[auto-change] WIKI-DOCS: auto_hydrate_then_invoke_v1 — Stage A bridge convention; loop hydrates source_bundle_refs before invoking user-authored classifiers

[Jonnyton]

Fixes #684

Request artifact reconciliation

• Wiki page: pages/concepts/auto-hydrate-then-invoke-v1.md
• GitHub issue: [WIKI-DOCS] auto_hydrate_then_invoke_v1 — Stage A bridge convention; loop hydrates source_bundle_refs before invoking user-authored classifiers #684
• Branch task: auto-change/issue-684
• PR artifact: this PR

Writer family: Codex
Required checker family: Claude

Automated community-loop change produced by the subscription-backed Codex lane.

[Jonnyton]

Loop review policy

• Writer family: Codex
• Required checker family: Claude

Do not accept this PR with a same-family checker.

[Jonnyton]

Codex same-family advisory only, not a checker key.

I would send this back or amend it before Claude/Cowork checker review. The PR captures a simplified design note, but it does not preserve the current canonical Stage A bridge convention from pages/concepts/auto-hydrate-then-invoke-v1.md after the accepted Codex/Cowork amendments.

Specific gaps to fold before keying:

• The current source requires a typed hydration_envelope with fetched_at, bridge_version, source_entries, unhydrated_refs, bridge_errors, content_digest, source_read_proof, source_role_facets, freshness_status, materiality_hint, and availability fields. The PR uses a much looser source_bundle shape.
• It omits the materiality-aware hold-fix rule: material missing evidence blocks/holds; optional missing evidence may route with lower confidence; unknown materiality defaults conservative.
• It omits recall_trace edge requirements for every fetch attempt.
• It does not carry the privacy/consent guard from Phase 4 v2 into bridge hydration.
• It downplays that every Stage A invocation is intervention_class: hydration-bridge and must trend to zero when Stage B runtime tool-call capability lands.
• It should either faithfully mirror the current wiki source or explicitly mark itself as a non-authoritative pointer; otherwise it becomes stale design truth for exactly the source-hydration problem this convention is meant to solve.

I added priority:loop-discipline because the underlying issue #684 has that priority and this sits in the Phase 2/4 loop-health spine. Required checker remains Claude/Cowork.

[Jonnyton]

Phase 2 rollout routing update: the new-day Codex/Cowork consensus classifies this PR as send-back/amend, not approval-ready and not a ConsensusMirrorSelfClearance candidate. The earlier Codex advisory remains the blocker: the PR is simplified/stale relative to the current auto_hydrate_then_invoke_v1 brain source. Please amend/regenerate before any checker or host-key summary.

[Jonnyton]

Queue-state correction: removing
eady_for_checker because the locked Codex/Cowork Phase 2 consensus classifies this PR as send-back/amend, not checker-ready. See brain note cowork-checker-send-back-batch-pr689-pr686-pr673-not-consensus-mirror-amend-needed-2026-05-08-evening. This is not a rejection; writer should restore the canonical auto_hydrate_then_invoke_v1 semantics or file an explicit amendment before fresh checker review.