@@ -122,5 +122,6 @@ closed; no further implementation slice in it is admitted.
 
 [Routing-fact ergonomics](routing-fact-ergonomics/plan.md) owns the separately admitted
 focused input improvement, its consumer cutover and actual-agent qualification.
-The accepted six-slice refactoring remains closed. Review-evidence packaging and
-broader test strengthening are later, separately admitted work.
+It is Accepted after locked CI, independent review and connected interface-45
+qualification. The accepted six-slice refactoring remains closed. Review-evidence
+packaging and broader test strengthening are later, separately admitted work.
@@ -148,3 +148,31 @@ in `reports/r2-live-agent.md`. These disposable client runs do not substitute fo
 the actual connected-session check: that connection still reports interface 44 and
 `restart-and-reconnect`. Keep R-A7 and overall acceptance pending until the host
 refreshes the catalog and the connected tools exercise the changed surface.
+
+## R2 acceptance and closure
+
+After the session restart, both the connected native tool declarations and
+`runtime_info` reported interface 45, with `installation_state: current` and
+on-demand output delivery unchanged. Against one actual snapshot, the agent
+constructed a compact route from known values, answered the unresolved boundary
+question, and used the returned content-page and full-explanation requests
+directly. Three invalid calls returned bounded typed feedback; explicit unknown
+and a justified empty enum set retained their distinct meanings. Other unanswered
+facts remained unanswered. Production has no alias, boolean or nullable facts,
+so those cases retain the disposable-fixture CI evidence without fabricated live
+assertions. The [R2 live report](reports/r2-live-agent.md) records the connected
+observations separately from the paired fresh-model comparison.
+
+The acceptance owner reconciled R-A1–R-A8: exact-source locked Python 3.12 CI
+run `36465019273` covers the integrated production candidate `09d7829d` through
+HEAD `39d44f00` (tree `ca0e6c21`); independent review found no blocker, and
+RF-R1/RF-R2 have explicit non-blocking dispositions. The evidence and manifest
+commits `56303b79` and `e305c760` changed only documentation and the generated
+repository-index digest. The final acceptance records likewise change no
+production source or output-delivery configuration. The complete structural
+checkpoint passed 73/73 after regenerating the suite-input manifest for the
+updated plan-index digest; manifest freshness passed. An identical full Engine
+campaign is not required for documentation-only changes. R1 and R2
+are Accepted, the plan acceptance status is satisfied, and no further slice is
+admitted under this plan. Review-evidence packaging and broader test strengthening
+remain separate priorities requiring new admission.
@@ -64,6 +64,11 @@ blocker. The acceptance owner dispositions its two findings as non-blocking:
   and the reviewer found no reachable trigger; RF-10 and cold-transport tests
   exercise the known risk. Defer until a concrete valid-workflow failure exists.
 
-The source-review recommendation is preserved as evidence for R-A8, not a final
-acceptance verdict. [Connected-agent qualification](reports/r2-live-agent.md)
-remains the open gate; exact-head locked CI has since passed separately.
+The source-review recommendation remains reviewer evidence, not unilateral
+acceptance authority. The owner accepted R-A8 after the separate exact-head CI
+and [connected-agent qualification](reports/r2-live-agent.md) passed. Neither
+advisory opens another implementation slice.
+
+The production fact registry has no aliases, boolean facts or nullable facts.
+Their negative and state distinctions remain covered by disposable-fixture tests
+in the supported CI. They were not fabricated in the live production snapshot.
@@ -1,12 +1,12 @@
 # Routing-Fact Ergonomics
 
-**Plan status:** `Verifying`
+**Plan status:** `Accepted`
 
-**Current phase:** R2 exact-head CI and source review complete; connected interface-45 qualification pending
+**Current phase:** R2 accepted; no active implementation or integration slice
 
-**Next slice:** **R2 — Live agent comparison and accepted cutover**
+**Next slice:** `none`
 
-**Acceptance status:** `partial`
+**Acceptance status:** `satisfied`
 
 **Composed-design review:** `applicable`; see [the composed-design admission](#simplicity-and-ownership-review).
 
@@ -17,8 +17,8 @@
 **Admission:** this plan owns priority 2 of the user's new sequence **2 → 4 → 3**:
 routing-fact ergonomics, review-evidence packaging, then broader test strengthening.
 The prior six-slice refactoring is Accepted and closed. This plan does not reopen it.
-The user admitted `start` against this path; R1 is now implemented. The next
-admission is `verify` for the R2/integration evidence after checking current state.
+The user admitted `start` and then `verify` against this path. R1 and R2 are
+accepted; later priorities require separate admission.
 
