A reworded GPT-6 tool refusal is warned once, and a full effort memory forgets the oldest

Shipped

28 செப்டம்பர், 2026 அன்று 11:47 PM UTC

Author

Kamo

Commit

77f677e

Review follow-ups for the reasoning_effort none fix: - A 400/422 that names reasoning_effort and tools but is not the refusal ToolEffortFallback reads logs one WARN (model, param, code, the first 200 characters of the message, never the key), so a vendor rewording shows. - A full memory forgets the oldest learned model, not every one. - A test pins TOOL_CALLING routing to the adapter's tool round trip, which member chat's tool turns rely on.