[xiaoan17]

Scope

One vocabulary gets a typed owner.

The TypeScript host owner authors an ordered provider plan and the Python
transport validates the plan it receives against the same sequence, but the two
sequences were private hand-kept copies:

• loopx/control_plane/turn_driver/host_todo_completion.ts — a module-local const STEP_KINDS
• loopx/control_plane/host_adapter_settlement.py — a module-local _STEP_KINDS

Neither was exported, so the semantic inventory could not see either carrier, and
neither loopx/semantics/vocabulary_v0.json nor the drift smoke could tie them
together. A divergence between them would have surfaced at runtime as
TypeScript host provider step order mismatch, after the plan had already been
accepted and partly executed. The C4.3 settlement vocabulary confirmation
recorded exactly this gap.

What changes

• Both sequences become exported owners of a new cross-runtime vocabulary,
  host_adapter_settlement_step_kind, so the drift smoke checks each of them
  against the registry on every run.
• The vocabulary is deliberately distinct from settlement_step_kind: that
  one names the turn settlement phases, whose validation phase is never
  executed, while every kind here is a command the host adapter runs. The
  registry meaning records that difference, so the two are no longer
  confusable.
• The two owners keep different names per runtime, so no new multi-value twin
  appears: multi_value_twins stays 8/8 and every inventory budget is
  unchanged. Only the coverage floor grows, 28 → 29 vocabularies and 54 → 56
  owner symbols, in the registry and its anchor in the same diff.
• New tests pin the two runtimes against each other directly, and prove the
  transport still refuses a plan whose step order drifts (that tripwire had no
  test).

Evidence

• examples/semantic-vocabulary-drift-smoke.py → ok, coverage=vocabularies:29/29, owner_symbols:56/56, multi_value_twins=8/8, conflicting_values=16/16 — no
  budget moved.
• The diff-scoped advisory now sees both carriers, which it did not before, and
  points each at the new vocabulary:
  reuse hint: host_adapter_settlement_step_kind (exact_values) plus
  settlement_step_kind (overlapping_values) — the C4.3 overlap is now a
  machine-visible hint instead of prose.
• Semantic inventory, production and drift test suites pass; the MCP
  settlement, host vision and machine guard suites pass; the TypeScript host
  todo completion test passes (15) and the control plane typechecks.
• loopx canary premerge --from-git-diff --git-diff-base 2a2323756 green
  (19 checks, 0 failures), surfaces control_plane, docs_project_content, public_boundary, python.

Coordination

#5677 extends this same vocabulary: it adds delivery_context and
direction_context to both carriers and rewrites the surrounding plan schema.
Whichever of the two PRs merges second must add those two values to the
registry entry (and its glossary row), because the registry values are what the
drift smoke checks the carriers against. This PR does not add them, because
they do not exist on this branch.

Unrelated observation

tests/architecture/test_semantic_production.py::test_real_return_producer_rejects_an_unregistered_result[turn_route-…driver.py-return LoopXTurnRoute.CONTRACT_ERROR]
already fails on main (2a2323756) — the registry still names a turn_route
return producer whose literal driver.py no longer carries. PR #7621 carries
that fix; this PR does not touch it.

Signed-off-by: xiaoan17 anbc1224@gmail.com

Rebase and re-validation (2026-10-29)

• Base: main at 9dcdc0bacd455a8cf6f6146f052ffc4df0f5d04a
• Head: 3781b77d1bfb3622820fde72568c76fb874d1ffd
• The change itself is unchanged; only the parent moved.

This entry is independent of the merge order between #7617 and #7645: it adds no ledger file and does not touch the derived STATUS.md index.

All nine pull requests in this set were rebased onto main at 9dcdc0bacd455a8cf6f6146f052ffc4df0f5d04a and merged together in a local worktree. The merged state passes tests/architecture/, tests/canary/, tests/test_manager_context_roundtrip.py, tests/test_lark_native_reply_context.py, tests/control_plane/test_host_adapter_settlement_step_vocabulary.py and tests/control_plane/test_host_adapter_settlement_partial_steps.py — 1805 passed, 0 failed. python scripts/generate_rfc_status_index.py --check is clean.