chore(knowledge): retire models-explained blog doc-queue entry after slice completion - #1816
Merged
Merged
Conversation
…slice completion The claude-models-explained blog slice completed the full pipeline: rendered-channel fetch (raw-md confirmed absent) with firecrawl extraction, 7-unit digest fan-out executing the run-7 pairing cross-link contract (a)-(g), dual verification with two correction rounds (dominant class: assertion-specific vendor-attestation under-application), REVERIFY2: PASS from both verifiers with no degraded fallback, and interview handoff (8 CAs + 8 OQs + pairing results, all 32 digest questions mapped). The queue entry leaves the profile per the remove-on-completion rule, taking its inlined pairing contract with it (executed; results live in the slice's handoff). The vendor-blog attestation rule the context-engineering slice's handoff recommended (OQ-3) is codified in the profile, matching the form both verifiers enforced on this slice: assertion-specific `vendor-claimed (blog, <fetch date> fetch)` markers with an illustrative assertion-class list. Knowledge plugin 0.10.8 -> 0.10.9 with changelog. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Claude finished @kyle-sexton's task in 1m 20s —— View job PR Review: chore(knowledge): retire models-explained blog doc-queue entry after slice completionTasks
ReviewThree files changed; all are documentation and configuration — no executable code in scope. Review focuses on correctness, consistency, and adherence to established conventions in this repo.
|
|
Claude finished @kyle-sexton's task in 1m 7s —— View job Security Review — PR #1816
FindingsNo security issues found. The three changed files are all documentation or plugin metadata; none introduce executable logic, permission changes, external trust delegations, or new data-egress surfaces. Detailed assessment below.
The sole change is
Two changelog entries describing the removed queue entry and the new attestation rule. No executable content.
Two logical changes:
The only URLs present in the post-PR profile are |
kyle-sexton
added a commit
that referenced
this pull request
Jul 31, 2026
…ion (#1842) ## Summary The extended-thinking platform doc slice (<https://platform.claude.com/docs/en/build-with-claude/thinking>) completed the full docpage-digest pipeline, so its queue entry leaves the profile per the remove-on-completion rule. Pipeline record: raw-md fetch (`.md` appended to the page URL; HTTP 200, no redirect) of a 917-line source, 13-unit digest fan-out, dual verification, one correction round of ten items (C1–C10), re-verified PASS by both verifiers, no degraded fallback at any stage, and an interview handoff carrying 2 cross-doc divergences, 9 candidate artifacts, 12 open questions, and all 41 digest open questions mapped (verified by a scripted 41/41 citation sweep). Verification chain: | Stage | Verifier | Verdict | |---|---|---| | Round 1 | A — same-vendor Claude, fresh context | PASS, 5 findings (all MINOR) | | Round 1 | B — cross-vendor Codex CLI 0.146.0, gpt-5.6-sol, per the #1740 recipe | FAIL, 9 findings (5 MAJOR, 4 MINOR) | | Re-verify | A | PASS — C1–C10 all RESOLVED, 1 MINOR residual | | Re-verify | B | PASS — no findings | The re-verify MINOR residual (user-specific routing-lane context in digest 06) was fixed by applying the already-ratified generic-reframe treatment from finding B-5 to the sibling occurrence; it is disclosed in the handoff's verification-state paragraph and recorded as a round-1 addendum rather than silently absorbed. ## Profile change A single deleted bullet. The paired `thinking-steering-and-cost` entry and the "Thinking" category heading deliberately REMAIN — that page is a separate concurrent run under the category's one-page-per-run contract, and its own queue PR removes both. Keeping this diff to one line makes it disjoint from the sibling run's diff, so the two merge in either order. Knowledge plugin 0.10.9 → 0.10.10 with a matching changelog entry. ## Merge sequencing (for the orchestrator) The concurrent sibling run bumps from this same 0.10.9 baseline, so both PRs claim **0.10.10** and will collide on `plugin.json` and `CHANGELOG.md`. Whichever merges second needs a rebase onto 0.10.11. The profile edits themselves do not conflict. Note: the "two overlapping docs — two runs, one page per run by contract" parenthetical above the queue bullets reads as transiently inaccurate between these two merges. That is the chosen sequencing, not a defect for this PR to repair; the sibling PR retires the heading and its parenthetical together. ## Cross-doc divergences surfaced (recorded for the interview, not acted on here) - **Terminology drift.** `code.claude.com/docs/en/model-config` describes interactive sessions receiving "redacted thinking blocks by default", while this page's `display` field takes `"summarized"`/`"omitted"` and documents `redacted_thinking` as a *distinct* safety mechanism — and states that Fable 5 / Mythos 5 return regular `thinking` blocks, "not `redacted_thinking`". Either the harness doc uses "redacted" loosely for `display: "omitted"`, or one page is stale. - **Harness-vs-API thinking-off exception gap.** The harness thinking-off exception list names only Fable 5, while the API rejects `disabled` at `xhigh`/`max` effort on "Claude Opus 5 and later models" with a per-request 400. Open whether the harness guards the combination or a user can hit repeated 400s from settings alone. No linked issue ## Related - #1816 — run 8 (models-explained blog), the preceding queue PR and the 0.10.9 baseline this bumps from - #1740 — the elevated Codex Verifier B recipe used for cross-vendor verification Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
