Debouncing the Strike: Filtering Noise from Signal

A praying mantis doesn’t lunge at every flicker of movement. Wind shifts a leaf, shadows cross the terrarium glass, condensation drips—the mantis stays motionless. But when a fruit fly lands and stays visible for half a second, the strike happens in 30 milliseconds. The mantis has a built-in debounce filter: ignore transients, wait for the signal to stabilize, then commit.

In electronics, mechanical switches bounce. Press a button and the contacts make-break-make-break dozens of times in the first 10 milliseconds before settling. Early microcontrollers saw each bounce as a separate press. The solution was debouncing: read the switch, wait a settling period (usually 20-50ms), read again. Only count it if both reads agree.

The Perl approach is procedural and explicit. Track the last change time, ignore any event within the debounce window:

use Time::HiRes qw(time);

my ($last_event, $debounce_ms) = (0, 50);

sub time_ms { int(time() * 1000) }
sub trigger_action { print "Strike!\n" }

sub on_input {
    my $now = time_ms();
    return if ($now - $last_event) < $debounce_ms;
    $last_event = $now;
    trigger_action();
}

Swift can express the same idea with closure-based state and timer dispatch:

import Dispatch

func debounce(ms: Int, action: @escaping () -> Void) -> () -> Void {
    var timer: DispatchWorkItem?
    return {
        timer?.cancel()
        timer = DispatchWorkItem { action() }
        DispatchQueue.main.asyncAfter(deadline: .now() + .milliseconds(ms), execute: timer!)
    }
}
let debouncedStrike = debounce(ms: 50) { print("Strike!") }

The mantis version: wait 500ms of continuous visual lock before committing metabolic energy to a strike. False positives waste calories. False negatives just mean the next fly. The threshold is tuned by evolution the same way we tuned capacitors and resistor values on debounce circuits.