Regression to the mean is a statement about your baseline

What the phrase actually claims

Regression to the mean says that when an observation is extreme, the next observation from the same source will probably be less extreme. The reason is arithmetic rather than physical, since an extreme reading usually combines a genuine underlying level with an unusually favourable set of circumstances. Those circumstances are not persistent, so the next reading contains a fresh and probably less favourable draw of the same randomness.

Nothing is pulling the team back towards average, and no mechanism is punishing them for having done well. The phrase is therefore a statement about how measurements behave and not a description of anything happening to the competitors.

Noisiness determines how much to expect

The amount of expected regression depends on the ratio between genuine variation in ability and random variation in the measurement itself. A measure dominated by randomness regresses heavily, because almost all of any extreme reading was noise in the first place. A measure closely tied to repeatable skill regresses little, since an extreme reading there mostly reflects a genuinely extreme underlying level.

This means the same word describes wildly different expectations depending on which quantity is being discussed, and the distinction is usually omitted. Anyone invoking regression without saying how noisy the measure is has not actually made a quantitative claim at all.

Sample size interacts with everything

Small samples produce extreme readings far more often, so early-season figures contain a much larger proportion of noise than late-season ones. That is why the most dramatic apparent trends appear when the least evidence exists, and why they so often dissolve as evidence accumulates. Accumulation does not correct a figure so much as dilute the influence of any individual observation within it.

Waiting for more evidence is therefore the cheapest available remedy, and it is precisely the remedy that a weekly publishing cycle cannot afford. Much of what looks like poor analysis is really the consequence of having to say something before the evidence exists to say it.

The baseline can be wrong too

Regression assumes there is a stable underlying level to regress towards, and that assumption fails when a team has genuinely changed. A side that has altered its structure or personnel has a new underlying level, and the old baseline is no longer the right reference point. Distinguishing a real shift from a noisy reading is the hard problem, and no amount of appeal to regression resolves it.

The practical approach is to look for a mechanism, since a genuine change usually has a visible cause that noise does not. Where no mechanism can be identified, treating the extreme reading as noise is the safer default, though it is a default rather than a proof.

Using it honestly in a preview

Invoking regression responsibly means naming the measure, describing how noisy it is, and stating what level the estimate is regressing towards. It also means accepting that the correction applies to your own confident readings as much as to the ones you are criticising. The concept is most valuable as a discipline on enthusiasm rather than as a weapon against a particular team's supporters.

Used that way, it produces previews that are noticeably less exciting and considerably more likely to survive the following month. Used as a rhetorical device, it becomes a way of dismissing evidence selectively while sounding technical about it.