feat(ask): per-post Ask history stack with gap-baseline refresh

AGENTS.md

@@ -324,11 +324,14 @@ A run-bearing analysis-run registry empties only after an unrevoked
 (ADR 0020 / v0.87.0). The documented phrase is not a secret. Do not
 expose purge on a public HTTP route.
 
-`POST /api/analysis-runs` records Pending lineage only (ADR 0017 /
-v2.7.1). TEPP and period-report kinds 422 before any snapshot write.
-`POST /api/analysis-runs/{id}/start` reconstructs a Pending lineage
-cutoff bag through `reconstruct()` / `lineage_edge_specs` (ADR 0021 /
-v0.88.0). Do not invent a theta.
+`POST /api/analysis-runs` records Pending lineage, TEPP measurement, and
+topic-lineage requests (ADR 0017 / ADR 0022); period-report and unknown kinds
+422 before any snapshot write. A Failed measurement remains terminal, while
+Retry creates and starts a new current-snapshot request through the outbox and
+`tepp_client`. `POST /api/analysis-runs/{id}/start` reconstructs a Pending
+lineage cutoff bag through `reconstruct()` / `lineage_edge_specs` (ADR 0021 /
+v0.88.0) or submits supported measurement work through `tepp_client`. Do not
+invent a theta.
 Opening a cutoff-rewritten title shows **Body this run knew** from
 `source_post_revision` beside the live rewrite (ADR 0025 / v2.1.0).
 Do not invent the earlier sentence when no revision covers the cutoff.
@@ -375,8 +378,9 @@ A manual dispatch from another ref is not a publication path. The
 repository-case public namespace
 `https://contextualwisdomlab.github.io/LineageWeave/ontology#` is
 canonical ([ADR 0207](docs/adr/0207-repository-case-ontology-namespace-canonical.md),
-superseding ADR 0157, resolving issue #372); the lowercase form is a
-deprecated compatibility vocabulary with validated term-kind mappings.
+superseding ADR 0157 and amended by ADR 0236); the lowercase form is a
+deprecated compatibility identifier mapped by a published compatibility
+vocabulary, not a separately served namespace.
 New runtime values, exports, fixtures, and database rows mint only
 repository-case IRIs; `scripts/migrate_legacy_namespace.py` rewrites
 stored lowercase IRIs (dry-run by default, never touching provenance

ARCHITECTURE.md

@@ -20,11 +20,21 @@ authorization, provenance persistence, and UI projection only. The current
 Python IRT/report, residual-map, similarity, graph-ranking, and fusion paths
 are explicitly inventoried migration debt rather than evidence that this
 repository owns their mathematics. They move by construct to TEPP,
-fast-mlsirm, or RankWeave after versioned Rust CPU/GPU owner contracts pass
-recovery/equivalence checks; affected product paths fail closed during each
-cutover rather than substituting a local estimate. See
+fast-mlsirm, or RankWeave after each owner's versioned contract passes
+recovery/equivalence checks; TEPP and fast-mlsirm own Rust scientific compute,
+while RankWeave owns its existing retrieval and comparison implementation.
+Affected product paths fail closed during each cutover rather than substituting
+a local estimate. See
 `docs/doctoring/python-mathematical-compute-boundary-audit.md`.
 
+ADR 0237 also keeps accelerator deployment outside this repository. MLX runs
+as a native Apple-silicon inference service behind contextual-orchestrator;
+scientific CPU/CUDA/OpenCL profiles belong to TEPP or fast-mlsirm. RankWeave
+remains the dependency-free Python retrieval-fusion/evaluation owner behind its
+published contract. LineageWeave Compose therefore does not reserve devices or
+mount host drivers; its provider-neutral connectors consume versioned results
+and fail closed when an owning service is unavailable.
+
 ## Data flow
 
 ```mermaid

CHANGELOG.d/2.21.2-cancelled-analysis-next-actions.md

@@ -0,0 +1,5 @@
+# 2.21.2 Name the next action for every cancelled analysis kind
+
+Cancelled lineage, TEPP, topic-lineage, and period-report runs now tell the
+operator the kind-specific valid next action. A terminal cancellation no
+longer leaves the analysis list and detail without guidance.

