On Thu, Feb 4, 2016 at 5:56 PM, jl2012 via bitcoin-dev <bitcoin-dev@lists.linuxfoundation.org> wrote:

No, the "triggering block" you mentioned is NOT where the hardfork starts. Using BIP101 as an example, the hardfork starts when the first >1MB is mined. For people who failed to upgrade, the "grace period" is always zero, which is the moment they realize a hardfork.

Clients have to update in some way to get the benefit of this right?

An SPV client which fully validated the header chain would simply reject the hard forking header. Last time I checked, the Bitcoinj SPV wallet ignored the version bits, and just followed the longest chain. Is that still the case?

In fact, does Core enforce the 95% rule for the soft-forks before checking for long forks? I am assuming that it happens when checking headers rather than when checking full blocks.