Progress Without Going Harder: The One-Variable Rule

The context

Change one thing, hold the rest, and let the trend tell you whether it worked. This is a good rule and it is worth keeping. The number usually attached to it — wait two weeks before you read anything — is not one anyone chose deliberately, and no study establishes it. It arrived at some point by accretion, it is repeated in enough places to look settled, and it does not survive being traced.

Three things are usually offered as its basis, and each fails in a different way. Individual response variation is real and is genuinely why one session decides nothing — but it is a statement about how much people differ, not about how long to wait. Day-to-day fluctuation in a physiological marker is also real, and it is the reason monitoring works from a rolling trend rather than a single morning's reading — but it still says nothing about the length of the window. And the argument that a rolling load calculation "has to be allowed to settle" comes from a paper whose published conclusion is that the ratio should be abandoned as a framework, since it conferred no meaningful predictive advantage over a model that predicted nothing at all [7]. A window invented from those three is not a finding. It is a number that outlived its reasons.

This matters practically, because the two-week figure is not one a rower can check for themselves.

The core idea

The problem one-variable discipline solves is attribution, not motivation. A session has many inputs, and the number that comes off the monitor at the end is the joint product of all of them. Change two inputs and the output moves, but the movement carries no information about which input did the work. Change one and hold the others, and the same output becomes evidence.

This is not a coaching preference that happens to agree with the evidence. It is the condition the evidence was generated under. The dose-response curve underpinning essentially all exercise prescription is built from studies that varied one component at a time, which is why the ACSM position stand on exercise quantity and quality can prescribe incremental progression at all [5]. Take that design out and the curve you are being prescribed would not exist in readable form.

The reason a two-dial change is worse than merely slow is that the interaction is real, not hypothetical. Intensity, frequency and duration do not add independently, and the pooled evidence is specific about how: above roughly 35 minutes of work, a lower intensity produces much the same aerobic effect as a higher intensity over a shorter session [6]. So "I added two minutes and also went out a little harder" is not a session with two small effects in it. It is a session whose outcome depends on a trade-off you did not hold constant, and no amount of logging will separate them afterwards.

What replaces the two-week figure is smaller, more useful and not a constant: it is a test you can actually run. The monitoring literature's own instruction is that a change be judged against the error of measurement and the smallest important change of the measure being read [1] — a test on magnitude, applied to one specific measure, which is why near-daily resting readings are taken at all. There is no constant number of days in it, because the answer depends on how noisy the thing you are watching happens to be.

Your own noise has a number in it

This is the part most guidance leaves implicit, and it does not have to be. The Concept2 is a measuring instrument with a published standard error, and it depends on the distance you test over. In a reliability study of national-standard rowers, the percentage standard error of measurement in mean power over 2,000 m on a Concept2 was about 1.3%, and over 500 m it was about 2.8% — with the confidence limits on those figures running from roughly 0.9% to 2.9%, and 2.3% to 3.4% respectively [8]. The same rowers on a different machine produced a wider spread again.

That is the whole argument in one number. Reading a 2,000 m test, your instrument contributes about 1.3% of noise to mean power before you have done anything wrong; reading 500 m pieces, about 2.8% [8]. Any window has to be long enough that the change you made is larger than that. A two-minute addition to a 30-minute steady row is a small percentage of total load, and that error is a large fraction of the effect you are hoping to see.

So set the window from the measure, not the calendar. Ask how big a change would have to be to be worth reading at all, how noisy your own readings of that thing are, and then take as many readings as it takes for the change to clear the noise. On a daily morning reading that might be a handful of days. On a threshold or a race piece it is weeks. Both answers are correct, and neither is fourteen days.

The two trials that actually closed the loop

If you want to see what a defensible monitoring window looks like in practice, two trials are worth reading, and they disagree about the length. That disagreement is the point.

The shorter one gave 26 moderately fit men a daily heart-rate-variability reading each morning and prescribed the day's intensity from it, over a four-week block of 40-minute sessions. The rule was strikingly short: if the morning HRV fell below a reference value derived from the previous 10-day mean and standard deviation, or showed a decreasing trend for two days, the session dropped to low intensity or rest [3]. Two days is the decision window; ten days is the rolling baseline it was read against. Running velocity improved more in the HRV-guided group than in the group following a fixed programme.

