Who pays for the work that fails: costs that today's changes make explicit
Several of today's protocol changes are about costs that existed but were not stated. The amendment to SIMD-0529 is the clearest. Under the old text, a call to Solana's big-integer modexp syscall that failed validation charged nothing for the syscall, so the work a validator did before rejecting bad input was free to the caller. Now a flat base charge comes first, and the amended order is declared consensus-critical so that every implementation consumes exactly the same compute units on every abort path.

Bitcoin Core's PR 36864 deals with a smaller hidden cost: an RPC field, iswatchonly, that has answered false for every address since an earlier change but was still being returned, and that the v30 release notes wrongly described as removed everywhere. Moving it behind -deprecatedrpc makes the deprecation real, and the next release can drop it. The rippled change is about a dependency rather than a fee. Its supposedly static Linux binaries already needed the shared libgcc_s.so.1 at run time, and the combination of two unwinders turned an ordinary C++ exception into a silent abort. Linking libgcc dynamically states the dependency instead of hiding it.
IOSCO's quantum report adds the largest number of the day. A post-quantum ML-DSA-44 signature is about 2,420 bytes, 35 to 40 times a common elliptic-curve signature, and on a public chain that difference is paid in block space, bandwidth and fees. In each case the change does not create a new cost; it moves an existing one to where the people who bear it can see it.