[Atom-oh]

Removes ADR references from intro-deck architecture slides 10-19: 02-architecture.md (source), 02-architecture.html (per-part rendering) and the matching slide and speaker-note lines of the combined index.html.

Merge order: #196, #197 and the architecture slides 10-19 PR are independent and can merge in any order (their index.html edits touch disjoint lines). Each carries its Markdown source block, its per-part HTML rendering and the matching lines of the combined index.html together, so no review sees a rendering diverge from its source; splitting this way keeps each review under the 128 KiB AI review input limit. #195 is superseded once all three merge.

🤖 Generated with Claude Code

[github-actions]

🤖 AI Code Review (two independent reviewers)

_Reviewer responses: codex/ALL claude/ALL _

Status: PASSED — No blocking issues found

Verification is complete — I have everything needed to synthesize the chair review.

Chair Review — PR #205: docs-site: remove ADR references from the deck architecture section (slides 10–19)

1. Summary

This is a documentation-only PR that strips ADR identifiers (ADR-002/005/007/009/011/013/015) from slides 10–19 of the public awsops-intro presentation deck, editing the source Markdown and both generated HTML bundles in lockstep. It continues the merged de-ADR-ification program established by the immediately preceding commit bb1c8d6 (PR #201, same treatment for other docs-site pages). No runtime code, Terraform, IAM, or query logic changes; all substantive governance semantics on the posture slides are preserved.

2. Issues per lens

L2 — Code correctness

Both models found no correctness issues; I concur. The claude panel specifically verified the only executable surface (the inline speaker-notes JS object literals in both HTML bundles): edited strings remain syntactically valid, escaped quotes/\n are preserved, deleted phrases were removed whole, and the +11 note-key offset between 02-architecture.html and index.html is respected in every edited key — I confirmed the edited note keys (11/15/16/18/19 ↔ 22/26/27/29/30) map to the correct slides.

• MINOR (pre-existing, claude only): literal unescaped && in HTML text (workers_enabled && diagnosis_notify_enabled) on lines this PR rewrites. Browsers tolerate it and it predates the diff. Not gating.

L3 — Security / AWS mutation safety

No mutation path, IAM change, or secret is introduced — both models agree. I verified the posture slides still state the freeze, its non-permanence, the relaxation gate (new decision document + multi-AI panel + dated owner approval), and every substantive limit of the single carve-out (one action ecs:UpdateService force-new-deployment, one ARN, secret-rotation trigger only, fail-closed, default-off, everything else frozen). These match the enforced reality in base (terraform/foundation/secret-rotation.tf, flag default false, single-action/single-ARN policy, per claude's check). No weakened control or bypass.

• Downgraded to MINOR — "carve-out loses its audit handle" (claude MAJOR, codex MAJOR — agreement, but on a disproven premise). Both panels flagged removing ADR-015 (2026-07-01 owner-override) from the exception callout and narration. Claude's load-bearing claim — "this deck was the only artifact here describing the carve-out's provenance" — is false: the full provenance (owner-override, 2026-07-01, exact scope) remains in root CLAUDE.md:113, AGENTS.md:127, docs/architecture.md:55,68, CHANGELOG.md:292, and README.md. Codex's claim that the rewrite "conflicts with the baseline requirement for a new ADR, multi-AI review, and dated approval" also doesn't hold: the rewritten note retains all three requirements, merely calling the ADR a "결정 문서" (decision document) — a reasonable paraphrase in a public deck for an ADR series that CLAUDE.md:109 says is private and cited "for traceability only". Removing citations from a presentation changes no enforcement artifact, matches the merged PR docs-site: remove ADR references from overview, monitoring, security and compute pages #201 direction, and incidentally stops naming a private individual (오준석님) in a public deck. The residual traceability loss within the deck itself is a MINOR consistency note, not a gate.

