Distillation

Date: Sep 16, 2026

• 🟡 (16:29) Further excerpt from /home/byk/Code/opencode-lore-responses-accumulation-review-fix/packages/gateway/test/openai-responses-recall-aware-stream.test.ts covered lines 4734–5485, including content-part ordering, malformed stream rejection, terminal reasoning behavior, reference reuse, and recall-count checks.
• 🟡 (16:29) Content-part ordering coverage uses item_id: "msg_part_order", output_index: 0, and a "fix" delta; the resulting drained stream is asserted not to contain "response.failed".
• 🟡 (16:29) A parameterized test.each block beginning near line 4992 includes the case "content gap" and tracks recallCalls while processing a follow-up stream.
• 🟡 (16:29) Terminal reasoning follow-up coverage uses ID rs_terminal_reasoning_followup, tracks both recall and follow-up invocation counts, and asserts output does not contain "private contradiction".
• 🟡 (16:29) Related reasoning-stream coverage references summary_index: 0 and ID rs_text_changed.
• 🟡 (16:29) Reference-reuse coverage creates principal response resp_reference_reuse_principal with model gpt-5.6-terra.
• 🟡 (16:29) Test at line 5207 is titled "rejects content-part declarations after value deltas" and exercises late content-part declarations for item_id: "msg_late_part" before a response.output_item.done event.
• 🟡 (16:29) Later tests in the excerpt track a recalled counter, drain the client stream, and inspect parsed output events.