This one doesn't make any sense to me either.
Some quick tests show that:

1. citeproc-js indeed behaves this way (as one would hope given the test)
2. This is unrelated to the numbers in the title: Title A, Title B, Title C behave the exact same way

<sort>
  <key variable="title"/>
  <key macro="issued-date"/>
</sort>

Sorts as

Title 1
Title 2
Title 3

as expected (with or without the title being part of the issued-date variable).

It'd be great if @fbennett could chime in, but I'd currently call this as an incorrect test and suggest changing it (likely by just removing the title variable from the macro since the test is about dates)

Ah, I remember this one. Rintze and I discussed the case in 2013, back in the BitBucket days. In the discussion Rintze also questioned the discarding of the non-date text from the macro when generating keys, and noted that doing so is an off-spec invention.

As I note in the linked thread, intermixing date and string values in a single key can't be made to produce intuitive sort results. Dates are a bit special: keys for non-date variables are simple strings or numbers that can be derived directly from the variable; but date keys must generate a machine-readable sort key that takes account of the date-parts element of the rendered form. As a result, there are cases in which a macro containing a condition must be used to produce the date key (APA is mentioned in the discussion).

The need to express date keys in a macro can be addressed in one of two ways. One would be to assume that such macros will always contain exactly one date. The other would be to extract a date from an existing macro, and ignore everything else it contains. citerproc-js adopts the latter approach. While it (obviously) can create confusion, it has the advantage that, if a style designer mistakenly assumes that a macro combining two elements (creator and date, say) will treat the contents as two atomic keys, the attempt will throw bad sort results immediately. (There is an example of this in the repo, in the initial history of deutsche-sprache.csl, where the initial checkin of the style is followed immediately by an amendment to the sort keys.)

So that's the reasoning behind it. On a quick check of the repository (based on macro names alone), it looks like a few styles would need to be amended to add and call date-only macros if the former approach is preferred.

beltz-padagogik.csl
berlin-school-of-economics-and-law-international-marketing-management.csl
biuletyn-polskiego-towarzystwa-jezykoznawczego.csl
deutsche-sprache.csl
hochschule-hannover-soziale-arbeit.csl
keel-ja-kirjandus.csl
zeitschrift-fur-qualitative-forschung.csl
wirtschaftsuniversitat-wien-abteilung-fur-bildungswissenschaft.csl
wirtschaftsuniversitat-wien-unternehmensrechnung-und-controlling.csl
wirtschaftsuniversitat-wien-wirtschaftspadagogik.csl