The rider-1 real-client capture (n5 task, executed at the wave-5 gate)
corrected the wire emission shape: upstream ls-refs.c sends the
literal `unborn` token in the oid field; the zero-oid shape we served
since wave 2 made a real clone fetch-want the null oid
(`ERR object <zero-oid> does not exist`, captured live). The inbound
RefLine parser accepted both shapes all along; the emission pins to
the upstream form. Pre-publish draft-shape amendment (the OQ-03 freeze
is at the core's publish): ADR-018 gains the §5 amendment note;
transport.md §ls-refs records the validation + corrected shape; the
pinned rows re-byte-pin the literal (v2_machines renamed, the
stateless/parse-back rows updated); the RefLine/LsRefsArgs doc
comments and the CHANGELOG wave-2 bullet corrected.
Verification: cargo test (workspace 375 default / 378 all-features, -p
alkgit 376, -p alkgit --features sha256 379, 0 fail), both clippies -D
warnings clean, cargo fmt --check, cargo doc zero warnings, the fuzz
corpus gate 13/13, cargo publish --dry-run green.
glm-5.3-flash
8a10a678e0