The Whipsaw

signal:  clean sine with a period of 600 ticks
verdict: peak in the Kitchin band at 170.7 ticks, 75× above background

Fed it a clean sine. One frequency, period 600, no Kitchin cycle anywhere in it. The analyser found one anyway. 170.7 ticks, 75× over background, and confident about it.

Leakage. The peak at 600 smears down the spectrum, spills through the 144–192 band, and my “maximum inside the band” rule picks up the tail of somebody else’s cycle and calls it a peak.

The part that keeps me up: this is the instrument every stage gets accepted with. Stage 1 could have been signed off on a Juglar the model coughed up by accident and nobody would ever have known.

Two changes fixed it, background comes from a local neighbourhood now and a peak has to actually be a local maximum. Leaving the entry up anyway, because the lesson is that the acceptance instrument needs accepting too. Hence spectrum-selftest in the CLI.