A baseline is only meaningful if every number inside it was produced the same way. Platforms break that assumption regularly, and they almost never restate history when they do — the April 2025 views change is the clearest recent example.
The damage is quiet. Nothing errors. Your dashboard keeps producing a figure. It is simply a blend of two definitions, and every multiple measured against it is wrong by an amount you cannot see.
Finding the break
You do not need to be told a change happened. It is visible in your own data if you look for it. Line up your posts in date order and look for a step — not a gradual drift, a level shift where the typical value before and the typical value after are plainly different, with no change in what you were making.
A real step change has three signatures worth checking together: it happens on a specific date rather than over weeks, it affects posts you consider ordinary rather than just the outliers, and it does not correspond to anything you did.
Gradual drift is you getting better or worse. A step is the ruler changing.
The repair, in order of preference
Rebuild from the break. Discard everything before the change and start the window from the first clean post. Honest and simple, and it costs you stability until enough posts accumulate — see how many posts a baseline needs. For most accounts this is the right answer.
Hold two baselines. Keep the old one for reading old posts and build a new one for new posts. Never compare across them. This preserves your history as history, which matters if you are trying to learn from a post that predates the change.
Do not rescale. It is tempting to compute a ratio between the two periods and multiply the old numbers by it. Resist. A metric change is rarely a constant factor — the views change rewarded rewatchable content more than everything else, so one multiplier would overstate some posts and understate others. A rescaled baseline looks more precise and is less true.
What to record
Whatever you do, write down the date of the break and what changed, and keep it next to the numbers. The reason is practical: in six months you will look at a post from before the change, see a multiple that seems impossible, and have no idea why.
This is the whole argument for publishing a method rather than just reporting results. A number without its definition is not evidence, and definitions move.
How often this matters
More than people expect. Platforms change what counts as a view, a play, an impression, a reach, and they change which surfaces feed into which figure. Any one change is small. The accumulation is why year-over-year comparisons on social platforms are mostly fiction, and why we keep every claim on this site tied to a window and a date.