You have to be specific when you pick the algorithm if it's to work the way the author suggests, and preferably something that is not easily shifted, such as the domain name and not the site's title.
I was in complete denial about how bad the situation was until I encountered some light teasing from acquaintances who didn't even know me that well -- just well enough to know I couldn't ever remember my passwords. I started keeping tick marks on my calendar and found a ~75% success rate, although the 75% was composed of the handful of logins I used every day and the 25% was "everything else" so the reality was that the password derivation system was failing for a strong majority of passwords.
I gave up and started using lastpass. No regrets, but many positive surprises: that time their network was breached but it didn't matter because they didn't use shit encryption, the automated password reset feature for big-profile leaks (they give you a "todo" list which is often as simple as clicking a single button next to each item), the ability to easily store serial numbers, "verification questions", and other nonsense, a general lack of guessing passwords several times before success, and an ability to dramatically up the cross-entropy of my passwords.
I might move from lastpass to a physical repository some time. But I am never, ever going back to a password derivation scheme.
I wish you luck. Or an eidetic memory. You'll need it.
It is not very strong, I don't have some lookup table or complicated stuff just some small variations for each site.
I've started using 1Password as of late though, but that's mainly because I sometimes want to store notes and whatnot for websites and this makes that convenient.