 ## Objective
 
@@ -89,12 +89,12 @@ Analysis, native queries and stored contracts. It is not a retired renderer.
 | --- | --- | --- | --- | --- | --- | --- |
 | R-A1 | The new focused input and output fact grammar is explicit, generated, discoverable, and has no old-envelope fallback. | contract | supported locked runtime | automated | satisfied | [Verification and exact-head locked CI](reports/verification.md) |
 | R-A2 | Equivalent supported inputs produce identical canonical fact bindings, rule truth/unknown sets, selected dependencies, ordering and qualified content. | focused + differential | supported locked runtime | automated | satisfied | [Verification and exact-head locked CI](reports/verification.md) |
-| R-A3 | Static and snapshot-dependent failures are bounded, useful typed input rejections; no invalid value becomes success, a transport exception or fabricated absence. | contract + negative | both MCP purposes | automated | satisfied | [Verification and remaining environment/host/review claims](reports/verification.md) |
-| R-A4 | Fact binding uses the selected snapshot after qualification; content and full-explanation continuations round-trip that binding without additional capture or compilation. | integration | real SQLite and replacement stdio | automated | satisfied | [Verification and remaining environment/host/review claims](reports/verification.md) |
-| R-A5 | Native query/preview, evidence-backed Analysis, authoring, stored identities/handles and prior retained workflows preserve their contracts and results. | compatibility/retention | baseline producer and candidate consumers | automated | satisfied | [Verification and remaining environment/host/review claims](reports/verification.md) |
+| R-A3 | Static and snapshot-dependent failures are bounded, useful typed input rejections; no invalid value becomes success, a transport exception or fabricated absence. | contract + negative | both MCP purposes | automated | satisfied | [Verification](reports/verification.md) |
+| R-A4 | Fact binding uses the selected snapshot after qualification; content and full-explanation continuations round-trip that binding without additional capture or compilation. | integration | real SQLite and replacement stdio | automated | satisfied | [Verification](reports/verification.md) |
+| R-A5 | Native query/preview, evidence-backed Analysis, authoring, stored identities/handles and prior retained workflows preserve their contracts and results. | compatibility/retention | baseline producer and candidate consumers | automated | satisfied | [Verification](reports/verification.md) |
 | R-A6 | All actual affected producers, consumers, generated examples and live guidance are coordinated; complete required CI and structural checks pass on the relevant material. | integration | current hash-locked CI | automated | satisfied | [Verification and exact-head locked CI](reports/verification.md) |
-| R-A7 | The supported live agent can construct and refine routing facts with the smaller format, use discovery as needed and preserve unknowns, without new repair loops or a required extra call. | user workflow | actual authorized client/model/surface | observed | pending | [Verification and remaining environment/host/review claims](reports/verification.md) |
-| R-A8 | Independent review and acceptance-owner disposition establish the exact candidate's invariants and proportionality; unrelated work and previous acceptance are preserved. | review | independent reviewer and owner | manual | pending | [Verification and remaining environment/host/review claims](reports/verification.md) |
+| R-A7 | The supported live agent can construct and refine routing facts with the smaller format, use discovery as needed and preserve unknowns, without new repair loops or a required extra call. | user workflow | actual authorized client/model/surface | observed | satisfied | [Connected interface-45 navigation and paired comparison](reports/r2-live-agent.md) |
+| R-A8 | Independent review and acceptance-owner disposition establish the exact candidate's invariants and proportionality; unrelated work and previous acceptance are preserved. | review | independent reviewer and owner | manual | satisfied | [External review](reports/external-review.md), [owner findings](issues.md), [final verification](reports/verification.md) |
 
 No production standards proposal or publication is necessary for the navigation
 workflow qualification. Required existing authoring regressions and retained-state
@@ -308,7 +308,7 @@ its snapshot-dependent binding and multiple continuation producers.
 