kyle-sexton
added a commit
that referenced
this pull request
Jul 31, 2026
…ry (#1868) ## Summary The thinking-steering-and-cost platform doc slice completed, so its entry leaves the `docpage-digest` Anthropic profile's doc queue per the queue's remove-on-completion rule. The now-empty "Thinking" category heading goes with it — that category held two overlapping docs digested one page per run under its own contract, and the run-9 entry's PR (#1842) removed the first while anticipating this removal taking the heading. Knowledge bumps 0.10.10 → 0.10.11 so consumers receive the queue change. ## Verification Dual verification reached PASS on both arms against **identical frozen bytes** — a SHA-256 pin of all nine slice files, with the tree quiesced before either arm audited, per the frozen-tree rule this campaign adopted after a concurrent-correction race produced a spurious MAJOR. - **Arm B (cross-vendor, Codex)** — `VERDICT: PASS`, `FINDINGS: none`, six checks PASS, three round-1 findings RESOLVED. - **Arm A (same-vendor, fresh context, rationale withheld)** — `REVERIFY2: PASS`, six checks PASS, five prior findings RESOLVED, three MINORs, all record-keeping and none touching fidelity. Fidelity was swept mechanically: all 108 quoted spans across the six digests are exact substrings of the single source line each cites, 0 misses. Tag census 41 `cc-applicable` / 28 `mixed` / 12 `api-only` = 81 rows, exactly one vocabulary tag per row. The three MINORs were remediated after the verdicts without altering any claim row: two stale self-check footer records got dated "historical as of" annotations, and the missing round-2/3 applied record was written. The slice's `interview-handoff.md` was independently re-validated against the corrected digest bytes by a separate agent and its five stale items corrected. No degraded fallback was used at any point. Three findings from this run carry to the batched dispositions interview rather than this PR: the absence of any cross-digest citation-consistency check in the verification contract, the absence of a required applied-record artifact per correction round, and a genuine unharmonized cross-source tension about where a resolved effort value physically lives. ## Related - Follows #1842 (run 9, the paired thinking doc), which removed the first entry of this category. - Same campaign as #1816, #1788, #1779, #1775, #1765, #1761, #1756, #1752. No linked issue. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
kyle-sexton
added a commit
that referenced
this pull request
Aug 1, 2026
…e-digest queue (#1872) The `platform.claude.com/docs/en/resources/overview` slice ran end to end and reached dual verification, so its entry leaves the `docpage-digest` Anthropic doc queue under the queue's remove-on-completion rule. ## What ran - Raw-md fetch, per-unit digests (4 digests, 57 claim rows, 13 `api-only`), correction rounds, and Phase 5 interview handoff. - **Dual verification on identical SHA-256-pinned bytes**: verifier arm A and cross-vendor arm B both returned `VERDICT: PASS` with **no MAJOR findings**. - Arm B's coverage: all 57 claim rows and all 117 source lines examined; 31 distinct commands replayed with every count agreeing; all 174 `llms.txt` hits mechanically verified; 57/57 quote anchors, row/tag parity, stated tallies and placeholder scans all clean. Residual MINOR findings are disclosed in the slice's `interview-handoff.md` as Open questions for the batched dispositions interview, per the campaign's stopping rule. ## What this PR changes - Removes the resources-overview entry from the profile's Doc queue section. The **"Supplementary references" heading remains** — the system-prompts release-notes page sits under it as a separate concurrent run, and its own queue PR removes the heading once it empties. - Bumps the `knowledge` plugin `0.10.11` → `0.10.12`. - Adds the matching CHANGELOG entry, mirroring the 0.10.8–0.10.11 exemplars. No linked issue ## Related - Follows the same queue-drain pattern as #1868 (thinking-steering-and-cost), #1842, and #1816. - Two sibling run-11 slices (system-prompts release notes, verification-loops blog) follow in serialized PRs, since all three touch the same profile file. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
kyle-sexton
added a commit
that referenced
this pull request
Aug 1, 2026
