Correct reviewer URL pin numeric defect in ADR-078 + ADR-079 (tree/v1.2.8 → tree/v1.0.0); axis-only supersession on the reviewer-URL-pin axis

Published

May 22, 2026

ADR-080: Correct reviewer URL pin numeric defect in ADR-078 + ADR-079 (tree/v1.2.8 → tree/v1.0.0); axis-only supersession on the reviewer-URL-pin axis

Status

Accepted.

Context

A fresh-eyes audit performed 2026-05-22 against the live GH-Pages deployment at https://brandon-behring.github.io/prompt-injection-detection-prototype/ surfaced a factual numeric error propagated across three reader-facing surfaces. Each surface claims the historical reviewer URL pin is tree/v1.2.8 and cites ADR-033 as the authorizing decision. ADR-033 does no such thing.

Source-of-truth verification (live + source):

  • decisions/ADR-033-github-release-strategy-rehearsal-plus-submission.md:113 artifact-pin table: https://github.com/brandon-behring/prompt-injection-detection-prototype/tree/v1.0.0 is listed as “Canonical source pin | Never drifts”.
  • CHANGELOG.md v1.3.0 entry confirms: “Reviewer URL pin tree/v1.0.0 unchanged per ADR-033.”
  • index.qmd:82, README.md:213, READING_GUIDE.md:94-95, WRITEUP_PAPER.md:10, WRITEUP_NARRATIVE.md:11, ADR-079 frontmatter comment line 71 + Linked-ADRs section line 336, and ADR-078 frontmatter line 24 + Consequences line 131 + line 163 + line 183 + line 197 all cite tree/v1.0.0 correctly per ADR-033.

The three defective surfaces:

  • WRITEUP.md:48 — “the historical content remains accessible at the reviewer URL pin (tree/v1.2.8) per ADR-033.”
  • decisions/ADR-078:164 — inside the Consequences section: “Reviewer URL pin is now tree/v1.2.8; reviewers navigating to EXECUTIVE_SUMMARY through current-state links would hit a 404 — mitigated by the README anchor + the redirect that could be added at v1.3.x if needed.” Note: the SAME paragraph at line 163 says “the tree/v1.0.0 pin (per ADR-033)” — the two sentences contradict each other.
  • decisions/ADR-079:228 — inside the Stub-redirect §: “The historical WRITEUP.md content is accessible at the reviewer URL pin (tree/v1.2.8) per ADR-033.”
  • decisions/ADR-079:291 — inside the Consequences section: “Historical reviewer link to WRITEUP.md remains stable via the tree/v1.2.8 pin (per ADR-033). Current-state WRITEUP.md is a redirect stub; readers following links land on the chooser and pick a guide.”

Why this matters: ADR-033’s purpose is to give an external academic reviewer a stable URL pinned to the submission state, never to drift. Citing tree/v1.2.8 as that pin tells a reader to fetch a tree state that was the v1.2.8 Quarto-navigation-restructure intermediate state (per ADR-061), not the original submission state. A reviewer following the incorrect link sees the navigation-restructure state, not the submission-original. The semantic claim “per ADR-033” is also false: ADR-033 makes no v1.2.8 commitment.

Why this falls into a supersession (not a narrow-relaxation) case: The four CLAUDE.md narrow-relaxation classes (slug typo / broken external ref / publisher-URL canonicalization / render-only Markdown) all cover surface defects that don’t change the prose intent of the ADR. The v1.2.8 claim in ADR-078 + ADR-079 IS the prose intent — both ADRs explicitly assert a specific reviewer-URL-pin commitment in their body text. Correcting that commitment is a meaningful change to the ADRs’ stated claims (even though it harmonizes them with ADR-033 + the other reader-facing surfaces). Therefore axis-only supersession, not in-place edit, is the correct discipline.

The ADR-076-class frontmatter-backfill narrow-relaxation (per ADR-073 consolidated re-statement) DOES authorize editing the superseded_by: frontmatter on ADR-078 + ADR-079 to point to ADR-080 — that’s audit-trail completeness, not decision-content alteration.

