A re-ask is a plain question
When a tool needs one more detail from the caller, such as which size or the caller's number, the receptionist asks for it as an ordinary turn. It no longer speaks the line reserved for a failed request.
Quiet while a tool works
While the receptionist looks something up, it stays silent instead of saying a filler such as hold on a sec. Those fillers had taught it to read its own reasoning aloud on later turns.
Call page shows what a tool said
On the call page, each tool call that did not simply succeed shows what the receptionist was told, such as the re-ask or the spoken error, under the tool name.
Ratings no longer use a separate AI grading
The pass or fail verdict that a separate AI grading pass produced is no longer requested. A call rates as an error when a tool failed or the call ended on a failure; otherwise it stays neutral rather than judged.
Behind the scenes. Each call now keeps its turn-by-turn record with timing and transfers, so a rating can later be computed from facts we hold.