Upload 2 files

_dataset_schema.py

@@ -168,6 +168,18 @@ _QUICK_OPTS: list[dict[str, Any]] = [
     {"slug": "helpful",     "title": "Helpful",     "value": +1, "sentiment": "positive"},
 ]
 
+#: Set of slug values associated with quick (👍/👎) feedback options.
+#: Disjoint from all panel slugs — used for deterministic ratingMode detection
+#: when ``ratingMode`` is not explicitly provided in the payload (old records).
+_QUICK_SLUGS: frozenset[str] = frozenset(e["slug"] for e in _QUICK_OPTS)
+
+#: Set of sentiment strings used as quick feedback mode indicators.
+#: Old records written before the slug fix may carry "positive"/"negative" here.
+_QUICK_SENTIMENTS: frozenset[str] = frozenset(e["sentiment"] for e in _QUICK_OPTS)
+
+#: All identifiers that unambiguously indicate quick (👍/👎) rating mode.
+_QUICK_IDENTIFIERS: frozenset[str] = _QUICK_SLUGS | _QUICK_SENTIMENTS
+
 # Derived lookup tables.
 _SLUG_TO_TITLE: dict[str, str] = {
     **{e["slug"]: e["title"] for e in _PANEL_SCALE},
@@ -309,7 +321,13 @@ def normalize_rating(
         elif label_str in _KNOWN_TITLES:
             detected_mode = "quick"
         elif label_str and _SLUG_RE.match(label_str):
-            detected_mode = "panel"
+            # Slug-based mode detection: quick slugs ("helpful", "not_helpful")
+            # and panel slugs ("mostly_positive", "excellent", …) are disjoint
+            # sets — membership check is sufficient and deterministic.
+            # This handles contribution records where _feedbackStore.ratingMode
+            # is forwarded in ratingMode (new JS) but also back-compats old
+            # records that only carried ratingLabel (slug or Title Case).
+            detected_mode = "quick" if label_str in _QUICK_IDENTIFIERS else "panel"
         elif rating_value in (-1, 1) and not label_str:
             detected_mode = "quick"
         else:
@@ -546,8 +564,8 @@ def normalize_contribution_record(
     rating_fields = normalize_rating(
         rec.get("ratingValue"),
         rec.get("ratingLabel"),
-        rating_mode=rec.get("ratingMode"),     # new JS field (None if old)
-        rating_title=rec.get("ratingTitle"),   # new JS field (None if old)
+        rating_mode=rec.get("ratingMode"),     # from _feedbackStore.ratingMode (new JS)
+        rating_title=rec.get("ratingTitle"),   # from _feedbackStore.ratingTitle (new JS)
         feedback_id=None,                      # no per-event session ID for contributions
     )
 

app.py

@@ -1309,7 +1309,7 @@ async def contribute(request: Request) -> JSONResponse:  # noqa: PLR0912
                     path_or_fileobj=rows_jsonl.encode(),
                 )
             ],
-            commit_message=f"Add {len(records)} feedback record(s)",
+            commit_message=f"Add {len(records)} contribution record(s)",
         )
     except Exception as exc:
         logger.error(json.dumps({"event": "contribute.hf_fail", "error": str(exc)}))
@@ -1829,18 +1829,13 @@ async def feedback(request: Request) -> JSONResponse:
             # The resulting record is structurally identical to a contribution row
             # so both sources concatenate directly into one pandas DataFrame.
             # See _dataset_schema.py for the full canonical column list.
-            record = json.dumps(
-                normalize_feedback_record(
-                    payload,
-                    server_ts_ms=int(_time.time() * 1000),
-                ),
-                ensure_ascii=False,
+            _rec_dict: dict = normalize_feedback_record(
+                payload,
+                server_ts_ms=int(_time.time() * 1000),
             )
-            # Derive dedup_key from the normalised record (avoids duplicating logic).
-            import json as _json  # noqa: PLC0415 — already imported at module scope
-            _rec_dict = _json.loads(record)
-            conversation_id = _rec_dict.get("conversationId") or ""
-            answer_index    = _rec_dict.get("answerIndex", "")
+            record: str = json.dumps(_rec_dict, ensure_ascii=False)
+            conversation_id: str = str(_rec_dict.get("conversationId") or "")
+            answer_index: Any = _rec_dict.get("answerIndex", "")
             filename = f"feedback/{int(_time.time() * 1000)}.jsonl"
             api = HfApi(token=HF_DATASET_TOKEN)
             # Fix 3 (feedback): same event-loop fix as /v1/contribute — offload