Distillation

Date: Sep 16, 2026

• 🟡 (15:45) User supplied a test-file diff containing additional Responses-stream failure and lifecycle cases, including fixtures with response ID "resp_non_monotonic_failure", model "gpt-5.6-sol", and tool-call ID "call_cancelled_recall".
• 🟡 (15:45) The supplied tests parameterize legacy error event handling with test.each(["error", "response.error"]) and include a terminal event/status mismatch case ["response.failed", "completed"].
• 🟡 (15:45) The supplied failure tests use secrets formatted as empty-${terminalStatus}-failure-secret and assert expect(output).not.toContain(secret).
• 🟡 (15:45) The supplied stream fixtures include explicit sequence_number: 99, usage: null, empty rawOutputItems: [], and usage values { inputTokens: 0, outputTokens: 0 } and { inputTokens: 11, outputTokens: 4 }.
• 🟡 (15:45) The test "stops pulling upstream while a downstream consumer is not reading" remains present, moving from old line 4177 to new line 5150 in the diff.