 ### R1 — Coordinated focused contract and consumers
 
-**Status:** `Implemented`
+**Status:** `Accepted`
 
 **Goal:** implement the single smaller focused route representation with semantic
 and authority equivalence and useful error behavior.
@@ -341,7 +341,7 @@ current locked CI and generated freshness. Old focused input must fail usefully.
 
 ### R2 — Live agent comparison and accepted cutover
 
-**Status:** `Planned`; depends on the coherent R1 candidate.
+**Status:** `Accepted`; completed against the coherent R1 candidate.
 
 **Goal:** establish the externally meaningful reduction in construction burden and
 complete review/acceptance without a new feature slice.
@@ -393,10 +393,8 @@ priority 3 only defers broader unrelated test hardening.
 
 ## Blockers
 
-The source candidate is locally verified. Supported-runtime CI and actual host/model
-behavior and independent review remain unqualified; their owner and procedure are
-R2/integration. The missing environment cannot be
-silently marked passed. Source and static planning checks do not claim R-A7.
+No acceptance blocker remains. The supported-runtime CI, independent review and
+actual connected interface-45 agent check are recorded under their own evidence.
 
 ## Re-Plan Triggers
 
@@ -411,17 +409,18 @@ meaning and uncertainty rather than shipping a permissive fallback.
 
 ## Final Acceptance
 
-Source implementation and local verification are complete. R-A3–R-A5 have the
-requested local/real-process evidence. R-A1/R-A2/R-A6 still require supported locked
-CI; R-A7/R-A8 require actual host/model evidence and independent review/owner
-disposition. See the verification report rather than substituting baseline CI or
-planning probes. Preserve original failures and repaired candidate evidence.
-Use ordinary repository branching/commit/review rules; no prescribed commit count,
-retrospective PR or history rewrite is part of this plan.
+R-A1–R-A8 are satisfied. Exact-source Python 3.12 locked CI covers the integrated
+production candidate; the independent source reviewer found no blocker; the owner
+deferred two advisories with triggers; and the refreshed connected agent completed
+construction, refinement, continuation and supported error cases on interface 45.
+The [verification](reports/verification.md), [live-agent record](reports/r2-live-agent.md),
+[review](reports/external-review.md), [issues](issues.md) and [ledger](execution-ledger.md)
+retain the separate evidence. No authoring boundary changed, so no new proposal
+lifecycle run was required. The earlier accepted six-slice sequence remains closed.
 
 After R1/R2 acceptance, the user's next priority is **review-evidence packaging**;
 then **targeted test strengthening**. Neither is implemented or admitted as a source
 slice here, and neither reopens the closed six-slice refactoring.
 
-- Acceptance status: `partial`
-- Final status: `Verifying`
+- Acceptance status: `satisfied`
+- Final status: `Accepted`
@@ -1,12 +1,30 @@
 # R2 connected-agent qualification
 
-**State (2026-09-28): incomplete.** The connected authoring-purpose Codex tool catalog still presents interface 44. A direct `runtime_info` call returned instance `acd7ec54-b009-41a3-a5ee-62b2110681a9`, interface `44`, catalog digest `sha256:feedb6e08686a2b2036548c41c05459a7e72c4fd9d2cd217ecc22bf18b6c3d17`, `installation_state: restart-required` and `action: restart-and-reconnect`. The saved registration selects the repository source, purpose `authoring`, and `--output-schemas on-demand`; it has no new routing flag. No store, handle or registration argument was changed.
+**State (2026-09-28): completed after session restart.** The connected authoring-purpose Codex catalog and `runtime_info` both report interface **45**. The live instance is `7b218cb0-a70a-4325-934c-743b3f312e96`, catalog digest `sha256:8235cb21f7d1937be1a003eb2eef570345242194dac36d79463b16e26bb6d65d`, `installation_state: current`, `action: reuse`. The saved registration still selects the repository source, purpose `authoring`, and `--output-schemas on-demand`; it has no new routing flag. No store, handle or registration argument was changed.
+
+## Initial stale connection (historical observation)
+
+Before restart, the connected catalog still presented interface 44. `runtime_info` returned instance `acd7ec54-b009-41a3-a5ee-62b2110681a9`, catalog digest `sha256:feedb6e08686a2b2036548c41c05459a7e72c4fd9d2cd217ecc22bf18b6c3d17`, `installation_state: restart-required` and `action: restart-and-reconnect`.
 
