How Adaptive Training Actually Works: TSS, CTL, ATL Explained

Every adaptive triathlon app talks about "adapting to you." Almost none of them explain what that actually means in terms you could check for yourself. Underneath the marketing language, there are three numbers doing the real work: TSS, CTL, and ATL. They are not new. They come from sports science and cycling coaching that predates any training app on the market, and once you understand what each one tracks, you can tell the difference between a plan that's genuinely adapting and one that's just reshuffling a template.

This is a mechanism explainer, not a pitch. The goal is to leave you able to read your own numbers and understand what a plan is doing when it changes your week, whether or not you ever use Triathlify to do it.

The raw material: what TSS actually scores

TSS (Training Stress Score) is a single number attached to one workout. It combines how long a session lasted with how hard it was, so a short, brutal interval session and a long, easy ride can land on a similar score for very different reasons. An hour ridden at your functional threshold power scores exactly 100 TSS by definition. Running and swimming use pace-based or heart-rate-based versions of the same idea, usually written rTSS, sTSS, or hrTSS, since power meters aren't standard equipment in either sport.

A single day's TSS tells you almost nothing on its own. A 150-TSS day could be a huge outlier for a beginner or a routine Tuesday for someone deep into an Ironman build. What TSS is actually for is being summed and averaged over time, which is where CTL and ATL come in.

The slow number: CTL, your accumulated load

CTL (Chronic Training Load) is a rolling, weighted average of your daily TSS, usually calculated with roughly a 42-day time constant. Recent training counts more than training from a month ago, but nothing falls off a cliff overnight. It's the closest thing endurance sport has to a single "fitness" number, and it's the slow-moving line on a training chart.

A few things worth being precise about, because CTL gets misread constantly:

• CTL is derived entirely from logged training. Nothing about it is measured from your body directly. It's bookkeeping, not a physiological test.
• A rising CTL usually tracks with rising aerobic fitness, because sustained training load and fitness gains are correlated, but the relationship isn't perfect.
• Two athletes with identical CTL can have different real fitness if the composition of their training differed, for example all easy volume against a mix that included real intensity.
• The number itself matters less than the rate of change. Climbing CTL aggressively, several points a week instead of gradually, is one of the more consistent risk factors for injury and non-functional overreaching in the training-load literature. Ramping CTL as fast as mathematically possible is optimizing for the wrong thing.

The rate-of-change point isn't just a caution from one glossary entry. Sports scientist Tim Gabbett's 2016 paper in the British Journal of Sports Medicine, built on his work with elite team-sport athletes, popularized the acute-to-chronic workload ratio (comparing a short recent window of training load against the longer-term average) as a rough injury signal, and found that athletes with a well-built chronic base tolerate hard acute blocks better, not worse. It's not a precise dial you can read to the decimal point for any one person, and later researchers have pushed back on how tightly a specific ratio should be trusted. But the direction of the finding lines up with what coaches were already doing by hand long before software existed: build the base gradually, then load on top of it, rather than spiking volume from a standing start.

The fast number: ATL, your current fatigue

ATL (Acute Training Load) is CTL's short-term counterpart. It's the same kind of rolling, weighted average of daily TSS, but over roughly a week instead of roughly six. Where CTL moves slowly and represents accumulated fitness, ATL moves fast. It climbs quickly during a hard block and drops quickly once volume or intensity backs off.

Subtract ATL from CTL and you get a number sometimes called Training Stress Balance, or informally "form":

• A strongly negative number means ATL is running well above CTL. Recent training has outpaced what your body has adapted to yet. That shows up as heavy legs, flat power, a session that should feel easy but doesn't.
• This isn't automatically bad. Some accumulated fatigue is the normal cost of a training block doing its job. The problem is when it never gets a chance to resolve.
• A near-zero or positive number usually means you're fresh, which is the state you want walking into a race, not the state you want to live in every week of a build phase.

The gap between these two numbers is the entire mechanical justification for a taper, and it's also the main thing an adaptive plan is actually watching week to week.

What happens when a plan actually adapts

This is the part most explanations skip. Here's roughly what changes inside a system that recalculates around TSS, CTL, and ATL rather than a fixed calendar.

On a normal week, the plan compares each completed session against what was planned: did it happen, how hard was it relative to the target, did the athlete report anything unusual. That comparison updates the athlete's current CTL and ATL, which become the starting point for building the sessions still ahead.

When a session is missed or shortened, the training load for that period is lower than the plan assumed. A plan that just keeps stacking the original template on top of that gap is quietly setting up the exact kind of sharp acute-to-chronic load jump the Gabbett research above warns about, because the next planned increase in volume now lands on a lower base than it was designed for. A plan that adapts instead recalculates the upcoming block from where CTL and ATL actually are, not from where the calendar assumed they'd be.

