Backfill missing frontmatter fields + Alternatives sections + Status headings for ADR-051 + ADR-052 per ADR-067-style narrow-relaxation discipline

Published

May 20, 2026

ADR-072: ADR-051 + ADR-052 frontmatter and structural backfill

Status

Accepted.

Context

The user’s 2026-05-18 self-audit (decisions/audits/REPO_AUDIT_2026-05-18.md §P1.4) flagged ADR-049 + ADR-050 for empty frontmatter fields (closing_commit:, supersedes:, superseded_by:). The 2026-05-20 audit re-verification confirms ADR-049 + ADR-050 have since been populated (closing_commit: 423c2c8 and 3b16036 respectively). The user’s flag was stale.

The actual structurally incomplete ADRs are ADR-051 and ADR-052, both 2026-05-18 governance ADRs that ship with:

  • ADR-051: empty closing_commit: field (line 102), empty transcript: field (line 116), no ## Status body section (jumps straight from frontmatter to ## Context), no ## Alternatives Considered section, and an opening line # ADR-051 — ... that uses an em-dash instead of the template-convention colon (# ADR-051: ...).
  • ADR-052: empty closing_commit: field (line 86), empty transcript: field (line 98), no ## Alternatives Considered section.

Both ADRs document substantive supersession decisions in the ADR-034 + ADR-039 + ADR-050 chain. Their own frontmatter doesn’t conform to the schema they invoke — the immutability discipline applies to other ADRs but these two skipped fields.

Decision

Treat the empty-frontmatter-fields gap as a fourth narrow-relaxation class adjacent to ADR-067 (Class A slug typos), ADR-068 (Class B broken external references), ADR-069 (Class C publisher-URL canonicalization), ADR-070 (Class D render-only markdown corrections). The same reasoning applies: no decision content changes, only audit-trail completeness is restored.

ADR-051 backfill:

  • closing_commit: set to v1.0.9 (Block A; Block B carryforward to v1.1.x). Block A (T0 score-match wiring) closed via ADR-058 which landed at v1.0.9. Block B (38 invariant-test stubs unskip per ADR-039 gate 3) remains carryforward to v1.1.x per ADR-051’s own §Decision Block B narrative.
  • ## Status body section added (value: Accepted. — matches frontmatter).
  • ## Alternatives Considered section added retroactively, documenting the three alternatives surfaced at 2026-05-18 lock time: (1) drop both T0 + invariant commitments outright (rejected: methodology contracts can’t be silently dropped); (2) fix-forward inline before v1.0.0 tag (rejected: time pressure at submission close + the implementation surface for both blocks was ~100 LOC + 38 test bodies, too large for the v1.0.0 rehearsal window); (3) no carryforward ADR, let ADR-034 + ADR-039 stand unmet (rejected: violates immutability discipline; the original commitments need an explicit superseding record).
  • Opening line fixed: # ADR-051 — → # ADR-051: per ADR_TEMPLATE.md convention.

ADR-052 backfill:

  • closing_commit: set to v1.0.3. ADR-052 is a methodological reframing ADR; the act of writing it is itself the closure event. Per git log, ADR-052 was introduced at commit 3ba4636 which sits under the v1.0.3 tag.
  • ## Alternatives Considered section added retroactively, documenting the four alternatives surfaced at 2026-05-18 lock time: (1) re-fire full-FT OOD inference on a fresh DC (rejected: methodological reasoning was already load-bearing per LoRA’s paired-bootstrap evidence; cost
    • repeat-FUSE-risk + ~$5-12 GPU spend didn’t justify); (2) restart from clean checkpoint (rejected: same reasoning, same cost); (3) abandon outright with no ADR (rejected: documentation contract requires explicit reframe of ADR-050 Revision 2); (4) no ADR-052 reframing, let ADR-050 R2 stand as operational forced-drop (rejected: the methodological reasoning was real and load-bearing for the characterisation conclusions; same-day retcon optic is the trade-off for honest documentation of the actual decision logic).

transcript: fields remain empty for both ADRs per the project’s “transcripts gitignored except transcripts/README.md” discipline; the field is reserved for future-session continuations and need not be populated retroactively.

Consequences

  • Audit-trail completeness restored: scripts/regenerate_audit.py --check now sees populated closure metadata + complete body structure for ADR-051 and ADR-052
  • Future reviewers checking frontmatter completeness no longer find a gap that contradicts the immutability discipline these ADRs themselves invoke
  • The ## Alternatives Considered sections retroactively document the decision space that was genuinely considered at lock-time (the transcripts capture it; this ADR surfaces a sanitized summary into the public record)
  • ADR-051 + ADR-052 decisions are unchanged + remain immutable; only the frontmatter + missing-sections debt is closed
  • Closes the loop on what the user’s prior 2026-05-18 self-audit was trying to flag (the audit was right about the gap class; the targets shifted from ADR-049+050 to ADR-051+052 after the v1.0.x patch cycle populated 049+050)

Alternatives Considered

  1. Leave ADR-051 + ADR-052 as-is. Rejected — the SDD discipline applies to all ADRs including the meta-governance ones. Empty closing_commit is a methodology gap a reviewer checking frontmatter completeness would surface; ADR-051 also lacks template-required body sections.
  2. Backfill via direct edit, no ADR-072. Rejected — the immutability rule means even narrow editorial edits need an authorizing ADR. ADR-067-070 established the precedent; ADR-072 extends it to the frontmatter-backfill class.
  3. Two separate ADRs (one for 051, one for 052). Rejected — same class of fix, same scope, applied at the same time. One ADR is cleaner; the audit trail isn’t materially improved by splitting.
  4. Treat ADR-029 misattribution refs (handled in ADR-071) and this frontmatter backfill as one ADR. Rejected — they’re different narrow-relaxation classes (slug-sweep vs frontmatter-backfill) with different supersession scopes. Cleaner to keep them as separate ADRs in the 071 → 072 → 073 → 074 → 075 sequence.

Linked ADRs

  • ADR-051 (v1.0.x-carryforward-of-t0-and-invariant-scaffolds): superseded on the empty-frontmatter axis only; decision content unchanged
  • ADR-052 (full-ft-ood-drop-methodological-reframing-of-adr-050-r2): same supersession scope as ADR-051 here; ADR-075 separately supersedes ADR-052 on the narrative axis
  • ADR-058 (eval-from-hub-non-dry-run-body-narrow-supersession-of-adr-051-block-a): closed ADR-051’s Block A; cited in ADR-051’s new closing_commit value
  • ADR-067, ADR-068, ADR-069, ADR-070 (the 4-ADR immutability-relaxation chain): ADR-072 follows the same narrow-relaxation discipline (no decision content changes; audit trail strictly improved)
  • ADR-075 (planned: full-ft-ood-drop-rationale-unified-narrative): supersedes ADR-052 entirely on the narrative axis; orthogonal to this ADR’s frontmatter-axis supersession

Transcript

Transcript file (gitignored per AGENTS.md): the 2026-05-20 audit re-verification cycle that surfaced the stale REPO_AUDIT_2026-05-18.md flag + identified the ADR-051+052 frontmatter gap; captured in ~/notes/prompt-injection-audit-2026-05-20.md (consumer side) and ~/notes/prompt-injection-audit-2026-05-20-adr-appendix.md (full ADR governance audit).