TL;DR: There should be an option, taproot=lockintrue, which allows users to set lockin-on-timeout to true. It should not be the default, though.
As stated in my previous post, we need actual consensus, not simply the appearance of consensus. I'm pretty sure we have that for taproot, but I would like a template we can use in future without endless debate each time.
- Giving every group a chance to openly signal for (or against!) gives us the most robust assurance that we actually have consensus. Being able to signal opposition is vital, since everyone can lie anyway; making opposition difficult just reduces the reliability of the signal.
- Developers should not activate. They've tried to assure themselves that there's broad approval of the change, but that's not really a transferable proof. We should be concerned about about future corruption, insanity, or groupthink. Moreover, even the perception that developers can set the rules will lead to attempts to influence them as Bitcoin becomes more important. As a (non-Bitcoin-core) developer I can't think of a worse hell myself, nor do we want to attract developers who want to be influenced!
- Miner activation is actually brilliant. It's easy for everyone to count, and majority miner enforcement is sufficient to rely on the new rules. But its real genius is that miners are most directly vulnerable to the economic majority of users: in a fork they have to pick sides continuously knowing that if they are wrong, they will immediately suffer economically through missed opportunity cost.
- Of course, economic users are ultimately in control. Any system which doesn't explicitly encode that is fragile; nobody would argue that fair elections are unnecessary because if people were really dissatisfied they could always overthrow the government themselves! We should make it as easy for them to exercise this power as possible: this means not requiring them to run unvetted or home-brew modifications which will place them at more risk, so developers need to supply this option (setting it should also change the default User-Agent string, for signalling purposes). It shouldn't be an upgrade either (which inevitably comes with other changes). Such a default-off option provides both a simple method, and a Schelling point for the lockinontimeout parameters. It also means much less chance of this power being required: "Si vis pacem, para bellum".
This triumverate model may seem familiar, being widely used in various different governance systems. It seems the most robust to me, and is very close to what we have evolved into already. Formalizing it reduces uncertainty for any future changes, as well.