Reading a lactate curve: every method, and the coach's final call

A lactate curve does not have one true threshold, it has several defensible ones, because established methods each read a different real point on it: the segmented turning points, the maximal-curvature Dmax, the log-log aerobic break, a fixed lactate value, or a fixed rise above the athlete's own baseline. Labthlete now computes every method, lets you choose one for LT1 and another for LT2, shows them all side by side so you can see where they agree, and lets you drag the final line yourself. The algorithm proposes a range; the coach decides the number.

Give the same lactate step test to ten experienced coaches and you can get ten slightly different threshold numbers. That is not sloppiness, and it is not a bug in anyone’s software. It is because a lactate curve does not contain one threshold. It contains several real transition points, and the established methods each read a different one.

What the curve actually contains

As intensity climbs, blood lactate stays near baseline, then starts to rise, then at some point accelerates and takes off. Hidden in that shape are several genuine landmarks: the first departure from baseline (the aerobic threshold), the point where the slope changes again and accumulation accelerates (the anaerobic turning point), and the highest intensity you could actually hold in a steady state for half an hour. Each is a real physiological event. Each sits at a slightly different place on the curve.

So the honest question is not “which method is right and which are wrong”. It is “which landmark do you want to read, and which method reads it well”. Those have real answers.

Every method, and what it is good at

Labthlete now computes the whole family on every test and shows them side by side. Here is what each one is actually good for.

On a clean curve the anaerobic estimates land within a tight band of each other, and the segmented turning point, Dmax and OBLA will often agree to within a few watts. Where they spread apart, on a noisy curve, a short protocol, or an athlete with unusual kinetics, the spread itself is the useful signal: it tells you the number is uncertain and deserves a second look.

Four curves, four different winners

The best method is a property of the curve in front of you, not a fixed decree. Here are four real shapes a coach meets, and the method that reads each one best. The winning method’s pick is marked in petrol, the alternatives in grey.

When the curve has two genuine, well-separated break points, the segmented fit locks straight onto them, exactly where the eye sees the flexion. A very steep last sample tilts the first-to-last line Dmax leans on, pulling its pick a little low, so here the turning point is the truest read.

Some athletes, often the well-trained, bend smoothly with no crisp elbow. There is no sharp break for the segmented method to grab, so it turns ambiguous and can drift high. Dmax’s maximum-curvature construction does not need a break point; it stays steady and individual, and here it wins.

When the curve is a little scattered, or when what you care about is comparing the same athlete across a season or against lab norms, you want a number that means the same thing every time. A fixed lactate value read off the smooth fit does exactly that: it is stable and comparable even where the break-point methods wobble from test to test.

For the aerobic threshold the winner often changes again. When the baseline drifts and the first rise is gentle, the segmented first break can sit too high. Log-log works on the shape of the curve regardless of the absolute lactate scale, so it pins LT1 most cleanly here, with baseline-plus-0.5 a close, individualised second.

The coach has the final call

This is the part that matters most. A threshold is an estimate of a physiological state, and no algorithm has ever seen your athlete train, race, or fall apart in the last hour of a long day. You have. So Labthlete is built so the software proposes and the coach decides.

You pick which method to trust for LT1 and, separately, for LT2, because different landmarks suit different methods. You see every other method’s estimate next to your choice, so nothing is hidden. And if your eye says the real turning point is two watts to the right of every algorithm, you drag it there. That manual value, not the model’s, is what gets saved on the athlete and fed into the race predictor.

So how do you know the numbers are trustworthy?

Because the methods themselves are validated, not invented. We tested the whole family against a mechanistic model of lactate production and clearance, where the underlying physiology is known, and against the published maximal-lactate-steady-state literature and 835 real graded tests from an open dataset. That work confirmed something important and slightly humbling: the methods target genuinely different points (the segmented turning point reads the visible flexion; Dmax and Dickhuth’s IAT sit closer to the maximal steady state, which is a little lower), and each is sound within its own definition. That is why we surface all of them rather than hard-coding one and calling it the truth.

The quickest way to see it is on your own curve: import a test, switch methods live, and drag the line where your judgement puts it. Then read how those thresholds become a race prediction.