L4 — Observability / data-integration correctness

No findings from either model; I concur. Steampipe/Powerpipe/CIS/AgentCore content is untouched, and the one rewritten factual claim remains accurate: claude verified against scripts/v2/workers/diagnosis_digest.py that the digest really does batch manual runs beyond the approved scheduled-summary scope — the deck still honestly reports the known overrun rather than hiding it.

• MINOR (claude only, out of diff scope): the identical diagnosis_digest claim still carries the ADR-13 attribution in the demo section (03-demo-scenarios.md:899, 03-demo-scenarios.html:639, index.html:1332 — verified present in base and untouched by this PR), so the published bundle now states the same scope overrun twice, once with and once without a tracking reference. Those copies are in the demo section, outside this PR's stated slide range — excluded from the gate.

L5 — Docs/ADR consistency

• Downgraded to MINOR / excluded from gate — "sweep stops mid-deck" (claude MAJOR, codex did not flag). I verified slide boundaries against the base files: every ADR reference inside slides 10–19 — bodies, the two edited slide titles, and speaker notes 10–19 in both HTML bundles — is removed by this patch. All survivors claude cites (02-architecture.md:114–391, HTML lines 83–184 and notes 4–9, plus 01-*/03-* files) sit in slides 3–9 or other deck parts, outside the PR's explicitly scoped range ("slides 10–19"). The prompt's gate rules exclude out-of-diff-scope findings, and the predecessor PR docs-site: remove ADR references from overview, monitoring, security and compute pages #201 establishes this is deliberate incremental work. The interim mixed state of the single published index.html is real but transitional — a follow-up matter, not a defect of this diff.
• MINOR (claude only, out of scope): the PPTX generator (docs-site/scripts/pptx/build-awsops-intro-pptx.js) still tells the governance story in ADR terms and even glossaries "ADR"; per the presentation README the PPTX needs a refresh when web content materially changes. Worth folding into the follow-up sweep.
• MINOR (claude only, in scope): dropping the scoping noun leaves a slightly ambiguous antecedent — 나머지 ADR-005 전부는 그대로 동결 → 나머지는 전부 그대로 동결 can be misread as "the remaining exceptions". An explicit noun (e.g. "나머지 변경·자율 기능 전부") would be cleaner.
• No README freshness regression; md ↔ generated-HTML pairs stay faithful for every edited string (verified for the edited hunks).

3. Suggestions

1. In the follow-up sweep covering slides 1–9 and parts 01/03, decide whether the posture slide should keep a dateless-but-identifiable provenance hint for the single exception (e.g. "2026-07 소유자 승인") — the date, unlike the ADR number, is public-safe and preserves the deck's internal "dated owner approval" logic.
2. Update the two remaining ADR-13 copies of the diagnosis_digest scope-overrun claim (03-demo-scenarios.md:899, index.html:1332) in the same follow-up so the bundle states it consistently, and keep some tracking handle (issue link) for the overrun.
3. Include the PPTX generator in the de-ADR program and rebuild awsops-intro.pptx in the same commit, per the presentation README's sync requirement.
4. Opportunistic: escape the literal && as && next time those lines are regenerated.

4. Verdict

Both panels declared IMAGE_COVERAGE: NOT_REQUIRED; the manifest lists zero images, deletions, omissions, and unavailable entries with status: complete, so no HEAD pixel inspection was in scope for the chair either.

IMAGE_COVERAGE: NOT_REQUIRED

Both panel MAJORs were verified against the base tree and do not survive: the carve-out's provenance remains fully recorded in CLAUDE.md/AGENTS.md/docs/architecture.md/README.md/CHANGELOG.md, the rewritten slide retains every substantive governance requirement and exception limit, and the "incomplete sweep" consists entirely of out-of-scope slides consistent with the merged incremental program. No CRITICAL or MAJOR remains in the gate.

---

Triggered by commit 108e61b24a06a7085ac22ab04c509eddc4cb3e1e · workflow: .github/workflows/pr-review.yml