Here is my setup:

PIC14000 acting as I2C master sending 5 bytes to a PIC16C63 acting as I2C slave.
PIC16C63 buffers the I2C data and sends it to a PC via RS-232.  I can look on a
scope and see the 5 I2C bytes followed by the 5 (plus checksum, total 6) bytes
on the RS232 line.

One of the bytes from the PIC14000 indicates how many data bytes are in the
message.  There is a three byte header and two data bytes for this test.
I know the protocol of the I2C doesn't need a "number of bytes" parameter; that
info is used by the RS232 portion of the 16C63, so it has to stay.

Everything works fine until I place my hand between my 100W halogen desk lamp
and the PIC14000.  The PIC14000 still only sends 5 bytes but the PIC16C63 starts
sending lots of data (lots and lots...).  Presumably the 14000 is telling the
 16C63
that there are more than 2 data bytes.  This is strange, but what is also
 stranger
is that it only works with the light _on_.  When I block the light or turn it
 off the
PIC losses its mind.  This doesn't bode well for a dark, windowless OTP 14000.
BTW, the 16C63 doesn't have any problems (that I can detect).

The first think I thought of is the internal 4Mhz clock of the 14000 is
 fluctuating
enough that the I2C master clock and data timing are moving outside the limits.
That isn't happening.  I confirmed it on the scope.

Any ideas?  Will the OTP parts exhibit the same behavior?

-tim drury