ADR-0125 — Re-mint malformed-requirement-heading as C019; the C018 reservation stands

Context

The malformed-requirement-heading check shipped on 2026-07-02 under the id C018 (contract 0.9.0 → 0.10.0) with no minting ADR — and C018 was not free: ADR-0097 explicitly reserves C018 for the oversized-packet check ("the CheckId set stops at C017"), and ADR-0094 repeats the reservation. A same-day multi-lens panel review (DP-8 in the family workspace's FINDING-review-gate-measurement) caught the collision, plus two behavior defects in the shipped check: the detection regex false-fires on prose group headings whose first token is NAME-digits-LETTERS (### UTF-16LE handling captures UTF-16LE), and the diagnostic can quote a truncated token (### AC-4a-extra reports AC-4a).

The owner's decision (2026-07-02): re-mint under the next free id rather than reinterpret two accepted ADRs, and fold the regex tightening into the same change.

Decision

D1 — C019 malformed-requirement-heading, warning (contract 0.10.0 → 0.11.0)

The check keeps its meaning and severity; only the id moves. A ### heading shaped like a requirement id but carrying a lowercase split-suffix (AC-004a) parses as plain prose — the requirement silently vanishes from scope and coverage — and C019 makes the disappearance visible. The fix remains a digits-only id: a split requirement gets its own number, never a suffix.

D2 — C018 stays reserved

The ADR-0097/ADR-0094 reservation of C018 for a future oversized-packet/decomposition-predictive signal is untouched and back in force. checks.yaml carries an explicit reservation comment so the next mint cannot collide by accident.

D3 — the detection shape narrows to the lowercase split-suffix

The regex requires the first suffix character to be lowercase (/^###\s+([A-Z][A-Z0-9]*-\d+[a-z][A-Za-z0-9_]*)/), because the split-suffix convention the check exists to catch is lowercase (AC-004a, AC-004b) while the measured false-positive class is uppercase-continuation prose (UTF-16LE, C-3PO). The capture continues through word characters (no trailing \b), so ### AC-004a_note is caught whole and the diagnostic quotes the full token.

Recorded residual risks (accepted, not silent): an UPPERCASE-suffixed id-like heading (### AC-004A) reads as a prose group heading and is not flagged — no instance observed in the family corpus, and warning-severity noise on prose was measured the worse error; a tab after ### follows CommonMark for the parser but is outside this check's shape; a malformed SOL opener (REQ AC-004a:) is structural-check territory for the SOL catalogue (SOL-S*), not a spec-form C-check — recorded there as future work, not minted here.

Consequences

• checks/checks.yaml is the registry of record: version: 0.11.0, C019 minted, C018 commented as reserved. suspec-cli pins CONTRACT_VERSION = '0.11.0' and its sibling drift guard reconciles the two on every CI run.
• The oversized-packet advisory in suspec review (neutral size info, no band) is unchanged and its in-code references to "C018 (oversized-packet)" are correct again under the restored reservation.
• Minting discipline is restated by example: a contract id ships with its ADR — the missing-ADR gap this re-mint closes is the collision's root cause, not the id arithmetic.

Status

Accepted (2026-07-02).