Every multiple on this site is a post divided by a baseline, so the baseline is doing at least half the work. It is worth being precise about what it is.
A baseline is one number that answers: what does a normal post from this account do? The whole question is which definition of “normal” survives contact with real data.
Why not the mean
The average is the obvious choice and the wrong one, because the thing we are trying to detect is precisely the thing that ruins an average. One breakout post pulls the mean up, and the inflated mean then makes that breakout look smaller relative to normal, while making every ordinary post look like an underperformance.
This is not a small effect in short-form. Distributions here are heavily skewed — a small number of posts do many multiples of the rest — which is exactly the condition under which means stop describing anything.
Using a mean to find outliers is using a measure that outliers break.
Why not the plain median
The median is robust against precisely that problem: the middle value does not move when the top value doubles. For most purposes it would be enough, and it is what we would use if we had to pick a one-word answer.
It has one weakness for this job. It is too stable. A median ignores not only the freak successes but also sustained, genuine changes at the top of an account's range — and it treats a dead post at the bottom, one that was suppressed or badly timed or never distributed at all, as equally informative as a normal one.
What trimming does
A trimmed median discards a slice from each end of the distribution and takes the middle of what remains. Cutting the top removes the breakouts, which is the point — a breakout should not be part of the definition of normal it is being measured against. Cutting the bottom removes posts that did not really get a chance, which would otherwise drag the definition of normal below anything the account actually does on a working day.
What is left is the account's ordinary operating range, which is the thing we want to compare against.
The window
The baseline is calculated over a rolling recent window rather than an account's whole history, for a reason that has nothing to do with statistics: accounts change. Audience size changes, posting cadence changes, the format changes, and the platform's distribution changes underneath all of it.
A baseline drawn from two years of history is measuring an account that no longer exists. A window that is too short, on the other hand, has too few posts to be stable — see how many posts a baseline needs.
What this does not fix
Two things, stated plainly.
It does not fix a step change. If an account genuinely levels up — a format finds its audience, a collaboration adds a large new following — the rolling baseline lags behind for a while, and posts in that period will read as larger multiples than they are. The baseline catches up; it does not catch up instantly.
It does not make small accounts comparable to large ones in absolute terms, and it is not trying to. A multiple says something happened relative to this account's normal. It says nothing about scale, which is the correct division of labour — see why the bigger number is usually the worse lesson.