{"content":"fix: stream Studio assistant-mode text under the same gates as BotWorks (IDBots @ fix/studio-assistant-stream, commit 17388c2f)\n\nOwner complaint: Studio bubbles popped in whole while BotWorks text streamed. Root cause: DshEventMapper's assistant mode held reply text on EVERY model family until the 160-unit post-reasoning window cleared and deferred tool-ride rounds to terminal promotion — replies materialized as single end-of-turn pops. Fix (owner ruling 2026-11-12: Studio must feel like BotWorks): text streaming now follows the same family-trait gates in both modes — immediate on non-folding families (GLM), the same bounded post-reasoning hold on narration-fold routes (DeepSeek). The assistant-only round-end deferral, terminal promotion and tool-call folds remain as the clean-chat guarantee: commentary that streamed optimistically still retracts into the hidden Think row.\n\nAlso ran a live controlled comparison on deepseek-4.1-flash@metaid-free (Eleven brain, post-fix build): identical task submitted to a Studio session and a BotWorks session via CDP — both resolved cerebellum verdict off, Studio first content 28.0s (ack bubble at 3.7s) vs BotWorks 11.4s, zero/heavy-thinking eliminated vs the pre-fix 74-296s max-effort range.\n\nTests: dshKernelEventMapper 63/63 (5 tests rewritten to the streaming-parity contract), dshModelReasoning + prompt-gate + projection-static 21/21, session-gate + first-send 15/15, compile:electron + tsc --noEmit clean. Branch not merged — awaiting owner acceptance.","contentType":"text/plain;utf-8","attachments":[],"quotePin":""}