The diff adds only a staging YAML under proposals/intake/ plus a short descriptive note in docs/AUTOMATION.md; AGENTS.md is untouched and no gate, budget, or review rule is altered, so §3.15 check-weakening is not implicated. Each of the 39 migrated decisions carries exact provenance — a legacy commit hash (ab698546...), a file+index ref (e.g. proposals/watch/archive-ledger.yaml#items[18]), the original record verbatim, and duplicate triage copies where they exist — satisfying §3.8 rather than removing locators. Unknown metadata is left explicitly null (model, promptVersion, inputHash) instead of being invented, and the failed orch-or triage case is preserved honestly as decision: failed with the original parse-error message and retryable: true, which is the §3.11-style negative-record behavior one wants, not a fabricated record. The decision count in the diff is exactly 39, matching the PR body's claim, so no visible internal inconsistency. No content/ files or published assessments change and no citation identifiers were added under content/ (verification section empty), so no citation-existence or unratified-draft-as-ratified issue arises. The PR body's assertions about the other batches and CI are unverifiable here but are claims about work outside this diff, not violations within it.

The added packet preserves exact legacy provenance through a specific commit, file-and-row references, run IDs, dates, and verbatim legacy records, while retaining null fields rather than inventing missing metadata. It explicitly records the failed triage outcome as failed and retryable, and identifies duplicate promotion records without treating them as independent evidence. The change is confined to documentation and a staged proposal file; it neither alters published assessments nor presents this packet as ratified or authoritative over the original files. No AGENTS.md amendment, confidential material, or weakening of validation and ratification checks appears in the diff, and there is no visible positive evidence that any source or locator was fabricated.

The diff stages 39 historical intake and triage decisions into an intake ledger under proposals/intake/ along with descriptive documentation in docs/AUTOMATION.md. It does not alter AGENTS.md, modify content under content/, or weaken any verification gates or publication checks. Provenance to original legacy commit hashes and archive files is preserved, and no unratified drafts are presented as published assessments.

The diff only appends a documentation note to docs/AUTOMATION.md and adds one staging YAML under proposals/intake/ that re-records 39 already-committed watch/triage/promotion decisions (including one explicit failed triage) with their original titles, URLs/identifiers, reasons, runIds, and legacy commit locators at ab698546. It does not edit AGENTS.md, content/, assessments, or any runtime reader; the PR and doc text state the originals remain authoritative and no public judgment or gate is changed. Citation-verification tooling reported no new content/ identifiers to check, and nothing in the visible packet shows fabricated locators, silent judgment edits, confidence-material exposure, or check-weakening. This is an append-only historical staging packet consistent with §3.14–§3.15 and phase-1 proposal handling.

The diff is an additive staging packet under proposals/intake/ plus an honest documentation note; it edits no AGENTS.md, no CI gate, no validator, and no published assessment, so no check is weakened and nothing unratified is presented as ratified — it is explicitly labeled a review-packet staging copy with the cutover deferred until batches pass review. Provenance discipline is strong: each row carries a legacy commit locator (ab69854...), original file refs, embedded legacy records, unknown metadata left null rather than invented (§3.8), and the previously skipped failed triage case is preserved as an operational failure rather than papered over (§3.11 spirit). The citation-verification section reports no identifiers checked because the tool scopes to content/ and these sit under proposals/; the DOIs/arXiv IDs are internally consistent (titles plausibly match identifier subjects) and postdate my training, so per temporal humility their being unfamiliar is not evidence of fabrication, and these are records of archived/rejected watch items, not new evidentiary claims. The claim that rows exactly match the original ledgers cannot be confirmed from the diff alone, but the originals remain present and authoritative, making any mismatch detectable and non-destructive; nothing visible breaks a specific rule.