[NOTICKET] fix(retrieval): no page citation for TXT documents

- New helper `citable_page(data)` in retrieval/base.py: the chunk's page to show, or None when there is no page or the file type has no real pages
- TXT ingest stores every chunk as page_label 1 (ingest_v2.chunks_from_text), so without this the chip and the answer would always say "p. 1"
- Used for the traceability chip label, the chunk labels sent to the answer model, and the planner's document context; PDF and DOCX unchanged
- Raw `page_label` stays in the traceability payload; file type falls back to the filename extension
- Local tests 12/12 for this change, ruff clean; the 6 failing unit tests fail the same way before this change (stale retriever fake, one planner validator test)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

src/agents/chat_handler.py

@@ -37,7 +37,7 @@ from typing import TYPE_CHECKING, Any
 from langchain_core.messages import BaseMessage
 
 from src.middlewares.logging import get_logger
-from src.retrieval.base import RetrievalResult
+from src.retrieval.base import RetrievalResult, citable_page
 from src.traceability import TraceabilityScratchpad, TraceabilityToolInvoker
 
 from ..charts import chart_envelopes
@@ -1080,8 +1080,9 @@ def _render_document_context(results: list[RetrievalResult]) -> str:
     for r in results:
         data = r.metadata.get("data", {}) if isinstance(r.metadata, dict) else {}
         where = data.get("filename") or "document"
-        if data.get("page_label") is not None:
-            where += f", p. {data['page_label']}"
+        page = citable_page(data)  # None for TXT (every chunk is "page 1")
+        if page is not None:
+            where += f", p. {page}"
         text = r.content.strip()
         if len(text) > _DOC_CONTEXT_MAX_CHARS:
             text = text[:_DOC_CONTEXT_MAX_CHARS].rstrip() + " …"
@@ -1110,12 +1111,12 @@ def _normalize_chunks(raw: Any) -> list[DocumentChunk]:
                 DocumentChunk(
                     content=str(item.get("content", "")),
                     filename=item.get("filename"),
-                    page_label=item.get("page_label"),
+                    page_label=citable_page(item),
                 )
             )
         elif isinstance(item, RetrievalResult):
             data = item.metadata.get("data", {})
-            page = data.get("page_label")
+            page = citable_page(data)  # None for TXT (every chunk is "page 1")
             chunks.append(DocumentChunk(
                 content=item.content,
                 filename=data.get("filename"),

src/retrieval/base.py

@@ -13,6 +13,26 @@ class RetrievalResult:
     source_type: str  # "document" | "database"
 
 
+# File types whose `page_label` is not a real page: TXT ingest stores every chunk as
+# page 1 (`document/ingest_v2.chunks_from_text`), so citing it would always say "p. 1".
+_NO_PAGE_TYPES = frozenset({"txt"})
+
+
+def citable_page(data: dict[str, Any]) -> Any:
+    """The chunk's page to show the user (chip label, answer citation), or None.
+
+    `data` is a chunk's `metadata["data"]`. None when there is no page, or when the
+    file type has no real pages (TXT).
+    """
+    file_type = str(data.get("file_type") or "").lower()
+    if not file_type:
+        name = str(data.get("filename") or "").lower()
+        file_type = name.rsplit(".", 1)[-1] if "." in name else ""
+    if file_type in _NO_PAGE_TYPES:
+        return None
+    return data.get("page_label")
+
+
 class BaseRetriever(ABC):
     @abstractmethod
     async def retrieve(

src/traceability/scratchpad.py

@@ -17,6 +17,7 @@ from typing import Any
 from pydantic import BaseModel
 
 from src.middlewares.logging import get_logger
+from src.retrieval.base import citable_page
 
 from .schemas import (
     KnowledgeUsed,
@@ -356,7 +357,7 @@ class TraceabilityScratchpad:
                 continue
             self._doc_seen.add(key)
             filename = data.get("filename", "Unknown")
-            page = data.get("page_label")
+            page = citable_page(data)  # None for TXT (every chunk is "page 1")
             source: dict[str, Any] = {
                 "type": "document",
                 # Display label. The FE citation chip reads `name` (then title/source/
@@ -365,7 +366,7 @@ class TraceabilityScratchpad:
                 "name": f"{filename}, p. {page}" if page is not None else filename,
                 "document_id": data.get("document_id"),
                 "filename": filename,
-                "page_label": page,
+                "page_label": data.get("page_label"),
                 "query": _truncate(query),
             }
             snippet = data.get("snippet") or data.get("content") or data.get("text")