So, the Frank Overton article about TrainingManager, and the blog posts from folks who've been beta-testing it, has made me so hungry to use the thing that I've ended up trying to replicate it in a spreadsheet. I've played around with different metrics for "chronic training load" and "acute training load"; e.g., rolling averages vs. straight sums over a range of days, using 7 days for acute and 28 for chronic vs. 10 and 45, etc. None of the metrics seems to make too big a difference; the gross shape of the TSB curve seems similar regardless. I settled on rolling averages, since it seems to make more sense to me than having workouts count for 100%, and then suddenly fall off a cliff to 0% after 7 or 28 days.
I'm considering "TSB" to simply by CTL - ATL. So, in my little model, positive TSB values indicate relative freshness, while negative values indicate relative staleness.
A few observations:
1. Grossly, the model "works." Looking back at mean maximal power data, strides in measured fitness tend to come after a month or two with high CTL followed by some sort of training interruption. For the most part, these mini-peaks have been unplanned on my part; they were more often the result of bad weather, not having time to train, etc., than a structured "recovery"
2. Layoffs have unexpected consequences. The model predicts the huge bump in performance I saw coming back from my 10-day layoff in Colorado; my TSB at that time was the largest it had been thus far in '06. The ATL fell through the floor, while the CTL takes a good while longer to decay.
A less obvious consequence of the layoff comes after resuming training. For a week or two, my CTL is still falling as it "digests" the long layoff. However, my ATL is back to my more typical 110-120 values, and this is causing a strongly negative TSB. However, I don't seem particularly stale; I'm putting up good, sometimes PB numbers for 3 minute power, 1 minute power, and 20 minute power, and I subjectively feel eager to train. Perhaps I'm not taking CTL over a long enough window?
3. Most discussions of "TSTWKT" have envisioned an application to peaking for a target event. It seems likely that it would work for this purpose. But what does the model have to say about build periods? Getting a good "picture" of the gains you've made seems to require a brief drop in ATL; in the long term, this drop in ATL translates into a drop in CTL as well. So, during a build period, is it better to just keep plugging away, in an effort to achieve the maximal possible CTL before tapering? Or, is it better to be able to accurately assess the effectiveness of the training you've been doing so far, and throttle back a bit to see what you can do when you're fresh?
