Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
docs: clarify shadow failure capture
Bruce-Yiif8b3d8071f9c
diff --git a/docs/staged-adoption.md b/docs/staged-adoption.mdindex 860d27c0..9f3bed9a 100644--- a/docs/staged-adoption.md+++ b/docs/staged-adoption.md@@ -38,14 +38,19 @@ class ShadowLog:router = Router(hooks=[ShadowLog()])def handle(request):- laya_result = router.predict(request.state, request.questions)+ try:+ laya_result = router.predict(request.state, request.questions)+ except Exception as exc:+ record_laya_failure(request, exc)+ return run_incumbent_action(request)# The shadow result is recorded by the hook; do not execute it here.return run_incumbent_action(request)```-Keep sensitive fields redacted according to the application's policy. If a Laya call fails before-or during inference, record the application-visible failure and continue or fall back according to-the incumbent policy; a shadow logger must not turn logging into a new user-facing action.+Keep sensitive fields redacted according to the application's policy. Catch and log exceptions+around `router.predict(...)` at the application boundary. The prediction hook covers the prediction+lifecycle, but failures before that lifecycle require application-level capture; do not assume+`on_predict_end` saw them. A shadow logger must not turn logging into a new user-facing action.See [Prediction hooks](hooks/index.md), the [hook lifecycle](hooks/lifecycle.md), and[Tracing](hooks/tracing.md) for the event order and `run_id` correlation.