I agree, but your analysis is missing something: the canary-threatener may be secretly, without a public ruling, within the refresh interval, be convinced that if he doesn't keep updating (falsely) the canary, he'll go to jail. In other words, the precommitment to stop producing the canary signal isn't fully credible (though it seems much more likely to get the message out than a promise to actively say if something happens).