md-codec-v0.35.0

md-codec [0.35.0] — 2026-05-17

Closes md-codec-decode-with-correction-supports-non-chunked-md1
(filed at v0.22.x follow-ups cycle Phase B.8). Lockstep with
mnemonic-toolkit-v0.24.0 Tranche D.

Added

• decode_with_correction non-chunked-form detection (crates/md-codec/src/chunk.rs).
  Pre-pass routes strings.len() == 1 inputs whose first-symbol bit-0 is
  0 (non-chunked header sentinel) directly into decode_payload,
  bypassing the chunk-reassembly path. Falls through to the existing
  reassemble for len > 1 OR len == 1 && bit-0 == 1 (preserves
  v0.34.0 count-1-chunked-of-1 semantics; multi-strings-supplied-as-one
  still surfaces Error::ChunkSetIncomplete).
• 5 new unit cells in crates/md-codec/tests/bch_decode.rs covering the
  non-chunked-form detection branch.

Fixed

• mnemonic repair --md1 / md repair invocations against non-chunked
  single-string md1 (the form emitted by plain md encode for small
  payloads) previously surfaced as wire-format-mismatch errors. They now
  decode through the BCH correction pipeline like any other md1 input.

Notes

• Wire-format change is purely additive: non-chunked md1 was previously
  rejected by decode_with_correction; now decodable. No corpus refresh
  required; no downstream-codec migration burden.

Resolved (FOLLOWUPS)

• md-codec-decode-with-correction-supports-non-chunked-md1 (cross-repo
  primary). GUI + toolkit + ms-codec mirrors transition open →
  RESOLVED in md-codec-v0.35.0 lockstep.