Dashboard › opencode-lore › Distillation
3d762075-67d5-4a0b-8be9-f42eabcfefab["lore_tm_v1_0rDe7Zh-M3fTb-o1IDi37_x6jBm1Z4P-q6vEOxx3wwc","lore_tm_v1_m9957dvupZ4aXSJaBDNEEZL0IObp1Ea_ep_pjKKACi8","lore_tm_v1_7yEYrDEHGpPFCsfTCJDzZfWcZqaiUdd4Z6vmT0L3PG0","lore_tm_v1_mMrzo8sBVAgLaERvvHLjoEaAm7flafi-uadQ-1NJVFU","lore_tm_v1_prVSN9DFA35i8ZR21vjAdCadIRSMf8CNUG0DVKIaNP8","lore_tm_v1_-wvDYPc9E26NxidrDPWV8BjLe7QSWTR13GYSyRMDpQw","lore_tm_v1_sPtq8_QWTbwiP0Cj1ETwbzFeLC7HSyKQ_3rD1Rq4VDA","lore_tm_v1_lmr0yiProgqXPHQIix5HIueTVetM202n8p31hjWAQRk","lore_tm_v1_YgnLZECppGzuQ2hhsQhprYHy5FEo_XUKnzb3GzkaaSY","lore_tm_v1_4T9Ojaco7nbQUQD7upRUv5wqkBkws4sxPaXbLZn464U","lore_tm_v1_nIZshW-oogWbkRtt70gEuxXN9VlJRsFYsaGBCZr4d54","lore_tm_v1_IM32b-h8CSqzfqEW4TDFK50fI2s0-JWKN3CdT_e86Ec","lore_tm_v1_mwJhJqmFceaH9MqtNtCcx6NE9A6NupXptHNuJCIl1O4","lore_tm_v1_H7PShyFCa0WUOnRdItsbiqKUIyZ3uy0xaJE2FLZnXZc","lore_tm_v1_Y12ZfyWrROJTJXAVLhts9neyMaVEPzrrYukoKC-tsv4","lore_tm_v1_kP6z5tr-GLV9ADHMhMbfgPvG7Uh08rOzHYYBJNAiFss","lore_tm_v1_F2ybfJ1Foo1o3YnmHCZN1gQ8jrw38BEV9ZyD9JwaDSk"]
Date: Sep 16, 2026
never executes recall from an incomplete principal is in packages/gateway/test/openai-responses-recall-aware-stream.test.ts:9783.never executes a chained recall from an incomplete continuation is in packages/gateway/test/openai-responses-recall-aware-stream.test.ts:9809.packages/gateway/test/openai-responses-recall-aware-stream.test.ts has failed-terminal assertions expecting { response: { status: "failed", output: [] } } at lines 1131, 1262, 7401, 7593, 7663, and 7890.consumeReferenceEvent() in packages/gateway/src/pipeline.ts:9572-9629 privately tracks item_reference lifecycles by output_index; it requires a nonnegative safe-integer index, a nonempty string id, and no fields other than type and id. It rejects index/identity reuse against references, raw items, synthetic identities, output identities, existing id/call_id values, and referenceIdentities; completion must be a matching response.output_item.done.assertReferenceLifecyclesComplete() in packages/gateway/src/pipeline.ts:9630-9640 rejects any item_reference lacking completion with Responses stream ended before item_reference completion for index ${outputIndex}. stripHiddenReferenceOutput() at lines 9654-9674 removes tracked item_reference entries from terminal response.output.item_reference lifecycle events must never be forwarded. Test never forwards response-side item_reference lifecycle events in packages/gateway/test/openai-responses-recall-aware-stream.test.ts:2497-2539 requires neither streamed output nor the completed GatewayResponse to contain item_reference, while also requiring no response.failed.item_reference ID must never be logged. Test never logs an item_reference id is in packages/gateway/test/openai-responses.test.ts:878; the related warning text is dropping unresolvable Responses API item_reference; gateway is stateless full-history and cannot resolve server-side item references.rejects an item_reference missing output_item.done in packages/gateway/test/openai-responses-recall-aware-stream.test.ts:2541-2561 creates msg_reference_incomplete with only response.output_item.added and requires the client stream to contain response.failed.response.created, response.in_progress, and terminal lifecycle events must NEVER reach the client because the principal already emitted response.created and the gateway rebuilds response.completed; forwarding them would duplicate initialization/terminal events and violate the SSE protocol.recall-only: emits marker, pipes the continuation inline, rebuilds completed in packages/gateway/test/openai-responses-recall-aware-stream.test.ts:6534-6597 requires continuation text Here is the answer from the continuation. and buildAnchor("architecture"), suppresses Searching and raw recall function_call output, emits exactly one response.created, exactly one response.completed, zero response.in_progress, and requires all emitted sequence_number values to equal their zero-based positions.sequenceChunk() in packages/gateway/src/pipeline.ts:9708-9734 parses formatted SSE frames and injects monotonically increasing sequence_number values using nextSequenceNumber++; non-event, data-less, or unparsable frames are retained unchanged.applyResponsesEvent() in packages/gateway/src/stream/openai-responses.ts:582-812 accumulates response.created, response.in_progress, output items, output-text/refusal/function-argument deltas and done events, and terminal variants response.failed, response.done, response.incomplete, and response.completed. It maps failed/cancelled terminals to response.failed, incomplete terminals to response.incomplete, and otherwise to response.completed; other events such as response.content_part.* and response.reasoning_summary_* are ignored for accumulation.SUPPORTED_RESPONSES_OUTPUT_ITEM_TYPES in packages/gateway/src/stream/openai-responses.ts:95-127 contains, in order: 1. message, 2. function_call, 3. function_call_output, 4. reasoning, 5. item_reference, 6. web_search_call, 7. file_search_call, 8. computer_call, 9. computer_call_output, 10. computer_tool_call, 11. computer_tool_call_output, 12. code_interpreter_call, 13. image_generation_call, 14. local_shell_call, 15. local_shell_call_output, 16. shell_call, 17. shell_call_output, 18. mcp_call, 19. mcp_list_tools, 20. mcp_approval_request, 21. mcp_approval_response, 22. custom_tool_call, 23. custom_tool_call_output, 24. apply_patch_call, 25. apply_patch_call_output, 26. program, 27. program_output, 28. tool_search_call, 29. tool_search_output, 30. additional_tools, 31. compaction.packages/gateway/src/pipeline.ts:12460-12464 rejects item_reference with upstream Responses request did not complete, because no streamed item lifecycle exists to resolve it and accepting it would silently erase provider output during normalization.packages/gateway/src/pipeline.ts:12473-12495 allows incomplete reasons only max_output_tokens and content_filter; any other reason, or a malformed non-string reason, throws upstream Responses request did not complete.