Hopefully as we have really gone in deep on this, some people might find this interesting.
So for my normal schedule, over 6 weeks, I ran with the wrist paced power set up, linked to my nearest weather station 1 minute before I started the run and pressed start on the Garmin. This was a mixture of easy runs, long easy runs, 10x1k , 6x 1600 and 3x 3200.
My FTP was set up with the assumption ( this is what I will continue to use) that a 5k is equivalent to a 10 mile TT or 20 mins and that 94% of that will suffice. It's not my real FTP, it's just a guess like anyone really but that is going to be my protocol from now until the end of time. For what it's worth, I always used this when cycling and my 10 mile PB is near 18 min flat and my 5k is around 17 mind flat. So in terms of aerobic effort, they are very close. We can argue on the 94%. If coggan was here, he would argue. But I've known plenty of people use this and it works over time. FTP is only relevant to me and the only thing I consider important really is it's in the range of realistic and highly repeatable in how I work it out.
Anyway, onto the data. So all my runs are either giant loops, or out and back repeats with an even number of runs into the wind and back or vice versa. I thought I would add that in.
So here is the huge issue. The watch based power, requires elevation in its calculations. Whilst that is usually accurate, as part of this testing I recalibrated manually before the start of each run. However every now and again there might be a random 5 foot spike up or down ( just as an example) despite clearly me being on a flat, sea level plain with literally the sea at eye level to my left or right, during the run.
So you would get a huge spike (or decrease) in power for a short duration, if you were looking at the power gauge in real time. This would obviously completely skew the data, over say a relatively short 1km repeat. However, the caveat to that, is when it measured the elevation as pan flat, say picking HM effort (not pace) as the example, it measured the power VERY accurately both ways, compared to the styrd. So it must accept my frontal area as less. Again, this must be factored into the equation, as it has no actual idea of the force being applied. So the "power" number is just an approximation of a giant algorithm, rather than a mechanical measurement. Note, if anyone ever tried the Cyclops "chest" based cycling power meter about 10 years ago, it reminds me slightly of this. Sometimes it plays ball, sometimes it doesn't.
Ok, so the problem. How to deal with the spikes. What I did was load all my runs into golden cheetah and use the editing software to flatten out the spikes in elevation/ power That's when things became much more interesting and usable.
All easy runs then fitted neatly into between 76-79% of FTP.
All 1k repeats were within 102- 103% of FTP.
All 1600 repeats were within 99- 100% of actual FTP .
And 3200 repeats were within 96- 97% of FTP.
This suddenly is incredibly useful information on a narrow range that is working for me. The pace was a bit scattered, so in this sense power is a much more useful and narrow margin of error metric to use.
Ok, so the big problem. I had to smooth out all those power spikes. It had zero use to me on the go or in the session in real time. Sometimes, before the data was smoothed out, there could have been 20w difference between workouts of the same duration. So on the go, really I would have no idea where I am. Anyway, I found it incredibly helpful to look back in retrospect. It's just such a shame, there is no way to see it with this accuracy in real time.
A final note, these are "harder" than cycling sweetspot zones. But the final say on that, cycling sweetspot you might do 3x 15 or 3x 20, so in a lesser percentage of FTP range, (96% was the lowest I did in running) which is offset by the longer reps and for a longer amount of time. Whereas in my running, they are shorter but a little bit harder, even the 3ks are probably 5 mins shorter than any sweetspot reps I would do on the bike (where I would target maybe 89- 93% of FTP).