Decision

Adopt the following two-part discipline for correcting the reviewer-URL-pin numeric defect:

Part A — ADR-080 axis-only supersession of ADR-078 + ADR-079

This ADR (ADR-080) supersedes ADR-078 + ADR-079 on the reviewer-URL-pin numeric axis only. The corrected position is:

The historical reviewer URL pin is tree/v1.0.0 per ADR-033. No other tag is the historical reviewer URL pin. Any reader following a “historical reviewer URL pin per ADR-033” link should go to tree/v1.0.0.

The historical content that ADR-078 + ADR-079 describe as accessible at the reviewer-URL pin (the pre-v1.3.0 EXECUTIVE_SUMMARY.md content + the pre-v1.3.0 jumbled-hybrid WRITEUP.md content) IS preserved at tree/v1.0.0 — the never-drift submission state contains both files unchanged. The historical-content-preservation claim that ADR-078 + ADR-079 make is correct; only the tag they cite was wrong.

ADR-078 + ADR-079 bodies are unchanged per CLAUDE.md immutability. Their superseded_by: frontmatter is backfilled to ["080"] per the ADR-076 / ADR-077 narrow-relaxation discipline (extended from ADR-072 frontmatter-backfill precedent).

Part B — WRITEUP.md in-place correction

WRITEUP.md is a mutable Markdown file (not under CLAUDE.md ADR-immutability). The single-line correction is:

  • WRITEUP.md:48 — tree/v1.2.8 → tree/v1.0.0 (and remove the parenthetical-citation confusion; clarify that the historical pre-v1.3.0 content is preserved at the ADR-033 reviewer pin per ADR-080).

Cascade — 79 → 80 ADR-count claims across reader-facing surfaces

This ADR’s creation moves the ADR count from 79 → 80. The scripts/audit_adr_count_claims.py invariant (added at v1.2.14) catches reader-facing ADR-count claims that would otherwise go stale. Cascade hit-set:

  • README.md (2 hits: line 173 + line 189): 79 → 80
  • CLAUDE.md (line 13): 79 ADRs at v1.3.0 close → keep as historical v1.3.0 statement (the count was 79 AT v1.3.0; ADR-080 lands at v1.3.1). The /exploring-options Q1 lock decision is to update CLAUDE.md’s v1.3.0-close clause to “79 ADRs at v1.3.0 close (80 at v1.3.1 including ADR-080)” to reflect both points-in-time.
  • WRITEUP.md (line 33): 79 → 80.
  • WRITEUP_NARRATIVE.md (line 553): 79 → 80.
  • READING_GUIDE.md (line 83): 79 → 80.
  • WRITEUP/methodology-guarantees.md (line 12): 79 ADRs accepted across Phase 0-00 through ADR-079 → 80 ADRs accepted across Phase 0-00 through ADR-080.
  • docs/for-hiring-managers.md (line 84): 79 → 80.
  • CHANGELOG.md (line 126): the historical v1.3.0 release-note reference to “79 ADRs” is left unchanged (it documents a past state).

SUBMISSION_AUDIT.md regenerates via scripts/regenerate_audit.py; CLAIM row count goes 79 → 80.

Consequences

Positive:

  • The factual reviewer-URL-pin claim is internally consistent across all three previously-defective surfaces + ADR-033 source-of-truth + every other reader-facing surface.
  • A reviewer following any “historical reviewer URL pin per ADR-033” link now lands at the correct tag (tree/v1.0.0) — the original submission state ADR-033 designed the never-drift discipline to preserve.
  • The SDD discipline is preserved: ADR-078 + ADR-079 bodies stay immutable; the correction is recorded as a new ADR with full audit trail rather than via an in-place rewrite that would erase the defect from history.
  • The audit_adr_count_claims.py invariant fires its 6th correct time across the v1.2.13 → v1.2.14 → v1.2.15 → v1.2.16 → v1.3.0 → v1.3.1 trail — validating the v1.2.14 design intent.

Negative / cost:

  • ADR count cascade across 7 reader-facing surfaces (vs the 5-surface cascade in v1.3.0); the cascade is mechanically caught by the invariant + applied in a single commit.
  • ADR-078 + ADR-079 bodies retain their v1.2.8 prose; a reader reading those ADRs in isolation will see the wrong claim, then must follow the superseded_by: ["080"] frontmatter pointer to ADR-080 for the correction. This is the standard cost of the axis-only supersession pattern (same trade-off ADR-076 + ADR-077
    • ADR-079 inherited).

