It was the radial cap. Which is odd, because it measured fine.
Guess I really have to get round to buying an ESR meter one of these days...
If you would only use it occasionally, you could use a siggen, resistor and scope.
Apply a square wave amplitude dV to the C in series with R, and measure the voltage across the C. If the capacitor is perfect then the voltage across it will be the standard RC asymptotic exponential rise dV(1-e
-t/RC). With a series resistor, you will see a step at t=0, followed by that standard rise. Assuming R>>R
ESR, initially current I through the capacitor is I=
dV/
R, and the size of the step is simply I*R
ESR.
Choose the frequency, dV and R to get a convenient step, and have a DC offset voltage if that is appropriate.
This technique has the advantage that it closely mimics the actual operation in an SMPS, which a generic ESR meter won't.