OK, I disconnected all the rails and it still wouldn't come up. Replaced
the TEA1522 and it came up OK with only the 5V rail active. Fair enough,
reconnected each rail in turn, turns out the SR560 Schottky on the 12V
rail had shorted *again*.

I'm now onto my last TEA1522 and getting more will be "entertaining" to
say the least (and probably involve buying £10 worth of them). What gets
me is that the shorted diode did NOT cause the PSU to go into shutdown:
it blew the TEA instead...

The shorted-turn trigger trips when the voltage at SOURCE is 0.75V. Thus,
the 1R resistor that's in there now will cut the power when the primary
pulls 0.75A, and the chip is rated to a drain current of 1A. This seems a
bit close to me -- so I'm thinking of adding about 10R to this (two 0.7W
22R resistors in parallel to share the load). Good idea, bad idea?

I'd try the Lightbulb Trick, but I'm willing to bet the contents of my
wallet that it won't offer much protection against a secondary side short.

--
Phil.

http://www.philpem.me.uk/
If mail bounces, replace "10" with the last two digits of the current year