Neutral:

  • The v1.2.8 tag itself is unaffected; it remains a valid repo tag pointing at the Quarto-navigation-restructure intermediate state per ADR-061 (which is not the reviewer URL pin).
  • ADR-078 + ADR-079 decision content (EXECUTIVE_SUMMARY absorption
    • two-guide reader architecture) stands as locked. ADR-080 does not revisit those decisions.

Alternatives Considered

  1. Extend the narrow-relaxation set to a fifth class (“numeric tag-reference defect”) and correct ADR-078 + ADR-079 bodies in place. Rejected: the v1.2.8 claim IS the prose intent of those ADRs’ Consequences + Stub-redirect sections (not a typo / broken ref / publisher canonicalization / render-only delimiter). Editing those in place would erase the defect from history. The four existing narrow-relaxation classes are all surface defects; this one is a factual claim about a tag commitment.
  2. Don’t fix the ADRs; only fix WRITEUP.md and leave ADR-078 + ADR-079 with their incorrect v1.2.8 claims. Rejected: a reader following any “tree/v1.2.8 per ADR-033” link from ADR-078 + ADR-079 would land at the wrong tree state + see the false ADR-033 citation. The defect is reader-visible on the live site (both ADRs render to live HTML); fixing only one of three surfaces leaves the other two misleading.
  3. Bigger fix: rewrite ADR-078 + ADR-079 entirely to reflect the correct pin + all other minor v1.3.0 polish. Rejected: scope creep + violates the axis-only supersession pattern that has served the v1.2.x → v1.3.0 trail (ADR-076 / ADR-077 / ADR-078 / ADR-079 all axis-only-supersession). The single numeric defect warrants a single axis-only supersession.
  4. Treat as a CLAUDE.md amendment (extend the narrow-relaxation set to authorize this specific class). Rejected: same reasoning as Option 1 + amendments to CLAUDE.md are themselves load-bearing governance changes that warrant their own ADR + transcript trail. Single ADR-080 is the cleanest option.

Linked ADRs

  • ADR-033 (GitHub release strategy + reviewer URL pin): the source-of-truth ADR pinning tree/v1.0.0 as the canonical reviewer URL. ADR-080 cites this as the authority being correctly applied by the post-fix state.
  • ADR-067 / ADR-068 / ADR-069 / ADR-070 (the four narrow-relaxation classes consolidated in ADR-073): cited to clarify that ADR-080’s axis-only-supersession discipline is distinct from the in-place narrow-relaxation classes.
  • ADR-072 (ADR-051 + ADR-052 frontmatter + structural backfill): the frontmatter-backfill narrow-relaxation precedent that authorizes editing superseded_by: on ADR-078 + ADR-079.
  • ADR-073 (immutability rule consolidated re-statement): the canonical immutability framework ADR-080 invokes for the body-vs- frontmatter distinction.
  • ADR-076 (superseded_by + closing_commit frontmatter backfill): direct precedent for the frontmatter-backfill pattern applied here.
  • ADR-077 (supersession-backlink + frontmatter octal-quoting backfill): the audit-tool ADR (scripts/audit_superseded_by_backlinks.py) that catches missing backlinks — will correctly classify ADR-080 → ADR-078/079 as axis-only via comment heuristic.
  • ADR-078 (EXECUTIVE_SUMMARY absorbed into README): superseded on reviewer-URL-pin numeric axis only. EXECUTIVE_SUMMARY-absorption decision content unchanged.
  • ADR-079 (two-guide reader architecture): superseded on reviewer-URL-pin numeric axis only. Two-guide-architecture decision content unchanged.

Transcript

Audit + decision trail captured in the v1.3.1 audit-fix session (2026-05-22), driven by /exploring-options walk-through of 9 decision-shaping questions (Q1-Q9). Plan file: ~/.claude/plans/i-want-to-audit-abundant-meerkat.md. Per the gitignore-by-default transcript discipline, the transcript stays private; /save-transcript 2026-05-22__v1-3-1-audit-fix will land the file post-v1.3.1 close.