Register the pending model routing layer for its covering sweep - #1126

Superseded by the CC-approved split (2026-10-15 01:12Z). This combined PR is closed. The rules-only prerequisite is #1130, head 5277796785096286d86c83d2b2dc0ad1ace3fddd, with no catalog or verdict-row/wave edits. The separate model-routes registration successor is held until #1130 lands, then will be rebuilt on current main with the exact CC scope and no candidates/winners; its actual PR link will be added here after it exists. The durable registration snapshot is 0f9e90b1a31f10c213857b36a6d4aa5d5ee4290a. This chooses the close-and-replace path rather than rewriting the combined branch or weakening the base-trust gate.

Scope

Register foundation/model-routes with the CC's exact provider/model/account-pool/effort scope so a covering sweep can start after review and landing. The row remains pending_lanes with no candidates or winners; the validator now admits that explicitly unsealed, gap-bearing registration without inventing a selection.

• Base commit: a339425ca56ec3fb87dff0fa8ad5dffc83354168 (origin/main).
• Lane: lane:foundation.
• Owned paths: foundation taxonomy, catalog/research/quality coverage, the narrow pending-empty validator path and regression, derived matrix/final-stack outputs and evidence registration.
• The sweep runs with --slots 2 after this PR lands and the CC read is complete. This PR dispatches no model, installs no runtime, changes no gateway/account state and makes no terminal claim.

SOTA sources

• Native reference at the base: scripts/landscape.py:854-859, pending-lane empty verdict containers; build_landscape:1264-1318 and quality coverage:1556-1578.
• Record v2 convergence angles with FINAL admission disabled #1058 original head 252babe4040e39173cc178317c67699371806df6, catalogs/landscape/final-stack.json foundation/model-routes entry and catalogs/landscape/angles/foundation-model-routes.json: retained unswept, angle-only source review with no catalog winner.
• docs/decisions/2026-10-11-layer-resolution-protocol.md, Decision and Terminal states: an absent verdict/sweep is an explicit gap, and pending selection does not qualify as terminal.
• CC FINALIZE-ALL plan SHA256 5692d4bfb6f30dbee76d79c45733c3ac2236f995bb690e69edfe40caa851562c; ROWS SHA256 872e4b9b748b4c0967465ac08c90ead44e9f8f6c8d7455d2825e4c335c902c1a; CC D1/F5 and 00:29Z pending-empty ruling retained in the lane questions report.
• python/cpython v3.13.16, Lib/unittest/case.py assertRaises/assertEqual: existing regression harness, matching installed Python.

Evidence-class table

Local commands run

The final outcomes are recorded in the lane R blocks:

python3 -m unittest tests.test_pending_catalog_registration.PendingRegistrationTests tests.test_landscape -v
98 tests passed
python3 scripts/landscape.py
passed: 21 foundation layers / 33 catalog rows, no candidate added
python3 scripts/component_matrix.py --write
33 rows, no platform flip violations
python3 scripts/final_stack.py --write
python3 scripts/final_stack.py --check
python3 scripts/validate.py

Final python3 scripts/validate.py exits 1 with ONLY the nine inherited model-currency date-rollover lines (nemotron embed, qmd reranker/query expansion, Potion, MinerU, the Kompress pair, Magika and background-haiku). The co-op confirmed this CI-wide incident; #1124 supplies its separately owned fix. No own file/hash/coverage errors remain and no currency-age or scanner bypass is added. Native landscape and final-stack checks pass; the final 44681-byte staged diff passed redacted gitleaks with zero findings.

Decision record

docs/decisions/2026-10-15-model-routes-catalog-registration.md. The scope text is verbatim, the older native-clients sentence is explicitly superseded, and the readiness input is review plus landing before the paced covering sweep.

Host evidence

No evidence/hosts receipt or platform-status change.

Checklist

• No workflows, secrets, billing surface or candidate installation added.
• No selected placeholder candidate, winner or frozen comparison invented.
• Peer worktrees and the separately held F1 branch preserved.
• CC reads this scope PR before its covering sweep.

Review correction at 0f9e90b

Fix round 1 addresses all three local-review P1 findings from reviewed head 5da12bca3a90237c2f4155814c05b2498c333062 (review SHA256 5a03c17995ca4684322d00b452355390c1d3e2deaea92fece93736eb8af49764), in one commit:

• The foundation validator recognizes 21 layers. A marked taxonomy registration requires an exact matching unrun catalog scope, and cannot borrow a capability decision.
• Explicit registration_only scope has no wave. The shared reader returns null; frozen-wave output excludes it. The 20260922 wave, registry and handbook remain byte-identical. Final-stack has wave: null and wave_doc: null.
• The new-host grand list admits a missing sweep row only for the marked empty pending matrix state, retaining an explicit unswept gap in JSON and Markdown. Ordinary/selected missing sweep coverage still refuses.

Each finding has a regression executed before its fix: foundation 20 required layers refusal, wrong 20260922 wave identity, and missing report layers: model-routes. After the fix, 195 affected consumer/landscape tests and 31 final-stack repository/status/registration tests pass. The three originally failing required commands now pass, along with standalone landscape, matrix, final-stack, catalog validation and validate_convergence.py --all-recorded.

Full validate.py was run again before this commit and fails with ONLY the same nine inherited model-currency lines tracked by #1124. No own integrity failure is hidden by that incident; the individual required consumer checks were run directly. Final 36632-byte staged diff SHA256 2eb7bda1a175d3e8f94bfa8a8cae118e8d7d8dc40c89ce5a54168f76bbf6a601 passed redacted gitleaks idle-class, with no findings. The exact CC scope and empty candidates/winners remain unchanged.

Additional pinned mechanism sources at the reviewed head: validate_foundation.py:35-42, its taxonomy/capability coverage at :88-94 and :216; build_verdicts.py:164-170 and frozen-wave checks :467-486; new_host_grand_list.py:91-151. The dated decision records the correction and boundaries; no upstream E2E/adoption or sweep result is claimed.