feat: incorporate modern-delivery ideas, judgment-scaled (ADR 0023; counter to #22)

[vpciii]

Summary

The methodology-consistent alternative to PR #22. Accept one or the other, not both — if you take this, close #22.

PR #22 proposed six new mandated practices and was BLOCKed by adversarial review (docs/reviews/pr22_adversarial_review.md) for duplicating existing practices/ADRs, contradicting standing exclusions, and inverting scale-to-work. This captures what's genuinely worth keeping, the methodology's way.

What it does

• Adds §12 Deterministic onboarding — one documented, reproducible setup command; "works on my machine" is a defect; the tool (Make / devcontainer / Nix / script) is a per-project choice, not a mandate. The one idea from feat: expand methodology with 6 modern delivery practices (ADR 0023) #22 with no existing home (running-system determinism is §6/§10; the developer's entry path wasn't stated). Tool-agnostic, judgment-scaled.
• Sharpens "Automate what is checkable" — formatting/style is machine-enforced, and if a machine settles it, review attends to behavior not style. Folds in the useful core of feat: expand methodology with 6 modern delivery practices (ADR 0023) #22's automated-gates + code-review ideas without a new practice.
• Adds nothing else — continuous refactoring (ADR 0014), docs-as-code (ADR 0020), review substance (ADR 0015), and CD (per-project; already implied by §4+§11) are covered. Re-adding would violate ADR 0022.

Self-consistency (what #22 missed)

Updated in this same PR: §1–§11 → §1–§12, the README practice count, and the global-CLAUDE.md / methodology.mdc summaries (ADR 0018). Cuts 0.12.0.

Provenance

This whole exchange — Gemini authors an over-reaching expansion → Claude adversary BLOCKs it → Claude drafts the judgment-scaled counter — is the cross-model review practice (trial round 8) doing exactly what it's for. Fair game to run the adversary on this PR too.

🤖 Generated with Claude Code

[claude]

Claude finished @vpciii's task in 3m 30s —— View job

---

Review: ADR 0023 — judgment-scaled delivery ideas (counter to #22)

• Read governing artifacts (methodology.md, ADR 0014/0015/0018/0020/0022, README, templates)
• Verify the PR's self-consistency claims
• Check cross-references and conventions
• Post review

I reviewed this against the methodology's own standards (ADR 0022 single-source, ADR 0018 summaries-in-same-PR, ADR 0005 navigability, the standing exclusions). The judgment call at the core is sound: §12 is the one genuinely-new, tool-agnostic gap (developer entry path, distinct from §6 config and §10 reproducible builds), and folding the review-attention nugget into "Automate what is checkable" rather than minting new practices is exactly the ADR 0022 discipline. The self-consistency work is real and verified — see below. One blocking defect, one optional note.

🔴 Dangling reference — docs/reviews/pr22_adversarial_review.md does not exist

ADR 0023 cites this file three times and the PR body links to it, but it is not in this branch (or anywhere on main):

• adr/0023-incorporate-delivery-ideas-judgment-scaled.md:9
• adr/0023-incorporate-delivery-ideas-judgment-scaled.md:20
• adr/0023-incorporate-delivery-ideas-judgment-scaled.md:102

$ ls docs/reviews/
pr18_adversarial_review.md  pr19_adversarial_review.md  pr21_adversarial_review.md

The established convention is the opposite: ADR 0022 cites docs/reviews/pr18_adversarial_review.md, and that file is committed to the repo (adr/0022:106,129). An ADR that grounds its entire Context and Decision on a BLOCK review should ship that review alongside it, per ADR 0005 (dual-audience navigability) and the "keep artifacts honest" rule.

This is sharpened by the PR's own premise: it asks reviewers to close #22. If #22 is closed unmerged, the pr22 review file lives only on that abandoned branch and these references become permanently dangling — the evidence the ADR rests on would never reach main.

Fix: commit docs/reviews/pr22_adversarial_review.md in this PR (same as pr18/19/21).

Fix this →

🟡 Optional — templates/project-CLAUDE.md not updated

The downstream-project hard-rules template (templates/project-CLAUDE.md) lists "Reversible by default", "Dependencies are decisions", etc., but not deterministic onboarding. It's explicitly "a pointer, not the methodology" and a curated subset (it omits several practices), so this isn't a strict ADR 0018 inconsistency — that ADR governs global-CLAUDE.md and methodology.mdc, which you did sync. Adding a one-line bullet there is a judgment call; mentioning so you've considered it, not as a blocker.

✅ Verified clean

• No stale practice references. §1–§11 → §1–§12 is complete (methodology.md:59), README count updated (README.md:26); grep finds no surviving §1–§11 outside ADR prose describing the change.
• ADR 0018 summaries synced in-PR — templates/global-CLAUDE.md and templates/methodology.mdc both carry the new §12 bullet and the sharpened automate-what's-checkable line.
• No tooling-mandate self-contradiction — §12 and the automate rule name formatting/style as machine-enforced but stay tool-agnostic (Make/devcontainer/Nix/script as examples, not mandates), consistent with the "no premature tooling" exclusion the ADR invokes against feat: expand methodology with 6 modern delivery practices (ADR 0023) #22.
• No ADR 0022 duplication — §12 differentiates itself from §6/§10; the review nugget sharpens an existing rule rather than adding a practice.
• SemVer + CHANGELOG — 0.11.0 → 0.12.0 minor bump (added practice) with compare links correct; Conventional Commit subject is well-formed.

Verdict: The judgment is on-point and the self-consistency is genuinely better than #22's. Ship the pr22 review file (blocking) and this is clean.
· feat/delivery-ideas-judgment-scaled

[vpciii]

Withdrawing. A round-9 adversarial review (docs/reviews/pr24_adversarial_review.md) BLOCKed this counter-proposal, and most of it holds: the 'automate what is checkable' refinement creates a second home for ADR 0015; §12 largely overlaps §6/§10 (a stricter standard than I applied to #22); CD was treated asymmetrically; and the ADR-number collision is real. (One finding — 'stale §11 references' — was a false positive: §12 was appended, not renumbered, so §11 references to the Reversibility practice remain valid.) Conceded. Resolution: status quo — neither #22 nor #24 lands.