…ocpage-digest queue (#1873) The `claude.com/blog/building-verification-loops-in-claude-code-with-skills` slice ran end to end and reached dual verification, so its entry leaves the `docpage-digest` Anthropic doc queue under the queue's remove-on-completion rule. It was the last entry under **"Blog posts"**, so that heading is removed too. ## What ran - Raw-md fetch, per-unit digests (6 digests, 78 claim rows, **0 `api-only`** — every row is positive-tagged), correction rounds, and Phase 5 interview handoff. - **Dual verification on identical SHA-256-pinned bytes**: arm A and cross-vendor arm B both returned `VERDICT: PASS` with **no MAJOR findings**. - Arm B's coverage: 78/78 rows examined including all 44 `vendor-claimed` rows; 7/7 pin hashes matched; 50 commands replayed with every count agreeing; 78/78 source quotes exact including ellipsis placement; a structural platform-scope sweep over 68 candidate paragraphs found no unscoped absence claim remaining. ## Why this slice took extra rounds It exercised the **vendor-blog attestation rule** (profile 0.10.9) at scale. The rule's predicate names two properties — an assertion exists only in the blog when *"no harness **or platform** doc states the same assertion"* — so a `code.claude.com`-only search never establishes it. All 44 `vendor-claimed` rows now carry a **targeted row-local `platform.claude.com` check**, using `curl -sSL` for pages outside the snapshot subset per DIGEST-BRIEF §5. Exhaustiveness language stays scoped to `code.claude.com`; the platform half is stated as the check performed, never as corpus-wide platform absence. Residual MINOR findings are disclosed in the slice's `interview-handoff.md` as Open questions for the batched dispositions interview. ## What this PR changes - Removes the blog entry and the now-empty "Blog posts" heading from the profile's Doc queue. - Bumps the `knowledge` plugin `0.10.12` → `0.10.13`. - Adds the matching CHANGELOG entry. No linked issue ## Related - Second of three serialized run-11 queue PRs; follows #1872 (resources overview). All three touch the same profile file, so they merge one at a time. - Same queue-drain pattern as #1868, #1842, #1816. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
kyle-sexton
added a commit
that referenced
this pull request
Aug 1, 2026
…ge-digest queue (#1874) The `platform.claude.com/docs/en/release-notes/system-prompts` slice completed. It is the **last entry in the doc queue** — after this merges, the queue section holds only the deferred task-budgets trigger entry, which was never queued. ## What ran The largest slice this pipeline has processed: **18 digests over a 2,548-line source, 659 claim rows, 561 `api-only`.** - Seven correction rounds and five dual-verification rounds, each on SHA-256-pinned bytes with the frozen-tree rule enforced — both arms re-hash before *and* after auditing, and no correction runs while a verifier holds the tree. - Final state: both arms `VERDICT: PASS`, with a confirming round after the last corrections. - Final gates: **1,895 recorded command/count pairs replay with 0 mismatches**, 0 `<corpus>` placeholders, 659/659 rows parse with `N. Quote (line` count equal to `- Tag:` count in all 18 digests, every stated tally equal to its anchored `grep -c`, `check-quotes.py` clean. ## What the run found worth carrying forward - **Narrow-slug false absences** remained the dominant defect — a basis testing one literal while the corpus documents the same concept in other words. Rows now disclose near-misses by name and line, each produced by a command recorded in that row. - **Prose is unparsed by every gate.** Summary paragraphs, Implications bullets and Open questions carried stale counts and false statements across several rounds; two sweeps (one whitespace-normalized with an offset map back to physical lines) closed the class. - **A contested tag-ordering question is escalated, not decided** — whether selecting a positive tag requires first establishing harness applicability. It is written up as a candidate profile amendment with both readings, the evidence, and a recommendation, for the dispositions interview. Residual MINOR findings are disclosed in the slice's `interview-handoff.md` as Open questions, per the run's stopping rule. ## What this PR changes - Removes the system-prompts entry and the now-empty "Supplementary references" heading from the profile's Doc queue. - Bumps the `knowledge` plugin `0.10.13` → `0.10.14`. - Adds the matching CHANGELOG entry. No linked issue ## Related - Third and final of the serialized run-11 queue PRs, after #1872 (resources overview) and #1873 (verification-loops blog). All three touch the same profile file, so they merged one at a time. - Completes the queue drain begun in #1752 and continued through #1756, #1761, #1765, #1775, #1779, #1788, #1816, #1842 and #1868. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
The claude-models-explained blog slice completed the full docpage-digest pipeline: rendered-channel fetch (raw-md confirmed absent per the profile's blog-post channel note) with firecrawl extraction, 7-unit digest fan-out executing the run-7 pairing cross-link contract (a)–(g), dual verification (Verifier A fable-5 + cross-vendor Codex per the #1740 recipe, two correction rounds, final REVERIFY2: PASS from both, no degraded fallback), and interview handoff (pairing results + 8 CAs + 8 OQs, all 32 digest questions mapped).
Profile changes:
vendor-claimed (blog, <fetch date> fetch)markers, illustrative assertion-class list — the rule the context-engineering slice's handoff recommended (its OQ-3), enforced end-to-end by both verifiers on this slice.Knowledge plugin 0.10.8 → 0.10.9 with changelog.
No linked issue
Related