I can recall in the mod-late 1990s when Mainstream America was starting to go online for the first time, everyone was training their kids "never give out your real name." From that simple bit of stranger-danger paranoia, we built a lot of communities as psuedonymous by default-- your AIM "screen mame" was rarely your given name, you could have different usernames on each forum, your email address probably referenced your favourite sports team or anime character.
This inherently constrains aggressive "passive" personalization. Without an obvious canonical identity, you don't want to try to cross-profile too aggressively, because user "hakfoo" on site A may well be different from user "hakfoo" at site B, and you had to assume that any ID you tracked was limited or transient: when you go off to college or apply for a job, you're probably not going to want to be slapping "PonyGirl1987" on your resume.
I wonder if it's that the algorithms are limited to being one-track or if they're overoptimized to being one-track though. I assume they have a profile somewhere that looks like "12% coin collecting, 31% travel to Paraguay, 9% 2004 Ford Focus Repair, ..." and then they offset that data with what content produces the best revenue/engagement/metric of the week. In the process, many secondary interests simply get demoted to the pont you see nothing but tire change tutorials.
The problem was (almost) no one was making money off it. Online advertising, as it was originally done, was a joke. Most of the early internet was based on ideals of quality, community, and freedom. None of these make much money.
So people started hunting for what DID make money, and they discovered data harvesting and targeted advertising. And like a cancer, that unholy pair devoured most of the web. Information really WAS power, like all those breathless articles and hacker manifestos in the 90s said. Power and MONEY. And the new masters of this realm have dedicated themselves to taking as much of YOUR information as possible. They need it, like Elizabeth Báthory would have needed the blood of virgins if she was an actual vampire and not just insane.
https://www.huffpost.com/entry/data-and-goliath-digital-surv...
https://support.mozilla.org/en-US/kb/containers
https://support.mozilla.org/en-US/kb/profile-manager-create-...
You should be able to trust a popular open-source project, because you trust that there are folks who will go in and take a sticky beak at these kinds of things. This why a lot of security folks prefer open-source software over close-sourced solutions.
Surely, if Chromium were cross-correlating profiles, then it'd be on the front page of Hacker News in short order.
I'm unaware of any potential problems relating to fingerprinting and so on, but for your use case it doesn't sound like it'd be a problem.
Realistically, on the modern net you need to: Use a VPN, kill cookies and ads aggressively, have multiple accounts (or avoid logging on at all in some cases), and never, ever use 2FA except with financial institutions. That's the bare minimum. No point in just complaining that tech companies are evil all the time. The smart organism adapts to its environment, and uses the tools at hand.
It's insane that I feel it helpful to take on "multiple personalities", but there it is.
Part of it is that these algorithms are fairly one-track. They can mix it up a bit, but it's always too much of one thing and too little of another. They can't truly comport with the reality that someone can have multiple interests and tastes.