feat: add evidence-honest Global Ask knowledge cutoff (v2.23.0)

AGENTS.md

@@ -178,6 +178,11 @@ invent a cited post.
 A linked Event Lineage node opened from that focused popup keeps the
 originating flags (ADR 0097). Do not invent a cited post.
 
+Ask Agent accepts an optional knowledge cutoff (ADR 0101). A dated
+question uses retained revisions and never substitutes a live body. A
+live query is never labeled as-of. Do not invent a cutoff body or a
+TEPP theta.
+
 
 ## Tests
 

ARCHITECTURE.md

@@ -286,7 +286,9 @@ Calendar commitments use the same Event Lineage focus path (ADR 0094),
 Customer master related posts use the same Event Lineage focus path
 (ADR 0095). Ask Agent cited posts use the same Event Lineage focus path
 (ADR 0096). A linked Event Lineage node opened from a focused popup keeps
-those flags (ADR 0097), and the full detail popup includes Korean
+those flags (ADR 0097). Ask Agent accepts an optional knowledge cutoff
+and uses retained `source_post_revision` bodies for that clock (ADR 0101);
+a live query is never labeled as-of. The full detail popup includes Korean
 summary/key-events/R&R, VOC evidence excerpts, an Event Lineage panel
 (direct vs. indirect links; a link opens that post), the Keyman
 affiliate tree (resolved ancestors plus unresolved org roots), Keyman +

CHANGELOG.d/2.23.0-global-ask-knowledge-cutoff.md

@@ -0,0 +1,6 @@
+# 2.23.0 Global Ask knowledge cutoff
+
+Ask Agent can name a dated question. Retrieval keeps only source posts
+that existed by that clock and matches the covering
+`source_post_revision` text. A missing historical body is an explicit
+limitation. Live queries stay live-only.

CHANGELOG.md

@@ -4,6 +4,16 @@ All notable changes to this project are documented here. Format follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/); versioning follows
 [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.23.0] - 2026-08-20
+
+### Added
+
+- Ask Agent now accepts an optional knowledge cutoff. A dated question
+  matches retained source-post revisions from that clock and never
+  substitutes a live body or live rewrite text. Fully, partly, and
+  live-only answers are named separately. No TEPP theta is invented. No
+  as-of label is applied to a live query (ADR 0101 / ADR 0016 / ADR 0025).
+
 ## [2.17.0] - 2026-08-19
 
 ### Added

CLAUDE.md

@@ -105,3 +105,11 @@ cited post.
 From a GNB-focused popup, open a linked Event Lineage node: Event Lineage
 stays focused and names the new post as current (ADR 0097). A home-list
 DAG walk does not. Do not invent a theta.
+
+## Ask Agent knowledge cutoff (v2.23.0)
+
+Open Ask Agent. Optionally set a knowledge cutoff. A dated question uses
+retained source-post revisions from that clock. A live query stays
+live-only and is never labeled as-of. A missing historical body is named
+and the live rewrite is not used (ADR 0101). Do not invent a theta or a
+cutoff body.

backend/app/main.py

@@ -73,8 +73,12 @@
     ChatSourceDocument,
     ContextualOrchestratorPostChatClient,
     NullPostChatClient,
+    ask_grounding_status,
+    ask_next_action,
+    cited_post_citations,
     cited_post_evidence,
     cited_post_summaries,
+    historical_body_limitations,
     render_global_ask_context,
 )
 from lineageweave.post_content_normalization import normalize_post_body
@@ -2533,6 +2537,7 @@ class GlobalAskRequest(BaseModel):
 
     question: str
     session_id: str | None = None
+    knowledge_cutoff: str | None = None
 
 
 def global_ask_timeline(sources: list[ChatSourceDocument]) -> list[dict[str, str | None]]:
@@ -2663,6 +2668,15 @@ async def ask_agent(
             UUID(request.session_id)
         except ValueError:
             raise HTTPException(status.HTTP_404_NOT_FOUND, "Global Ask session not found") from None
+    knowledge_cutoff = None
+    if request.knowledge_cutoff is not None and request.knowledge_cutoff.strip():
+        try:
+            knowledge_cutoff = parse_as_of_clock(request.knowledge_cutoff)
+        except ValueError as exc:
+            raise HTTPException(
+                status.HTTP_422_UNPROCESSABLE_ENTITY,
+                "knowledge_cutoff must be an ISO-8601 timestamp",
+            ) from exc
     client = _post_chat_client()
     if not client.available:
         raise HTTPException(
@@ -2681,6 +2695,7 @@ async def ask_agent(
             lambda row: _can_see_post(account, row),
             account.corporate_entity_ids,
             question=question,
+            knowledge_cutoff=knowledge_cutoff,
         )
     if conversation.compress_turns:
         compressor = getattr(client, "compress_context", None)
@@ -2712,7 +2727,11 @@ async def ask_agent(
         conversation.summary,
         conversation.recent_turns,
     )
-    if not sources:
+    grounding_status = ask_grounding_status(sources, knowledge_cutoff)
+    limitations = historical_body_limitations(sources)
+    cutoff_text = knowledge_cutoff.isoformat() if knowledge_cutoff is not None else None
+    llm_sources = [source for source in sources if not source.historical_body_unavailable]
+    if not llm_sources:
         async with pool.acquire() as conn:
             await persist_global_ask_turn(conn, conversation.session_id, question, "", ())
         await publish_operation_event(
@@ -2725,17 +2744,20 @@ async def ask_agent(
             "session_id": conversation.session_id,
             "answer_text": "",
             "cited_post_ids": [],
-            "cited_posts": [],
-            "source_post_ids": [],
+            "cited_posts": cited_post_citations(sources, [source.post_id for source in sources]),
+            "source_post_ids": [source.post_id for source in sources],
             "cited_post_evidence": [],
-            "timeline": [],
-            "next_action": "No authorized source posts are available for this question.",
+            "timeline": global_ask_timeline(sources),
+            "knowledge_cutoff": cutoff_text,
+            "grounding_status": grounding_status,
+            "limitations": limitations,
+            "next_action": ask_next_action(grounding_status, has_sources=bool(sources)),
         }
     try:
         answer = await asyncio.to_thread(
             client.answer,
             question,
-            sources,
+            llm_sources,
             conversation_context=conversation_context,
         )
     except (HttpClientError, KeyError, OSError, ValueError) as exc:
@@ -2762,10 +2784,14 @@ async def ask_agent(
         "session_id": conversation.session_id,
         "answer_text": answer.answer_text,
         "cited_post_ids": cited_ids,
-        "cited_posts": cited_post_summaries(sources, cited_ids),
-        "cited_post_evidence": cited_post_evidence(sources, cited_ids),
-        "source_post_ids": [source.post_id for source in sources],
-        "timeline": global_ask_timeline(sources),
+        "cited_posts": cited_post_citations(llm_sources, cited_ids),
+        "cited_post_evidence": cited_post_evidence(llm_sources, cited_ids),
+        "source_post_ids": [source.post_id for source in llm_sources],
+        "timeline": global_ask_timeline(llm_sources),
+        "knowledge_cutoff": cutoff_text,
+        "grounding_status": grounding_status,
+        "limitations": limitations,
+        "next_action": ask_next_action(grounding_status, has_sources=True),
     }