the slower average. A second weighted average changes at a 42-day rate. The app calls the raw value the Fitness base and converts it to displayed Fitness from 0 to 100. Fitness is training built over time from recorded workouts. It is not complete physiological fitness or everyday capacity.

4. Training Balance compares the averages. Training Balance divides the recent-training average by the Fitness base, with a minimum base for new or sparse histories. It describes the relationship between recent recorded training and the longer recorded pattern. It does not predict injury or overtraining.

5. Daily context brings the signals together. Incredible places how you slept and your overnight readings alongside recent training and workouts recorded today. Your optional check-in adds personal context. The view brings together Sleep and overnight readings, Recent training, Today’s recorded workouts, and Your check in.

Movement and daytime Stress remain separate from this context. Blood Oxygen can be displayed as a separate vital. None of these views proves recovery, predicts injury or overtraining, or prescribes a workout.

Why This Matters for Non-Athletes

You don't need to be a competitive athlete for training load modeling to be useful. The concepts apply to anyone who exercises regularly:

Comparing short and long patterns. A faster average responds quickly when recorded workout load rises or falls. A slower average changes more gradually. Their relationship makes a change in the training pattern visible without claiming to diagnose fatigue, injury, or overtraining.

Understanding recorded history. A flat slower average means the modeled recorded-workout history has been stable. It does not prove that physiological fitness has plateaued.

Seeing lighter periods. When recorded load falls, the faster average usually falls before the slower one. The model shows that mathematical change without predicting how the person will feel or perform.

Returning from breaks. Complete days without recorded workout load make both averages decline, with the faster one declining sooner. That describes the training record; it does not determine a safe or appropriate return plan.

Common Misconceptions

"Higher CTL is always better." No. A higher value means more modeled training stress has accumulated. It is not a health grade, safety threshold, or leaderboard.

"Negative TSB means I'm overtraining." No. It means the faster average is above the slower one. The ratio alone cannot diagnose overtraining.

"TSB tells me what I should do today." No. It describes recorded load. A training decision also depends on goals, symptoms, pain, professional advice, and information the model does not contain.

"A zero-load day is missing data." Not always. A complete day with no qualifying recorded workout contributes zero. Missing, stale, partial, or still-refining workout coverage is unavailable, not zero.

Getting Started

If you've never thought about training load modeling, here's a practical starting point:

Track your training consistently. The model needs data. Log your workouts — cardio via Apple Watch, strength via a training tracker. Gaps in data create gaps in the model.
Watch the patterns, not one daily number. A longer view shows how the faster recorded-load average has moved relative to the slower one. It does not determine whether physiological fitness is building, holding steady, or declining.
Treat the output as context. Training-load models describe recorded data. They do not prescribe a workout, establish safety, or replace how you feel and advice from a qualified professional.
Don't chase numbers. CTL is not a leaderboard. Its value depends on the input method, data coverage, and the person's recorded history, so a higher number is not automatically better.

Incredible calculates its shared training state from recorded Apple Watch cardio workouts and logged strength sessions. You do not need to understand the math to read Load, 7-Day Load, Fitness, Training Balance, or the daily context from sleep, vitals, and recorded training.

The Bottom Line

CTL/ATL modeling has been part of endurance coaching for two decades because it makes faster and slower recorded-training patterns visible. Applying that idea to strength and hybrid training requires a load estimate for logged gym work rather than treating workout heart rate as the whole session. Incredible combines recorded cardio and strength workout load, then reuses that shared state across its training metrics.

You do not need to become a sports scientist. The useful idea is simple: one recorded-load average changes quickly and another changes slowly. Their relationship provides context, while sleep, overnight readings, and today's recorded workouts add to the daily training picture. Your optional check-in adds personal context without claiming to predict performance, injury, or overtraining.