Every endurance platform tracks training load. Most stop there. The number on your screen tells you how much you did yesterday. It says nothing about whether that dose was right for you, or what it will cost you on Thursday.
The model underneath that number is more than fifty years old. Eric Banister and colleagues built it in 1975, and it has outlived almost everything published since (Banister et al., 1975). We use it at Athletica, but not in its original form. The original answers one question: how much stress did this session impose? The questions that matter to an athlete sit downstream of that. How much can I absorb this week. Where is my fitness heading. When will I be fresh. Those need the load model and a response model running together, which is the part most platforms skip.
This is Chapter 8 and Chapter 9 of the HIIT Science book (Laursen & Buchheit, 2019). Load, then response to load. The two are inseparable, and treating them as one thing is where a lot of training software goes wrong.
What the Banister model actually says
Training load is a single number for the stress of a session. A short easy jog might score 30. A long hard ride might score 200 or more. Banister's insight was that you could model an athlete's performance as the balance between two opposing effects of that load: a positive, slow-building fitness effect, and a negative, fast-decaying fatigue effect. Train, and both rise. Stop, and fatigue clears quickly while fitness lingers. Performance is what is left when you subtract one from the other.
The model isn't perfect, and we have never pretended otherwise. But it has stood the test of time, and it is still used across most endurance sports in some form. At minimum it gives you two principles you cannot coach without: progressive overload and supercompensation. You want a load progressively greater than what you have handled before, because that is what drives the adaptation. Push too far past it and you break. The luge track is the analogy we like. Stay on the fast line and you fly. Catch the wall and the run is over.
Fitness, fatigue, and form
In practice the model surfaces as three numbers, and they are worth knowing because the whole field uses them.
Fitness (CTL) is your training load averaged over roughly the last six weeks. It rises slowly, and it is what a training block is trying to build.
Fatigue (ATL) is your load averaged over roughly the last week. It rises and falls fast.
Form (TSB) is fitness minus fatigue. Positive means fresh and ready, which is what you want on race day. Negative means tired, which is normal and necessary in the hard weeks of a block. Plotted over time these three are the Performance Management Chart, and there is nothing proprietary about them. They are standard Banister-derived math, the same family of calculations TrainingPeaks and most of the industry use.
We say that plainly because transparency is the point. The base layer is shared science. What you build on top of it is where platforms separate.
Why a load model on its own isn't enough
Here is the problem the raw model can't solve by itself. Two athletes can look identical on paper, get the same prescribed load, and respond completely differently to it. The load is an input. The response is individual, and the response is what you actually care about.
Chapter 9 of the book is about exactly this, the training load response, and Athletica reads it through three channels. Your heart rate response to exercise. Your neuromuscular response, from GPS, power, and accelerometry. And your own perception, the comments and RPE and how the session felt. A load model that ignores these channels is guessing. It assumes the dose landed the way the textbook said it would, which for any given athlete in any given week is often wrong.
This is the failure mode of one-size-fits-all plans. The structure is fine. The blindness to individual response is not.
How Athletica modifies the model
When we describe the system as running an Athletica-modified load model (the phrasing we used in SPSR241, Zignoli & Laursen, 2024), three things make up the modification.First, it is continuous, and it runs from history rather than from test days. The model updates from how you actually performed and recovered across weeks and months, not from a lab visit every quarter. Your accumulated training is the context, and a session can only be read against it. A power number that looks impressive in isolation may be unremarkable, or a warning sign, in the context of three months of data.
Second, load and response are integrated rather than bolted together. The fitness and fatigue projections feed the prescription, and the response signals, HRV chief among them (Plews et al., 2013), feed back into how aggressively the next block builds. This is the load-response loop of Chapter 9 made automatic.
Third, the prescription runs inside guardrails. The model holds a fitness target for each phase and will move a session's load to hit it, but never past a set of safety limits. Fatigue is not allowed to run away from fitness. Form is not allowed to fall below a floor, and that floor is shallower for beginners and deeper for advanced athletes, because they tolerate different depths of fatigue. No single day is allowed to spike beyond a ceiling scaled to your current fitness. And as a race approaches, the model tapers you into a freshness window matched to how important the race is.
One distinction matters, because it gets muddled. The load prescription is deterministic. It is a rule-based calculator with explicit, auditable rules, not a black box and not a trained model guessing at your training. The machine learning in the system sits upstream, in building your power and pace profiles and detecting your thresholds and zones from raw wearable data. Machine learning finds your profile. Deterministic math sets your load against it. A coach should be able to ask why a session was prescribed the way it was and get a real answer, and with this architecture they can.
Where this sits in the Athletica methodology
The load model is the floor the rest of the building stands on.
Workout Reserve, durability, critical power and pace, HRV-guided readiness, every concept in our methodology assumes a load model underneath it. This is that model. It is old, it is well validated, and the work is in how faithfully you implement the response side and how honestly you constrain the projection.
Explore the full Athletica Methodology HubFrequently Asked Questions
What is the Banister model of training load?
A systems model from Banister and colleagues (1975) that represents performance as the balance between a slow-building fitness effect and a fast-decaying fatigue effect of training load. Train and both rise; rest and fatigue clears faster than fitness, leaving a performance gain. It remains the basis for training load tracking across most endurance platforms.
What is the difference between CTL, ATL, and TSB?
CTL (fitness) is training load averaged over about six weeks. ATL (fatigue) is load averaged over about one week. TSB (form) is CTL minus ATL: positive when you are fresh, negative when you are carrying fatigue. Plotted over time these three form the Performance Management Chart.
Is Athletica's training load model the same as TrainingPeaks?
The base layer is the same Banister-derived math the industry shares. Athletica modifies it by running it continuously from accumulated history, integrating individual load-response signals (heart rate, neuromuscular, and perceptual), and constraining the prescription with experience- and sport-dependent safety guardrails.
Does Athletica use AI or machine learning to set my training load?
The load prescription itself is deterministic, a rule-based calculator with auditable rules, not a trained model. Machine learning is used upstream to build power and pace profiles and detect thresholds and zones from wearable data. Machine learning finds your profile; deterministic math sets your load against it.
Why do two athletes on the same plan progress differently?
Because load is an input and the response to it is individual. The same prescribed load lands differently on different athletes. Athletica tracks each athlete's response through heart rate, neuromuscular, and perceptual channels (Chapter 9, Laursen & Buchheit, 2019) and adjusts the prescription accordingly.
What stops an Athletica plan from overtraining me?
A set of guardrails that override the fitness target when needed. Fatigue cannot run away from fitness, form cannot drop below an experience-dependent floor, no single day can spike beyond a fitness-scaled ceiling, and races are preceded by a taper into an appropriate freshness window.
What stops an Athletica plan from overtraining me?
A set of guardrails that override the fitness target when needed. Fatigue cannot run away from fitness, form cannot drop below an experience-dependent floor, no single day can spike beyond a fitness-scaled ceiling, and races are preceded by a taper into an appropriate freshness window.
Updated: 06/19/2026