CHANGELOG.md

@@ -8,6 +8,15 @@ All notable changes to this project are documented here. Format follows
 
 ### Added
 
+- Failed measurement and topic-lineage views can request and start a new
+  immutable current-snapshot run while retaining the Failed history.
+- Analysis-run next actions, empty-snapshot guidance, and selected-post
+  guidance now follow the active English, Korean, Chinese, Japanese, or
+  Vietnamese locale. Customer copy describes the next product action without
+  exposing internal measurement transport boundaries; Storybook includes
+  desktop and phone status matrices.
+- Account-owned per-post Ask conversations can be listed, reopened, and
+  continued with current authorization reapplied to cited evidence (ADR 0235).
 - Persist explicit paragraph, list, table, MathML formula, and caller-parsed
   conversation-turn semantic-unit kinds without inferring absent boundaries.
 - Event Lineage now persists each reconstructed connection's independent
@@ -105,6 +114,10 @@ All notable changes to this project are documented here. Format follows
 
 ### Changed
 
+- Ontology documentation now distinguishes the canonical dereferenceable
+  namespace from the published mapping document for legacy lowercase IRIs;
+  no customer or consumer is directed to a lowercase path that returns 404.
+
 - ADRs 0011 and 0065 now include APA 7th References for the dated W3C
   PROV-O and PROV-DM Recommendations (30 April 2013). Decisions are
   unchanged.

CLAUDE.md

@@ -27,8 +27,9 @@ period-report run on one snapshot. The TEPP path goes through
 Failed (`tepp_not_available` / `tepp_result_not_persisted`). Never
 invent a theta or a local psychometric substitute.
 
-- Failed TEPP is terminal -- open that row and connect a live TEPP
-  transport from it.
+- Failed TEPP is terminal. Retry creates and starts a new current-snapshot
+  request through the existing outbox and `tepp_client`; it never mutates the
+  failed row or invents a local measurement.
 - A failed lineage row retries reconstruction and does not mention TEPP;
   a failed period-report row rebuilds the report.
 - Pending rows claim nothing: pending TEPP is not a calibrated

backend/app/analysis_run_ingestion.py

@@ -6,10 +6,9 @@
 lookup labels come back; source SQL, DSNs, raw records, and provider
 payloads never do.
 
-``create_pending_analysis_run`` (ADR 0017) writes snapshot, counts, frozen
-membership, run, scope, and the first Pending event atomically. It
-records lineage only. It does not reconstruct lineage, accept a TEPP
-kind, or invent a score. ``enqueue_pending_analysis_run`` then
+``create_pending_analysis_run`` (ADR 0017 / ADR 0022) writes snapshot,
+counts, frozen membership, run, scope, and the first Pending event atomically.
+It records a request, never a result or score. ``enqueue_pending_analysis_run`` then
 ``deliver_queued_analysis_run`` later reconstruct lineage (ADR 0021 /
 ADR 0023) or submit TEPP through ``tepp_client`` (ADR 0022). Neither
 path invents a TEPP score.
@@ -333,6 +332,8 @@ async def _serialize_runs(
         }
         if row["scope_entity_name"]:
             item["scope_entity_name"] = row["scope_entity_name"]
+        if row["corporate_entity_id"]:
+            item["scope_corporate_entity_id"] = str(row["corporate_entity_id"])
         if row["scope_key"]:
             item["scope_key"] = row["scope_key"]
         grouping_key = scope_grouping_key(row)
@@ -629,36 +630,25 @@ def __init__(self, status_code: int, detail: str) -> None:
         self.detail = detail
 
 
-def _require_lineage_create_kind(run_kind_code: str) -> None:
-    """Reject TEPP, topic-lineage, and report writes so this path cannot fake those products.
+def _require_requestable_run_kind(run_kind_code: str) -> None:
+    """Allow requestable run kinds without claiming that work has completed.
 
-    TEPP and topic-lineage stay ``tepp_client`` wire paths (ADR 0022 /
-    ADR 0132). Period reports stay on the Reports panel rebuild. A Pending
-    TEPP or topic-lineage row that never called the transport is a
-    fabricated measurement request.
