Dashboard › opencode-lore › Distillation
a4a06e4f-092e-4c1e-ba8e-7efbafbbd2b8["lore_tm_v1_olFQq5JD-7kg3tWWqv_R82A5iA1mKc3QnlHsq4lzBo4","lore_tm_v1_dFvF1vvPIBQcEGYIHtUHMUEHDczi9l7cBgY6Xk9_eZ4","lore_tm_v1_Vzr54Ouq4Xi0ElrMHnUiLKigge-R3fomtnNnYfZzpAM","lore_tm_v1_OHKTEKnyR7oFDvKN48z1joSyP9DhJVgVT0BbcJZuvJw","lore_tm_v1_v41rTtPdAduFQ-ndS2qwnVYnBkwVCpy9_VJ1LtY31YA"]
Date: Sep 16, 2026
packages/gateway/test/openai-responses-recall-aware-stream.test.ts:4867-4919 test accepts indexed reasoning-text content lifecycles verifies a reasoning item can progress through response.content_part.added, response.reasoning_text.delta, response.reasoning_text.done, response.content_part.done, and response.output_item.done at content_index: 0 without emitting response.failed.rejects principal reasoning %s introduced only by %s before recall execution at packages/gateway/test/openai-responses-recall-aware-stream.test.ts:4921-4990 covers, in order: 1. summary introduced by output_item.done using summary_text, 2. summary introduced only by terminal output using summary_text, 3. content introduced by output_item.done using reasoning_text, 4. content introduced only by terminal output using reasoning_text. Under validation: "codex", untracked "private terminal reasoning" is suppressed and neither onRecall nor runFollowUp executes; output_item.done mismatches produce PUBLIC_GATEWAY_ERROR, while terminal-only mismatches produce PUBLIC_RECALL_ERROR.rejects terminal reasoning %s before nested recall execution at packages/gateway/test/openai-responses-recall-aware-stream.test.ts:4992-5093 covers, in order: 1. a sparse summary gap containing tracked index 0, undefined index 1, and private index 2; 2. a contradictory terminal summary; 3. a sparse content gap with the same shape; 4. contradictory terminal content. The principal recall and first follow-up each run exactly once ([recallCalls, followUps] = [1, 1]), but the nested recall is blocked; output contains PUBLIC_RECALL_ERROR and omits both "private gap" and "private contradiction".packages/gateway/test/openai-responses-recall-aware-stream.test.ts:5095-5135 test rejects reasoning deltas after text completion sends response.reasoning_summary_text.done with "safe" followed by a late delta "evil" at the same summary_index: 0; the stream must emit response.failed.packages/gateway/test/openai-responses-recall-aware-stream.test.ts:5137 begins test rejects output_item.done changing reasoning text, using reasoning item rs_text_changed and a completed reasoning text value "secret" at content_index: 0.packages/gateway/src/pipeline.ts found assertOutputLifecyclesComplete defined at line 9349 and assertTerminalOutputMatches invoked at lines 10584, 10621, 11192, and 11234.