 Through that same connected tool, `routing_facts` returned a real snapshot handle and eight registered enum-set categories. For the representative review/qualification task, known facts were activities `verification, documentation`, boundaries `generated-contract, ipc`, topics `contracts, diagnostics, performance`, and details `topic.contracts.schemas, workflow.verification.oracles`. Other categories remained omitted and unknown. The baseline envelope request was 563 JSON bytes including the returned snapshot and `content.limit=2`. The old process returned the transport error `Engine invocation failed; inspect server stderr` for `route`, so it supplied no valid baseline route or continuation. This failure is not evidence about the interface-45 process. No application or standards mutation occurred.
 
-The candidate connection has not yet been refreshed in this agent session. Interface-45 construction, refinement, continuations and mistake feedback through this connected tool remain unobserved, so R-A7 is still open. The separate paired model comparison below does not replace that connected-agent check. Keep the plan `Verifying` until the actual catalog reports 45 and the live workflow is observed. Existing automated disposable-fixture coverage of boolean, nullable and alias cases remains separate from the production vocabulary, which has none of those definitions.
+The paired model comparison below is separate from the connected-session result.
+
+## Actual connected interface-45 navigation and mistakes
+
+`routing_facts` returned snapshot `snapshot:v1:00d9a08b-4d75-4266-a00c-e172d95df3c1` and eight enum-set definitions. The agent supplied known verification, documentation and commit activities, library application form, contracts/diagnostics/performance topics, and schema/oracle details directly. It left boundaries, frameworks, languages and workflow profile unanswered. The first compact route succeeded with 12 reading entries and four unresolved questions; no `describe_input` call was needed. The boundary question was then answered with `generated-contract` and `ipc` against the **same snapshot**. The refined route succeeded with 18 reading entries; frameworks, languages and workflow profile remained unanswered rather than becoming empty or absent.
+
+The returned `content.next` request and the returned full `explanation` request were each passed directly to `route` without rebuilding their facts. Both succeeded, retained the same snapshot and compact fact values, and retained the three unanswered questions. The content page returned one exact item at offset one and a further continuation. A separate explicit-unknown enum-set assertion returned `{"state":"unknown"}` in the echo and kept that fact unanswered. A justified known-empty framework set returned `[]` and removed that question. These separate probes did not alter the main refinement's unknown facts.
+
+| Live mistake | Result |
+| --- | --- |
+| String supplied for enum-set `routing.activities` | `ROUTE.INPUT_INVALID`, exact `/facts/routing.activities`, useful unique-string-set feedback. |
+| Undeclared fact name | `ROUTE.INPUT_INVALID`, non-exact `/facts` pointer without echoing the caller's key. |
+| Old `{type,state,value}` envelope | `INTERFACE.INVALID_ARGUMENTS`, bounded `/facts/*` `oneOf` feedback and `describe_input(route)` hint. |
+
+All three were typed invalid outcomes (`isError: true`), not internal transport errors or successful routes. There were no route construction errors, repair calls or extra discovery calls in the successful connected workflow. The production registry has **no aliases, boolean facts or nullable facts**. Alias-collision, false, null and known-absence semantics therefore use the existing disposable-fixture tests in the supported 608/571-test CI, rather than inventing unsupported production facts or mutating standards. The live explicit-unknown and known-empty checks use actual enum-set definitions. No proposal, publication or production standards edit was made.
 
-## Paired fresh-model comparison (separate from the connected-session gate)
+## Paired fresh-model comparison (separate from the connected-session result)
 
 Two disposable clean checkouts used the same `gpt-6-sol` prompt and authoring/on-demand registration: baseline `ff2e13ed31a42dbe6cc70ee76f7e7a7e58eabe76` (interface 44) and candidate `39d44f007c36684e1ebc546f31a19e20e2e660e5` (interface 45). The model routed a new internal library feature for implementation, verification and documentation, explicitly no application forms, with representation boundaries initially unresolved. It then learned the IPC and generated-contract boundaries, refined against the original snapshot, and followed one returned content page and the full explanation. All four route calls per side succeeded without input repair. The initial candidate explicitly supplied `routing.boundaries: {"state":"unknown"}`; the baseline omitted that still-unknown category. Both preserved other unanswered categories. The candidate's later calls used the returned compact facts directly.