When a session goes harder than planned (a race simulation that turns into an actual race, an interval session where everything clicked), ATL rises more than expected. The near-term plan should account for that extra fatigue rather than ignoring it and stacking another hard day on top.

When an athlete requests an easier week, the plan can lighten volume, but it has to stay inside the coaching floor for that goal. A plan is not supposed to fall below the minimum training days a specific goal and timeframe requires, because a plan built on too little structure isn't a lighter version of a real plan, it's a plan that can't get the athlete to the start line prepared. The same logic applies to weekly hours: below a certain floor, the plan should say so and offer a way back to a schedule that works, rather than silently generating something that can't succeed.

Across a full training block, the shape of CTL should still rise toward the goal in controlled steps, following the coaching rules for that race distance and timeframe, phase by phase, rather than being pushed as high as possible as fast as possible.

None of this requires a human to manually redraw the calendar. It requires a system that treats TSS as the raw input, CTL and ATL as the state the athlete is actually in right now, and a set of coaching rules, day minimums, hour bands, phase structure, taper shape, as the constraints that state has to respect. That combination is what "adaptive" is supposed to mean. A plan that just moves a missed Tuesday session to Wednesday without touching anything else in the week isn't adapting in this sense. It's rescheduling.

Where a taper fits into this picture

A taper is the clearest example of CTL and ATL being used on purpose rather than just observed. A taper works by deliberately cutting training volume so ATL falls faster than CTL does. CTL decays slowly by design, so a well-timed taper lets fatigue clear while fitness stays close to intact.

That's why a real taper cuts volume and keeps intensity. In a well-built plan, a taper week typically lands somewhere around 30 to 55 percent of the peak week's training load, while still including some threshold or race-pace work rather than dropping to easy volume only. Cutting the intensity along with the volume would blunt the neuromuscular sharpness a race actually asks for, without meaningfully speeding up how quickly ATL clears, since the fatigue reduction comes mostly from the drop in total load, not specifically from removing hard efforts. A near-zero final block isn't a taper. It's detraining, and it shows up on race day as an athlete who feels rested but has lost the sharpness they built.

What the numbers can't tell a plan

TSS, CTL, and ATL are all derived from training data. None of them come from a direct measurement of your body. That has real limits worth naming plainly:

• They can't tell the difference between training fatigue and illness. A high ATL from a hard week and a high ATL from fighting off a cold can look identical in the data. An honest plan treats this as a gap to ask the athlete about, not something the numbers alone resolve.
• They can't see injury risk directly. They can flag a load pattern that correlates with elevated risk in population-level research, which is a very different thing from predicting what will happen to one specific athlete.
• They can't replace an athlete's own read on how they feel. A session that scores as easy on paper can feel hard on a day with bad sleep or outside stress. The numbers are an input to a decision, not the whole decision.
• They say nothing about technique, pacing skill, or race-day execution, which matter just as much as fitness and freshness on the day itself.

An adaptive plan that's honest about this treats TSS, CTL, and ATL as the load-management layer of coaching, not the entire job. The rest, goal-setting, technique work, race strategy, still needs the same judgment it always did.

A short history, because none of this is new software

The math behind CTL and ATL is older than any training app currently on the market. Exercise scientist Eric Banister proposed an impulse-response model of fitness and fatigue in 1975, describing performance as the difference between a slow-building fitness effect and a fast-decaying fatigue effect from training. Cyclist and exercise physiologist Andrew Coggan built a simplified, power-based version of that idea in the late 1990s, which became TSS and the Performance Manager chart later popularized by TrainingPeaks. What's changed since then isn't the underlying model. It's how much of the recalculation an app can now do automatically, and how quickly it can respond to what actually happened instead of waiting for a coach to notice and manually adjust the next block.

Common ways athletes misread their own numbers

A few patterns show up often enough to call out on their own:

1. Treating a single day's ATL spike as an emergency. A hard week is supposed to raise ATL. Panic-resting mid-block because of one heavy day often does more harm to the plan's structure than the fatigue itself would have.
2. Chasing the highest possible CTL. A bigger number isn't automatically better if it got there by climbing too fast for the athlete's actual recovery capacity.
3. Comparing raw TSS between different athletes. Two athletes' TSS numbers aren't comparable to each other, only to their own history.
4. Reading CTL without ATL, or ATL without CTL. Either number alone misses the tension a taper exists to resolve. They're meant to be read together.
5. Assuming a missed week just gets glued back onto the plan unchanged. That's the exact pattern that produces an unplanned load spike right when the plan should be recovering ground carefully instead.

Understanding this is also what makes it possible to judge whether a specific tool's "AI adapts your plan" claim means anything. See the training methodology page for how these mechanics are applied across an entire plan, or current plans if you're ready to see it running against your own training history.

Frequently asked questions