Custom user-agent fails bot detection

Not much. They're in the business of detecting tampering with the connection identification, which, apparently, you're trying to do, and they seem to be good at this. (They might cross–reference the platform claimed with e.g. fonts available or GPU details that don't match between the two. Or you've thrown resist–fingerprinting or strict ETP/PBM into the mix, so they see even less real values and more greasing, and have to guess more, so the mismatches might be more apparent?) That said, if I fake the UA the only thing that happens for me on the link is I have to manually check the turnstile, but that's all. (Using stock nightly, and faking macOS as Windows.) So it must be in tandem with something else you've hardened, and adding to the overall scoring. If using e.g. a VM like ''<code>browser.lol</code>'' with datacenter origin, I get the turnstile checkbox right away, and if I spoof as Windows there the turnstile keeps looping infinitely. So the non–residential IP apparently raises the bar somewhat. If you wanna get rid of some entropy, I'd first suggest using the non–Snap version of Linux UA for starters, which is more broadly used, and shared also with Android tablets and the like, so it's more generic than the Canonical–provided one — if that helps with your Turing checks.