the issue is that the two arms differ in several ways at once, including transport, flags and who started the browser, so the post can’t show which one matters. the debug port being ignored on the default data directory matches what they quote, tho the silent failure is only their experience.

old headless sent the headlesschrome token, and since chrome 112 only the headless shell build keeps that behavior. treating a capped query as not run rather than as a negative result is the most useful takeaway, since silent false negatives quietly spoil datasets