+    TEPP and topic-lineage execution still crosses ``tepp_client`` (ADR 0022 /
+    ADR 0132). A Pending row is an immutable request, not a calibrated result.
+    Period reports stay on the Reports panel rebuild path.
     """
-    if run_kind_code == _TEPP_RUN_KIND:
-        raise AnalysisRunCreateError(
-            422,
-            "Connect a TEPP transport from a Failed TEPP row; this endpoint "
-            "does not invent a measurement.",
-        )
-    if run_kind_code == _TOPIC_LINEAGE_RUN_KIND:
-        raise AnalysisRunCreateError(
-            422,
-            "Connect a TEPP transport from a Failed topic-lineage row; this "
-            "endpoint does not invent a topic model.",
-        )
+    if run_kind_code in {
+        _LINEAGE_RUN_KIND,
+        _TEPP_RUN_KIND,
+        _TOPIC_LINEAGE_RUN_KIND,
+    }:
+        return
     if run_kind_code == _REPORT_RUN_KIND:
         raise AnalysisRunCreateError(
             422,
             "Rebuild the period report from the Reports panel.",
         )
-    if run_kind_code != _LINEAGE_RUN_KIND:
-        raise AnalysisRunCreateError(
-            422,
-            "Only lineage reconstruction can be requested here.",
-        )
+    raise AnalysisRunCreateError(422, "Choose a supported analysis run kind.")
 
 
 @dataclass(frozen=True)
@@ -785,11 +775,12 @@ async def create_pending_analysis_run(
 ) -> dict[str, Any]:
     """Insert snapshot, counts, frozen members, run, scope, and Pending.
 
-    Lineage only. Does not reconstruct, call TEPP, or invent a theta.
+    Does not reconstruct, call TEPP, or invent a theta. Pending measurement
+    and topic-lineage rows become work only through the start/outbox path.
     Kind rejection happens before any snapshot or run insert.
     Idempotent retries compare ``configuration_sha256``.
     """
-    _require_lineage_create_kind(run_kind_code)
+    _require_requestable_run_kind(run_kind_code)
     if scope_kind_code != _CORPORATE_SCOPE:
         raise AnalysisRunCreateError(
             422,

backend/app/analysis_run_start.py

@@ -954,9 +954,9 @@ async def _claim_delivery_plan(
     if weights is None:
         raise AnalysisRunStartError(
             503,
-            "Channel weights are not estimated yet for this run's active "
-            f"channels ({', '.join(sorted(active_channels))}). Run "
-            "scripts/estimate_channel_weights.py, then start this run again.",
+            "Lineage calibration evidence is unavailable for this run. Ask "
+            "your workspace administrator to complete calibration, then start "
+            "this run again.",
         )
     return _DeliveryPlan(
         _LINEAGE_KIND,
@@ -986,9 +986,9 @@ def _execute_delivery_plan(
         except _AdjudicationProviderError as exc:
             raise AnalysisRunStartError(
                 503,
-                "The adjudication provider failed mid-reconstruction; nothing "
-                "was persisted. Check the contextual-orchestrator transport, "
-                "then start this run again.",
+                "Lineage reconstruction is temporarily unavailable. No result "
+                "was saved. Start this run again in a moment; if this continues, "
+                "contact your workspace administrator.",
             ) from exc
         return _DeliveryOutcome(plan.work_kind_code, plan.started_at, tuple(edges))
 

backend/app/global_ask_queue.py

@@ -76,6 +76,10 @@
 # trimmed stream) and are republished by the worker's recovery sweep.
 _REPUBLISH_AFTER_SECONDS = 60
 _RECOVERY_INTERVAL_SECONDS = 30.0
+_ASK_RETRY_MESSAGE = (
+    "Ask Agent is unavailable. Retry in a moment. If this continues, "
+    "contact your workspace administrator."
