The measure of uniqueness is effectively entropy (entropy = log base 2 of the number of possible passwords). Ease of use is just unnecessary (for entropy purposes) padding on top of that (constraints on symbol sets, order, etc). The necessary threshold is entirely dependent on your threat model, combined with the risks that your use cases are exposed to when using the password.
Your point that using entropy rich passwords is foolish, is incorrect most of the time, and even if it weren't people's general understanding of where specifically to draw that line is generally significantly underestimated as a collective. The evidence of this is the amount of data breach information available, and the actual attacks which are available using this information.
Service breaches happen where the password database is insufficiently secured, and that often correlates strongly with insufficient protection of the password hashes. This means that low entropy passwords can be cracked and used for a period of time before the data breach is discovered. Entropy (and MFA) are the only protections against that.
56 DES was cracked in a day in 1999. Your passwords should definitely have more entropy than that. Probably around 80 bits is enough (about 16 alphanumeric characters or so). Your much lower threshold on entropy is insufficient for pretty much any reasonable threat model except public access.