Divergence is a mismatch between what price does at its extremes and what the indicator does in those very same bars. Price prints a new high, and at that high the indicator reads lower than at the previous one — that is the whole mismatch.
The key words are "the same". What gets compared is not arbitrary points on two charts, but price peaks and troughs against indicator values in exactly the bars where those peaks stand.
How it is built
First, price extremes are located: local highs and lows, each with its own required number of bars to the left and right. Then two neighbouring extremes of the same type are taken — two peaks or two troughs.
After that two pairs of numbers are compared. What price did: the new peak is higher or lower than the previous one. What the indicator did in those same two bars: its value rose or fell. If the directions disagree, there is a divergence.
Why it is sensitive to settings
Divergence depends entirely on what counts as an extreme. Move the requirement for how many bars must sit to the right of a peak from two to five, and some peaks disappear — and with them the divergences that were built on them.
So two tools with different swing settings will show a different set of divergences on the same chart, and both will be right within their own rules. That is not a markup bug but a consequence of the definition.
For the same reason we use one extreme detector across all modules: the same one that marks up structure in Smart Money and Elliott waves. Otherwise a peak would exist on one page and be missing on the next.
What divergence does not tell you
A mismatch describes what already happened: the current move's impulse is weaker than the previous one by this indicator's measure. It promises neither continuation nor reversal.
A divergence can persist for a long time, renewing at every new extreme while the trend carries calmly on. That is why it is not an entry signal and is not used as one here — it is one characteristic of a stretch of chart, no more.