>Reset:
>MCLR shorted with Vcc.

I've been running into a lot of trouble with this,  and it could
be your problem,  too.

I built a "breadboard" prototype of my first PIC product with
the cheap 1-transistor "brown-out" circuit from the databook,
and it seemed to be working fine.  When we did the first
pre-production boards,  we decided to try to save a few cents
by just tying MCLR to Vcc,  and have had frequent problems
with the chips not resetting correctly.

Some of the symptoms have been bizarre,  suggesting that
*some* of the instructions in the reset routine are not being
executed correctly.  E.g.,  once we had a case of the system
appearing to run normally,  except that a couple of the
output bits on port B weren't being driven.  But a couple
were,  so it looks like the TRISB value got scrambled.

I know from putting a scope on the power supply that we're
not getting a clean turn-on,  mostly because we're using
an alligator clip for a switch at the moment  ;-).  There are
warnings in the databook about this,  as I recall.

Another fun problem showed up while we were testing the
system on the bench:  if we leave the pulse generator that
simulates tach running when the CPU is  switched off,
the protection diodes feed enough energy into the power
supply to prevent a proper reset.  Any input that can source
more than a couple of milliamps can probably do this.

We've also seen a few cases where it looked like maybe the
WDT wasn't being initialized properly (we have it off,  too),
but I can't be sure about that:  it just looked like the system
was resetting itself,  so we powered off and on,  and the
problem went away.

Try putting a better reset circuit on the board.  Something
like a Dallas DS1233 would probably be better than the
1-transistor circuit from the databook,  but anything is
probably better than nothing.

Ran