Post-sweep cleanups: re-home spice pair under test/claim/ (first EX-row relocation dissolution); fix runner roster paths; link orphaned design docs

[gunbai-bot]

What (post-sweep follow-ups, operator-directed — scope narrowed once mid-flight, see item 1)

1. claim_witness_corpus_ci_runner.dag — deletion WITHDRAWN, paths repaired instead. This PR originally deleted the roster as a zero-consumer parallel representation (581 lines, no importers, stale path since birth). Mid-review, general body producer: ingested if-body fn_decl → Arrow, green by execution (Wave 1 gate 1) #6558 (general-body-producer lane) actively curated it — added two pick-keystone rows and fixed a stale path — disproving the abandoned-premise. Kept per that evidence; this PR now just repairs its three spice rows to the relocated paths below, including the original missing-_test bug at the passive-deck row. Whether an unexecuted roster should exist at all belongs to that lane's conversation, not a unilateral delete here.
2. Re-home the spice witness pair (spice_rc_passive_deck_claims_test.dag, spice_rc_ngspice_oracle_test.dag) from src/v2/extdeps/ to src/v2/test/claim/extdeps/ — the honest dissolution the sweep note named: they leave the strict-resolve probe universe together with their fixtures, so their two EX rows delete per the note's own trigger ('a row that excludes nothing is dead weight'). First relocation-class dissolution, receipted in the note. They remain floor-discovered at the new path.
3. Link two orphaned design docs — shell-to-dag-residual-census-and-arc-completion.md (ShellProgram -> DAG (Wave 1-3, on 1) #6514) and dependency-fidelity-design.md landed without links from any doc-graph root, redding doc_graph_has_no_orphan_docs on the cold sweep. The census link lives in the .dag authority dag/gunbc/plans/shell_emission_model.dag (first attempt edited the generated .md projection; the drift gate refused it — working as designed); the dependency-fidelity link landed independently via Fix orphaned dependency-fidelity design doc (latent fleet-wide floor red) #6559. Masking-class receipts 5 and 6: both offending PRs predict-skipped the live-read doc-reachability witnesses (DIVERGENCE … class=node-frontier recorded) — routed to the live-read classification design lane (keen-ram-460).

Verification (by execution, post-merge-resolve)

• Drift gate run_generated_artifact_drift_gate_body: PASS (committed .md ≡ authority, byte-for-byte via main_wet).
• doc_graph_has_no_orphan_docs: PASS both trees.
• All 3 relocated spice witnesses: claim_batch PASS at the new path; present + PASS in the cold corpus run.
• Whole-tree strict-resolve probe (18-row EX list): green, 1,105 modules, EXIT=0.
• Full cold whole-corpus falsifier: EXIT=0, all witnesses PASS (pre-runner-restore tree; the runner is data-only — no witness surface).

🤖 Generated with Claude Code

[gunbai-bot]

CI red root-caused and fixed: docs/plans/shell-emission-model.md is a GENERATED projection of dag/gunbc/plans/shell_emission_model.dag — my orphan-doc link edit went into the projection instead of the authority, and generated_artifact_drift_gate_passes correctly refused the drift (the gate working exactly as designed). Fix: the link now lives in the .dag authority; main_wet regeneration confirms the committed .md matches it byte-for-byte, and both the drift gate and doc-reachability witnesses pass locally. — sent from sleek-deer-172