diff --git a/kalibr/__init__.py b/kalibr/__init__.py
index 587a86a..e2c5ae7 100644
--- a/kalibr/__init__.py
+++ b/kalibr/__init__.py
@@ -103,7 +103,7 @@ def call_openai(prompt):
 from .feedback import (
     KalibrFeedback, track_run, user_rejected, user_accepted, get_feedback,
     classify_satisfaction, classify_satisfaction_async, emit_signal,
-    report_pipeline, report_user_turn, report_action, request_feedback, submit_feedback,
+    report_pipeline, report_user_turn, report_action,
 )
 
 if os.getenv("KALIBR_AUTO_INSTRUMENT", "true").lower() == "true":
@@ -204,6 +204,4 @@ def call_openai(prompt):
     "report_pipeline",
     "report_user_turn",
     "report_action",
-    "request_feedback",
-    "submit_feedback",
 ]
diff --git a/kalibr/feedback.py b/kalibr/feedback.py
index 4c7a34e..9c9ee47 100644
--- a/kalibr/feedback.py
+++ b/kalibr/feedback.py
@@ -658,78 +658,6 @@ def _send():
         logger.warning("report_action failed: %s", e)
 
 
-_DIMENSION_DEFS = {
-    "length": {"label": "Length", "options": [-1, 0, 1], "labels": ["Too long", "Good", "Too short"]},
-    "tone": {"label": "Tone", "options": [-1, 0, 1], "labels": ["Too casual", "Good", "Too formal"]},
-    "content": {"label": "Content", "options": [-1, 0, 1], "labels": ["Off-topic", "Good", "Missing info"]},
-    "factuality": {"label": "Factuality", "options": [-1, 0, 1], "labels": ["Inaccurate", "Good", "Unsure"]},
-    "format": {"label": "Format", "options": [-1, 0, 1], "labels": ["Poor format", "Good", "Needs restructuring"]},
-}
-
-
-def request_feedback(
-    dimensions: Optional[list] = None,
-) -> dict:
-    """
-    SDK-only. Returns a structured feedback form dict the developer renders
-    to their users. No network call.
-    """
-    if dimensions is None:
-        dimensions = ["length", "tone", "content", "factuality", "format"]
-    return {
-        "question": "How was this response?",
-        "dimensions": {
-            d: _DIMENSION_DEFS.get(d, {"label": d.title(), "options": [-1, 0, 1], "labels": ["Bad", "Good", "Needs work"]})
-            for d in dimensions
-        },
-    }
-
-
-def submit_feedback(session_id: str, ratings: dict) -> None:
-    """
-    Submit explicit user feedback. Writes ONE signal row PER dimension
-    with a non-zero rating. Fire-and-forget.
-
-    ratings: {"length": -1, "tone": 0, "content": 1, ...}
-    Values: -1 (negative), 0 (neutral — skipped), 1 (positive)
-    """
-    try:
-        session = _read_session(session_id)
-        if session is None:
-            return
-
-        api_key, tenant_id, base_url = _fb_config()
-        if not api_key or not tenant_id:
-            return
-
-        trace_id = session.get("trace_id", "")
-        goal_val = session.get("goal", "")
-
-        def _send_all():
-            try:
-                for dimension, rating in ratings.items():
-                    if rating == 0:
-                        continue
-                    strength = 0.0 if rating == -1 else 1.0
-                    payload = {
-                        "trace_id": trace_id,
-                        "signal_type": "explicit_feedback",
-                        "signal_source": "explicit_feedback",
-                        "strength": strength,
-                        "confidence": 1.0,
-                        "dimension": dimension,
-                        "goal": goal_val,
-                        "session_id": session_id,
-                    }
-                    _emit_signal_http(base_url, api_key, tenant_id, payload)
-            except Exception:
-                pass
-
-        threading.Thread(target=_send_all, daemon=True).start()
-    except Exception as e:
-        logger.warning("submit_feedback failed: %s", e)
-
-
 def emit_signal(
     signal_type: str,
     strength: float = 0.5,
diff --git a/pyproject.toml b/pyproject.toml
index 3e65916..ce36f42 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "kalibr"
-version = "1.12.0"
+version = "1.12.1"
 description = "Outcome-aware LLM routing for production AI agents. Routes between models, tools, and parameters based on real success signals using Thompson Sampling. Automatic fallback, cost optimization, and continuous learning — no redeploy required."
 authors = [{name = "Kalibr Team", email = "support@kalibr.systems"}]
 readme = "README.md"