Skip to content

feat: expand methodology with 6 modern delivery practices (ADR 0023)#22

Closed
vpciii wants to merge 2 commits into
mainfrom
feat/expand-methodology-practices
Closed

feat: expand methodology with 6 modern delivery practices (ADR 0023)#22
vpciii wants to merge 2 commits into
mainfrom
feat/expand-methodology-practices

Conversation

@vpciii

@vpciii vpciii commented Jun 21, 2026

Copy link
Copy Markdown
Owner

Expands the methodology to explicitly cover:

  • Deterministic Dev Environments
  • Automated Quality Gates
  • Continuous Deployment
  • Continuous Refactoring
  • Code Review Culture
  • Documentation as Code

Also includes ADR 0023 which documents this decision and updates the CHANGELOG.md.

@claude

claude Bot commented Jun 21, 2026

Copy link
Copy Markdown

Claude finished @vpciii's task in 2m 25s —— View job


Review: PR #22 — expand methodology with 6 modern delivery practices

Reviewed against this repo's own standards (the honesty rules in methodology.md, ADR 0018, ADR 0020, ADR 0022, the ADR template). The ADR is well-formed — clear context, real alternatives (including the "embed vs. number them" call), honest consequences, and an adoption-impact line that matches the CHANGELOG. The CHANGELOG entry is correct and lands under [Unreleased] as expected.

The substantive issues are all process drift of the "docs change in the same PR" / single-source kind — which is notable because this methodology polices exactly that.

🔴 Stale cross-references introduced by this PR (same-PR honesty rule, §"Keep the artifacts honest")

The PR grows the list from 11 to 17 practices but leaves two now-wrong counts behind:

  • methodology.md:59 — the reading order still says "the practices (§1–§11) and why they exist." It should read §1–§17. This is made stale by the very diff that introduces it. Fix this →
  • README.md:26 — the layout table describes methodology.md as "The canonical development methodology (11 practices + the reasoning + a decision guide)." README is not in this PR's diff, so it now contradicts the document. The same-PR rule binds it. Fix this →

🔴 Global summary not updated in the same PR (ADR 0018)

templates/global-CLAUDE.md is the canonical source of the always-loaded ~/.claude/CLAUDE.md summary. methodology.md's own "Changing this document" section and ADR 0018 require it to change in the same PR as any change it summarizes; its header comment repeats this. Several of the new practices are hard-rule material the summary currently omits — most clearly Automated Quality Gates (CI style enforcement) and Continuous Deployment. Either fold the summary-worthy ones into the "Hard rules" / decision-guide tables, or state explicitly in the ADR why these six are deliberately left out of the summary. As written, this is precisely the drift ADR 0018 (and the reviewer/CI gap ADR 0022) exist to catch. Fix this →

🟡 Single-source tension — new practices restate guidance that already lives elsewhere

The methodology insists "a fact lives in one place." ADR 0023's Option A rejection covers giving these their own headings, but three of the six largely re-state existing content rather than add new facts, creating future drift hazards:

  • §15 Continuous Refactoring overlaps the existing decision-guide "restructure without changing behavior" row and §5's test-as-safety-net framing.
  • §16 Code Review Culture overlaps the review-scope rules the methodology repeatedly defers to CONTRIBUTING.md (§2: "review-scope rules live in each project's CONTRIBUTING.md").
  • §17 Documentation as Code restates the "Keep the artifacts honest" section and ADR 0020 (docs ride along in the same PR).

Not a blocker, but worth a sentence in each new section cross-linking the canonical home (and ideally the relevant ADR), so these read as emphasis rather than a second copy that can drift.

🟡 §14 Continuous Deployment vs. the tool-agnostic ethos

§14 mandates "Merging to main triggers a pipeline that deploys to production." That's more prescriptive than neighbors and brushes against "No premature tech-stack or tooling commitments" / "document the artifact, not the tool." The "(or a staging equivalent)" hedge helps; consider framing it as the principle (automated, zero-touch releases; deploy readiness) and leaving the pipeline itself to a project ADR, consistent with how §6/§11 treat tooling.

⚪ Minor formatting (consistency with the rest of the doc)

  • New sections §12–§17 omit the blank line between the intro sentence and the bullet list that every existing practice uses (e.g. methodology.md:462→463). Harmless in most renderers but inconsistent.
  • Trailing whitespace at methodology.md:471 (…Black, or GoFmt. ).

(If a markdown linter runs in CI, ignore the two above — I have no evidence of one here.)


Summary: The ADR and CHANGELOG are solid and the new practices are reasonable additions. The blocking-ish items are all the methodology's own same-PR/single-source rules turned on itself: two stale practice counts (methodology.md:59, README.md:26) and the un-updated global summary (ADR 0018). Address those three and the content tensions are minor.

Reversed-role run: Gemini/Antigravity authored the 6-practice expansion
(ADR 0023); Claude adversary-reviewed it cold. BLOCK — the practices
duplicate existing ones (ADR 0014/0020 → violates ADR 0022), contradict
the "no premature tooling commitments" + lightweight-ops exclusions
(§13/§14), lack deployable-service scoping (§14 breaks on libs/CLIs/docs
repos), are framed "non-negotiable" against the scale-to-work stance, and
leave the doc internally inconsistent (stale "11 practices/§1–§11", unsynced
ADR-0018 summaries) with the ADR pre-declaring its own "Accepted" status.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@vpciii

vpciii commented Jun 21, 2026

Copy link
Copy Markdown
Owner Author

Withdrawing — status quo (the ADR's own Option B). An adversarial review BLOCKed this (docs/reviews/pr22_adversarial_review.md), and the methodology-consistent counter (#24) was then also BLOCKed for milder versions of the same sins. Run in both directions, the adversarial process showed the six-practice expansion over-reaches: most of it duplicates existing practices/ADRs, and the genuinely-new bits don't clear the bar for a numbered practice. No expansion lands.

@vpciii vpciii closed this Jun 21, 2026
vpciii added a commit that referenced this pull request Jun 21, 2026
Round 9: Gemini BLOCKed Claude's counter-proposal (PR #24) — and most of it
held (the refinement duplicates ADR 0015; §12 overlaps §6/§10; asymmetric CD
treatment; ADR-number collision). Claude conceded those and refuted one
finding as a false positive (the "stale §11 refs" claim — §12 was appended,
not renumbered). That false positive is the trial's first, and useful
calibration: a verifiable factual overreach, refutable by grep.

Both PR #22 (Gemini's expansion) and PR #24 (Claude's counter) are now
withdrawn — the methodology resisted expansion in both directions. Records
the adjudication (pr24_adjudication.md) and the new named failure mode
(adversary-authored-replacement bias).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
vpciii added a commit that referenced this pull request Jun 21, 2026
A PR #25 review (the same-model GitHub reviewer) correctly flagged that
the v0.12.0 §10 onboarding bullet is a normative practice extension that
shipped without an ADR — against the methodology's own "practice changes
get an ADR" rule, and unlike comparable additions (ADR 0014, 0016). ADR
0003 (which defines §10's bullets) is append-only, so ADR 0023 records
the new bullet and makes §10 traceable again.

Also trims the 0.12.0 changelog entry's process-journal backstory (the
#22/#24 narrative), keeping the substance + adoption line and citing
ADR 0023.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant