What is the point of miner signaling in a soft fork activation mechanism? What should the threshold be on miner signaling?
The second part of this question was asked by devrandom on IRC.
A question on the risks of reducing the threshold to below 95 percent was asked back in 2017 here.
What is the point of miner signaling in a soft fork activation mechanism?
Miners are signaling readiness for a soft fork activation, they are not signaling to support (or vote for) a soft fork. It seems generally accepted at the time of writing (February 2021) that any opposition towards a soft fork should be raised and discussed before an activation mechanism is proposed. If arguments are raised that haven't previously been considered or that opposition is material and sustained across the wider community, activation for that soft fork shouldn't be considered.
Ideally you want as many miners enforcing the Taproot rules at the point it activates as possible. Otherwise an invalid Taproot spend could creep into a block and some miners wouldn't reject it as they would be treating it as an anyone-can-spend. It would need a small, naturally occurring re-org to get that invalid Taproot spend out of the blockchain. This wouldn't be a disaster but ideally you don't want a greater frequency or greater magnitude of re-orgs than usually occur. Ideally you want all miners enforcing the Taproot rules from the point of activation.
There is also the consideration of miners deliberately or inadvertently producing blocks with invalid Taproot spends in them and fooling SPV/non-upgraded clients that is discussed here.
As devrandom pointed out on IRC there is motivation for presumably unready miners who haven't yet signaled to get ready urgently if the soft fork is activating soon. They don't want to expend resources mining blocks (with an invalid Taproot spend in them) that will ultimately be rejected by a proportion of the network. But that doesn't guarantee they will be ready and may lead to miners rushing to get ready for activation which again is not optimal.
What should the threshold be on miner signaling?
This is harder to answer as the threshold chosen appears to be a trade-off between ensuring as many miners are ready to activate the soft fork as possible and preventing a small minority of miners from unnecessarily delaying the soft fork for political or strategic reasons. With the SegWit soft fork in 2017, a BIP 148 user activated soft fork needed to be proposed because miners were assessed to be deliberately blocking activation to attempt to force through an additional block size increase.
In this developer survey conducted by AJ Towns the question "What do you consider a reasonable threshold for activation by hashpower supermajority?" it appears 90 percent or 95 percent would be the preferred threshold. The threshold for the SegWit activation was set initially at 95 percent.