Conceptually, released versions have three-part version numbers such as 1.0.0, 1.1.0, and 1.1.1. However, trailing zeros are suppressed, so that the commercially published version (conceptually 1.0.0) was simply called the 1st edition and its successor (conceptually 1.1.0) was called revised edition 1.1. For your own forked versions, please add a hyphen followed by something identifying. For example, if you fork off of 1.1.1, you might identify your version as 1.1.1-somename.
Changes that do not affect the PDF file do not change the version number. This would adding comments into the LaTeX source files, refactoring the LaTeX source using macros, as well as changes in the web site, README.md, etc.
Changes that affect the PDF file but not the substantive content advance the third (rightmost) part of the version number. For example, fixing typos falls into this category. Even minor technical bug fixes would fit this category; the rule is that for a bug fix that just removes some roadblock or frustration from the reader's learning, but does not change the content of what is ultimately learned, only the third part of the version number advances. My intention is to batch together multiple fixes of this level, so that new releases normally are separated by months rather than days or weeks. However, any issue likely to cause serious annoyance will prompt an immediate release; for example, 1.1.1 is soon after 1.1 because broken links can be so annoying. This portion of the section number could in principle advance arbitrarily far; no number of accumulated minor fixes ever equals a substantive change.
Changes that affect substantive content will ordinarily advance the second (middle) part of the version number and reset the rightmost part to 0. Examples would include the addition of a new section or a substantial rewrite to an existing section. This could include new material or an update to old material to reflect changes in one of the example systems or in the general state of the art. If a student who studied the old version would thereby be at risk of a poor grade on an exam that was written using the new version, then the change definitely rises to this level. Once there are enough changes of this kind, a subjective determination may be made to advance the leftmost version number.
Changes that are so broad in scope that they would merit a new "edition" in traditional publishing advance the first (leftmost) part of the version number and reset the others to 0. This is an even more subjective determination than what contstitutes a change in "substantive content." Most likely, a change of this scope will result from the cummulative impact of enough smaller changes. If every single chapter has been brought up to date and several of them have been reworked to reflect new perspectives, rather than just to catch up with changed details, a new edition would be order. Entire new chapters might also prompt a new edition.
On the cover page, the version number is followed by the date when pdflatex was run to produce the PDF file from the LaTeX source files. This date is not part of the version number and can change without the version changing. Because the version number is supposed to change with any change that would affect the PDF file, there will ordinarily only be one released PDF file with any version number, and hence among released PDF files, there will be a one-to-one correspondence between version numbers and dates. However, if someone runs pdflatex themself to generate their own PDF file from the released source files, they would find up with a different date. This is intentional, because it makes it easy to spot the difference between a locally produced PDF file and the released one. There might be little differences, even if the source files were the same, because pdflatex can be influenced by aspects of the local environment.