+)
 # Hard ceiling on one job's answer computation. Without it a hung
 # orchestrator round-trip kept a job `running` indefinitely (observed:
 # 17+ minutes) and, before concurrent processing, stalled every job
@@ -168,11 +172,11 @@ def _verification_next_action(status_code: str) -> str:
 
     return {
         VERIFICATION_SKIPPED: "Enable public verification to check eligible public claims.",
-        VERIFICATION_UNAVAILABLE: "Configure public search and contextual-orchestrator, then retry.",
-        VERIFICATION_NO_PUBLIC_CLAIMS: "Inspect the internal cited posts; no public claim was eligible.",
+        VERIFICATION_UNAVAILABLE: "Ask a workspace administrator to enable public verification, then retry.",
+        VERIFICATION_NO_PUBLIC_CLAIMS: "Ask about a specific claim or narrow the time range, then retry.",
         VERIFICATION_COMPLETED: "Inspect public evidence separately before any governed graph review.",
         CLAIM_NOT_ENOUGH_INFORMATION: "Collect stronger authoritative evidence before accepting the claim.",
-    }.get(status_code, "Inspect the authorized cited posts and their evidence.")
+    }.get(status_code, "Ask about a specific claim or narrow the time range, then retry.")
 
 
 async def _verify_public_claims(
@@ -348,7 +352,7 @@ def can_see(row: asyncpg.Record) -> bool:
         record_server_failure("global_ask", exc, outcome="internal_error")
         raise HTTPException(
             status.HTTP_503_SERVICE_UNAVAILABLE,
-            "Ask Agent is unavailable: authorized evidence could not be assembled",
+            _ASK_RETRY_MESSAGE,
         ) from exc
     cutoff_text = knowledge_cutoff.isoformat() if knowledge_cutoff else None
     grounding_status = ask_grounding_status(sources, cutoff_text)
@@ -404,7 +408,7 @@ def can_see(row: asyncpg.Record) -> bool:
         record_server_failure("global_ask", exc, outcome="provider_unavailable")
         raise HTTPException(
             status.HTTP_503_SERVICE_UNAVAILABLE,
-            "Ask Agent is unavailable: contextual-orchestrator could not complete the answer",
+            _ASK_RETRY_MESSAGE,
         ) from exc
     except (KeyError, ValueError) as exc:
         # Contract/schema fault: the orchestrator responded but its payload
@@ -413,7 +417,7 @@ def can_see(row: asyncpg.Record) -> bool:
         record_server_failure("global_ask", exc, outcome="provider_unavailable")
         raise HTTPException(
             status.HTTP_503_SERVICE_UNAVAILABLE,
-            "Ask Agent is unavailable: contextual-orchestrator could not complete the answer",
+            _ASK_RETRY_MESSAGE,
         ) from exc
     except Exception as exc:
         # Unexpected defect. Keep the customer boundary and emit a full
@@ -422,7 +426,7 @@ def can_see(row: asyncpg.Record) -> bool:
         record_server_failure("global_ask", exc, outcome="internal_error")
         raise HTTPException(
             status.HTTP_503_SERVICE_UNAVAILABLE,
-            "Ask Agent is unavailable: contextual-orchestrator could not complete the answer",
+            _ASK_RETRY_MESSAGE,
         ) from exc
     cited_ids = list(answer.cited_post_ids)
     verification_status, external_claims = await _verify_public_claims(
@@ -552,7 +556,7 @@ async def process_global_ask_job(
         chat_client = chat_factory()
         if not chat_client.available:
             raise _SafeJobError(
-                "Ask Agent is unavailable: set ORCHESTRATOR_BASE_URL / ORCHESTRATOR_API_KEY"
+                _ASK_RETRY_MESSAGE
             )
         payload = await asyncio.wait_for(
             compute_global_ask_answer(
@@ -588,16 +592,13 @@ async def process_global_ask_job(
             # state / missing config) — never a provider-boundary leak.
             detail = str(exc)
         elif isinstance(exc, asyncio.TimeoutError):
-            detail = f"job exceeded the {JOB_DEADLINE_SECONDS}s deadline"
+            detail = _ASK_RETRY_MESSAGE
         else:
             # Provider responses/exceptions can carry credentials, gateway
             # diagnostics, or model output (ADR 0123): never persist the
             # raw exception text as a durable `failure_detail`. The
             # traceback just logged keeps it for operator debugging only.
-            detail = (
-                "Ask Agent is unavailable: contextual-orchestrator returned "
-                "no complete evidence object"
-            )
+            detail = _ASK_RETRY_MESSAGE
         async with pool.acquire() as conn:
             await conn.execute(
                 """