{"version":"2.1.225","anchor":"classifier-refusals-are-tagged-separately-in-blocking-teleme","canonical_anchor":"classifier-refusals-are-tagged-separately-in-blocking-teleme","heading":"Classifier refusals are tagged separately in blocking telemetry","tier":"internal","area":"Telemetry","scope":null,"heads_up":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.225\/e\/classifier-refusals-are-tagged-separately-in-blocking-teleme","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.225","markdown":"### Classifier refusals are tagged separately in blocking telemetry\n\nSafety-classifier refusals are now logged distinctly from ordinary parse failures.\n\n**What**\n\nThe two-stage prompt-injection and safeguard classifier now records when the model stopped with a refusal, which previously looked identical to an ordinary parse failure.\n\n**Details**\n\n- The parse-failure event now carries `stopReason` plus `failureKind: \"policy_refusal\"`.\n- The block result itself carries `refusedBySafeguard`.\n\n**Evidence**\n\n`refusedBySafeguard`\n\n- Area: Telemetry\n- Tier: Under the hood\n- Useful: 1\/5\n- Signal: 0\/5"}