[kwsantiago]

Draft, for review. No activation height is set on any network, so this is inert as it stands.

The retarget at R measures time(R-1) - time(R-2016): 2016 blocks, 2015 intervals. The interval between R-1 and R is measured by nothing, and that gap is what the timewarp attack spends.

Two rules, on a new unscheduled buried deployment:

• the retarget at R measures time(R-1) - time(R-2017), so consecutive windows share an endpoint
• a block closing a period must satisfy time(H) >= time(H-2016)

Both are needed. Contiguous windows alone leave the Murch-Zawy residual: a negative span is credited as nPowTargetTimespan/4, and that 4x rise pays for two 4x drops. Alone, the schedule reaches 0.0156x of starting difficulty over 31 seconds of chain time; with both, 336 days.

The floor is BIP-54's non-negative-duration rule re-anchored to the corrected window. BIP-54's other rule becomes unnecessary, which drops its 7200s grace.

Enforced in ContextualCheckBlockHeaderVolatile so ConnectBlock re-runs it, and re-derived at startup by FindInheritedInvalidBlocks, as the BLAKE2b fork already does.

Hardfork: old nodes compute different nBits and split at activation.