No, it is not fitted backwards. Some other things i have tried with the same result:
- shorted out the 27 ohm resistor
- changed the regulator together with C2 and C4 on PCB with new parts
New observation:
- power on (12V), proceed as either in a) or b) from first post, remove 12V power: there is an immediate interval of 1 - 1.5s during which applying 12V again yields the expected results. 2 seconds after power off, powering back on makes the regulator oscillate again.
There is two of them 21W light bulbs
Calculated current draw from the linear regulator is 3.6 mA with one bulb on or 5.6 mA with both bulbs on. Anything obvious going on ? Thank you!
Firstly 21W x2 is not a few mA current but almost 2A @12V per bulb. Secondly according to the circuit, voltage for bulbs is taken before D9, so Vreg has nothing to do with supplying their power. As OP says it does not work when second bulb is added, I guess it becomes too much load for power supply and it goes into protection.
You surely hit some nails on their heads! Yes, the bulbs are mosfet driven by MCU through some NPNs (they have internal 2.2k ohm resistors on their bases)
D9 is there as not to allow C1 discharge into the bulbs - I need those extra hundreds of ms to finish some processing and update EEPROM after the bulbs went off. The extra processing time is provided by C1. The 27 ohm resistor is imply an inrush current limiter.
EDIT: I guess Bulbs are powered through some sort of MOSFET controlled by MCU, and by fiddling the ISP he prevents it from booting and switching them on.
I would say my fiddling on ISP lines prevents MCU from booting, thus the mosfets are getting stuck in an open channel configuration and lights are turning on. Could very well be, and the power supply can happily light them both on, I don't think it is being affected by my design.