I woudl like to propose a BIP that works something like this:
1. Allow users to signal readiness by publishing an EB. This EB is an absolute upper bound, and cannot be overridden by miners. Current EB is 1MB, the status-quo. Maybe EB can be configured in a config file, not a UI, since it's an "advanced" feature.
2. Miners can also signal readiness by publishing their own EB in a block.
3. If 95% of blocks within a one month signalling period contain an EB greater than the previous consensus EB, a fork date is triggered at 6 months using the smallest 5th percentile EB published. (Other times can be selected, but these are fairly conservative, looking for feedback here). Miner signalling is ignored during the waiting period.
4. Block heights used for timing
5. After 6 months, any users which already have the new EB or greater begin actually using it to validate transactions. Users use the EB or the latest 95% consensus triggered value - whichever is less. This means that the portion of users that originally signaled for the increase do not have to upgrade their software to participate in the hard fork.
6. Core can (optionally) ship a version with a default EB in-line with their own perceived consensus.
7. Some sort of versioning system is used to ensure that the two networks (old and new) are incompatible... blocks hashed in one cannot be used in the other.
Any users which don't already have the new EB or greater should update their EB within the 6 month period - or they will be excluded from the majority fork.
It would be in the best interests of major exchanges and users would to publicly announce their EB's.
Users are free to safely set very high EB levels, based on their current hardware and network speeds. These EB levels don't cause those users to accept invalid blocks ever. They are safe because block size transitions behave like normal hard forks with high miner consensus (95%).
No code changes will be needed to fork the network as many times as both users and miners feel the need to do so. (Bitcoin core is off the hook for "scaling" issues...forever!)
If a smaller block size is needed, a reduced size can also be published and agreed upon by *both* users and miners using a the same mechanism, but the largest 5th percentile is used. In other words... the requires broad consensus to deviate from status quo and fork.
Any new node can simply follow these rules to validate all the blocks in a chain... even if the sizes changes a lot (at most twice per year).