All good questions.
By observation (with a scope) the DCSRC only generate a single watchdog burst about 3/4 second after it starts up (or following a reset). This burst is less than 1/10th sec with timing that varies a bit whether DC or AC track voltage as previously mentioned. After this single burst the DCSRC watchdog returns to his doghouse and goes to sleep!
It is the TIU watchdog which barks for 45 sec after initial power up; this period has apparently varied with different TIU revisions. But that is another investigation which is sort of irrelevant for the PBW.
So my latest thinking is reset the DCSRC as frequently as possible but not too frequently that it never lets that one-time bark come out.
I figure a 10% (or so) window is a reasonable objective (vs. a 1% window or whatever) easily obtained with a modest range of "tuning" values. Plus, since this is sort of a 3-blind-mice situation (no published requirements on watchdog timing)...and since paying customers are playing the role of beta testers 
, a tighter spec is better for controlling the rollout. That's just my opinion of course.
Ahh, I was going to offer a gold star to anyone who could second-guess why I did not use a 10-cent 555 timer IC which has better timing tolerance. The 555 IC chip has relatively high operating current as it is a bipolar device (vs. the CMOS 74HC14). Since I am stealing power from the DCSRC microprocessor circuit, I wanted minimal loading of the DCSRC circuit. Yes, there is a CMOS version of the 555 chip but it is relatively expensive and not so common in the average DIY'ers parts stash.
In retrospect, I never expected anyone else to make one given where the thread was going at the time! But it's all good. 
