Shared-goal-authority delivery records get a ledger directory¶
Non-normative for the shared Goal authority model: it adds no runtime invariant. It relaxes one condition the ledger check imposed — a ledger directory had to name an RFC whose Appendix A is the execution ledger — and points this RFC's future delivery records at files.
- The cost was measured by the maintainer, not by this entry. #4677 surveyed
every open PR against
origin/main = d8e7af141on 2026-09-17 and found 19 of 34 headsmergeStateStatus=DIRTY, three of them (#3820,#4061,#4672) colliding on this one file. This entry re-measures no head: GitHub computes mergeability lazily, so a same-day recount is not available without forcing a computation per PR. - Why records collide in this file. Delivery records for this RFC are dated
subsections inside Appendix C —
Provider-first terminal lifecycle checkpoint (2026-09-07),Cross-RFC semantic and presentation conformance checkpoint (2026-09-12),Next delivery and parallel provider work. They sit at the end of a 3,125-line English file and a 2,474-line Chinese one, so two branches recording a delivery insert at the same position by construction. - The convention already existed and could not be adopted here.
check_rfc_ledger_entriesinexamples/docs-governance-smoke.pyrequired a ledger directory's RFC to contain the literal headingAppendix A: Execution ledger. This RFC's Appendix A isWhat This Evidence Proves, so per-file entries would have meant renumbering Appendices A-C across both language files — a large mechanical diff that itself forces rework on the branches the change exists to help. The check now acceptsAppendix <letter>: Execution ledgerin the English document, and this RFC's Appendix D is the pointer. - Existing records stay put, for the reason the semantic-vocabulary round recorded on 2026-09-18: they are append-only history nobody edits, so they were never the thing a later branch conflicts with — the shared insertion point ahead of them is.
- What this does not do. It resolves no conflict that already exists; each affected branch still rebases once. It does not make a ledger entry reviewable evidence — entries remain non-normative records of what a change measured and what it did not establish. And it adds no navigation entry: the hosted-docs check requires a file for every nav entry, not a nav entry for every file.