The longer one ran 40 recreational endurance runners through a four-week preparation period and then an eight-week intensive block, and dosed the hard sessions off an individually determined smallest worthwhile change in HRV. The guided group took significantly fewer moderate and hard sessions than the fixed group — 13.2 against 17.7 — and improved its 3,000 m time by 2.1% where the fixed group's 1.1% fell short of significance, for a between-group effect size of 0.42. Maximum oxygen uptake improved in both groups [4].

An eight-week block and a two-day trigger are both defensible, because they are answering different questions: one is deciding when to schedule a hard session, the other is asking whether a change has taken hold. Notice also what the second trial did not find cleanly — the effect on running performance was modest, and VO2max rose regardless of how the sessions were timed. "Strongest evidence in the literature" is not what this is, and a coach who tells you otherwise is overselling a 0.42 effect size.

Quantity first, and why pace goes last

The order of the dials is not arbitrary, and the reason is where the dose-response slope is most trustworthy. Accumulated moderate volume has a shallow, reliable gradient, and the position stands are explicit that progression works by incrementing what you already tolerate [5]. That makes duration the safest place to start: the failure mode of getting it wrong is that nothing much happens.

Intensity is different in kind rather than degree. It is the dial where the dose-response curve steepens, which is exactly why it produces faster results and why its errors are harder to read back. It is also the dial with a physiological ceiling and the one where the room you are in can turn a small programming mistake into a safety event. Rowers sensibly reach for pace last.

The fitness-fatigue intuition behind that ordering is broadly right and frequently over-claimed: a hard session leaves a debt that is not paid off overnight, but the arithmetic people build on top of it deserves less confidence than the shape does.

The shape still earns one very useful conclusion, which is that subtracting is well supported. The tapering meta-analysis across 27 studies found the most efficient strategy to be a two-week taper in which training volume falls by 41 to 60%, with an overall performance effect of 0.59 ± 0.33 [10]. That is one of the stronger results in this literature, and it points the same way as everything else here: where the evidence will not tell you the right dose, taking less is the move with evidence behind it.

Note that the two-week figure in that finding belongs to a taper, not to a monitoring window — an easy number to misfile, and the sort that turns into folklore.

Power is a separate dial, not a harder version of the same one

Strength and power work transfer to erg performance in a way that steady aerobic volume does not, because they adapt something different. Over four weeks of strength training, researchers tracking individual motor units in the ankle dorsiflexor found discharge rates rose by 3.3 ± 2.5 per second with recruitment thresholds falling, in fourteen trained lifters against fourteen controls [9]. The force increase was traceable to increased motor neuron output, not to the muscle itself.

That is a neural adaptation with its own timeline, and it is why adding a sprint session mid-block is a genuinely new dial rather than a bigger dose of an existing one. It is also why the week a hard power piece lands is the week nothing else should move — and why the answer is usually not that week.

The trade-offs

Two honest limits, and both are worth sitting with.

The first is that the rule does not make the change work. Individual response to a standardised programme varies widely: it is more common for individuals to show a wide spread of responses than for the group average to describe a typical one, and part of that spread is measurement error rather than biology [2]. A fast responder and a slow responder can look like a success and a failure at the same moment on the same programme. The rule makes that difference legible over time; it does not remove it, and anyone who tells you a disciplined approach guarantees adaptation is selling something.

The second is that your read is incomplete even at its best. Heart-rate measures cannot speak to all aspects of wellness, fatigue and performance, which is why they are paired with daily training logs and simple performance tests rather than trusted alone [1]. Sleep, food and heat all move the same channels your training moves, and they move them without telling you. That is not a reason to abandon the read; it is the reason a flat trend is a reason to look at your sleep before you add a second dial.

What to do tomorrow

Pick one dial and write the other three down as constants — including stroke rate, flywheel setting and ergometer model, because a plan reconstructed from memory halfway through is not the plan you actually ran. Make the change small enough that you would expect to be able to see it: a minute or two of steady duration, a single repetition, a second of rest.

Then set your window from the measure rather than the calendar: take enough readings that your change is clearly larger than your machine's own noise. Log heart rate, perceived exertion and completed duration together, because any one of them can move for reasons that have nothing to do with the dial you turned.

And when the trend is flat, the correct next move is to extend, not to add. Extend until you reach a verdict, and only then decide whether the dial has run its course. Where the evidence cannot tell you the right dose, subtract.