Upload 5 files

src/api/admin.py

@@ -865,28 +865,39 @@ async def admin_worker_health(request: Request):
     else:
         out["verdict"] = (f"Unexpected analytics status {ana.get('status')}. "
                           f"See probes for details.")
-    return out
+    return out
+
+@router.get("/log-status")
+async def admin_log_status(request: Request):
+    """Outcome of recent analytics writes, without needing container logs.
+
+    'recent_attempts' is newest-first. Read it like this:
+      ok           -> the write reached the Worker (HTTP 2xx)
+      throttled    -> HTTP 403/429: the host is being rate-limited by Cloudflare
+      unreachable  -> no HTTP response; exception type is in 'detail'
+      rejected_401 -> CLOUDFLARE_WORKER_TOKEN does not match ANALYTICS_TOKEN
+    Buffered rows are retried automatically on later queries.
+    """
+    _require_session(request)
+    from src.agentcore.orchestrator import analytics_log_status
+    from src.config import get_settings as _gs
+    st = _gs()
+    data = analytics_log_status()
+    data["worker_url_configured"] = bool(st.cloudflare_worker_url)
+    data["worker_token_configured"] = bool(st.cloudflare_worker_token)
+    if not data["recent_attempts"]:
+        data["hint"] = ("No write attempts recorded since the last restart. "
+                        "Send a query through LibBee, then reload this page.")
+    return data
 
-@router.get("/log-status")
-async def admin_log_status(request: Request):
-    """Outcome of recent analytics writes, without needing container logs.
+@router.get("/canary-status")
+async def admin_canary_status(request: Request):
+    """Last 14 daily canary runs (the Worker's scheduled health probe).
 
-    'recent_attempts' is newest-first. Read it like this:
-      ok           -> the write reached the Worker (HTTP 2xx)
-      throttled    -> HTTP 403/429: the host is being rate-limited by Cloudflare
-      unreachable  -> no HTTP response; exception type is in 'detail'
-      rejected_401 -> CLOUDFLARE_WORKER_TOKEN does not match ANALYTICS_TOKEN
-    Buffered rows are retried automatically on later queries.
+    Each run is one real question through the full pipeline, chosen from a
+    fixed 14-question pool rotating by day-of-year. Canary traffic is excluded
+    from analytics, so these runs never appear in the D1 usage data.
     """
     _require_session(request)
-    from src.agentcore.orchestrator import analytics_log_status
-    from src.config import get_settings as _gs
-    st = _gs()
-    data = analytics_log_status()
-    data["worker_url_configured"] = bool(st.cloudflare_worker_url)
-    data["worker_token_configured"] = bool(st.cloudflare_worker_token)
-    if not data["recent_attempts"]:
-        data["hint"] = ("No write attempts recorded since the last restart. "
-                        "Send a query through LibBee, then reload this page.")
-    return data
+    return await _worker_get("/canary/status")
 

src/api/agent.py

@@ -41,10 +41,18 @@ async def agent_query(request: Request, req: AgentRequest):
     history = req.history
     model = req.model
     client_state = req.client_state
-    _safe_metrics_increment("agent_requests")
+    # Canary requests (X-LibBee-Canary header, sent by the Worker's daily
+    # health probe) run the full pipeline but are EXCLUDED from analytics:
+    # synthetic traffic must never contaminate the D1 usage data that the
+    # research evaluation reports. Canary results live in the Worker's own
+    # canary_runs table instead.
+    is_canary = request.headers.get("x-libbee-canary") == "1"
+    if not is_canary:
+        _safe_metrics_increment("agent_requests")
     response = await _agent_query_inner(question, history, model, client_state, start_time)
     response.answer = _sanitize_llm_response(response.answer)
-    _log_agent_query(question, response, start_time)
+    if not is_canary:
+        _log_agent_query(question, response, start_time)
     _rcfg = _get_runtime_config()
     _ann = _rcfg.get("announcement", "").strip()
     if _ann and response.intent not in ("social_greeting", "social"):