The states of a dictation Diagram

The main path

Idle to Pasted: hotkey, first audio, stop, transcript, cleaned text
whisper-server and a waiting refine start in Starting, so the later states are short
With cleanup off or in Raw mode, Transcribing goes straight to the paste (refine still applies the dictionary)

Stopping early

Esc or a second press while the mic opens cancels
Hold to Talk: a press under 0.3 s is a tap, and cancels
A mic that drops out mid-dictation stops, and what was captured is transcribed

After every ending

The state returns to Idle, and the overlay says what happened
A waiting refine gets empty input and exits, so its Claude does too
Command Mode uses the same states; the overlay says Editing while it polishes