Five Points of Learning Rate Turn Fifty Repetitions Into Four Hundred

The cumulative total is what a schedule needs

Thirty attempts at this rate add up to 341.8 minutes rather than the 600 that thirty times the first attempt would suggest — learning saves 258 minutes over the batch, and that is the figure to quote when someone asks how long a run of thirty will take. Quoting the final attempt's 9 minutes instead promises a pace nobody reaches until the end, which is how a plan built from the best observed time runs late from the first day. Estimate the batch from the cumulative curve, and the individual attempt only when scheduling the last one.

Where the model stops describing anything

Taken literally, the curve says the time keeps falling forever and eventually reaches zero, which nothing does. Real tasks hit a floor set by the physical minimum — how fast a machine cycles, how long a weld has to cool, how quickly a hand can move — and improvement stops there no matter how many repetitions follow. The model is also silent on forgetting: a gap of six months between attempts undoes part of the gain, and a curve fitted across a long interruption will fit badly. Use it for a continuous run of similar work, and refit it whenever the method, the tooling or the person changes, because each of those starts a new curve rather than continuing the old one.

Frequently asked questions

How do I find my own learning rate?

Time the first attempt and a later one whose number is a power of two — the second, the fourth, the eighth — and take the ratio, then the appropriate root. If attempt 8 takes 12.28 minutes against 20 for the first, the total reduction is 0.614 across three doublings, and the cube root of 0.614 is 0.85. Two data points give a rough answer and four give a usable one; below about eight attempts the noise between individual tries is larger than the effect being measured.

Does the same curve describe a person and a factory?

The shape does, and the mechanism does not. For an individual the gain comes from motor skill, familiarity and removing hesitation; for an organisation it comes from better tooling, better sequencing, and the fact that a hundred people each learn a little. Organisational curves usually have lower rates — improvement per doubling is larger — because there are more things left to fix, and they can be reset by turnover in a way an individual curve is not. Fit them separately even when both are running at once, or the personal gain gets credited to the process and the process stops being improved.