AJA #20: The Pipeline Problem

It's a strange time to be a cyclist, a bike shop, or in the bike industry. Pro teams are scrounging for components. New frames are difficult to come by. Bike shops, swamped with extraordinary demand, are closing due to lack of supply. As an innovator in the bike industry to talks to bike shops and pro teams every day, Josh has practical insight into making the best of a problematic pipeline.

We also talk about how Silca goes about greenlighting a new product, marginal gains on gravel, and what Josh really wants marginal gainers to think about when we're on our bikes.

Got a question you’d like to ask? Text or leave a voicemail at the Marginal Gains Hotline: +1-317-343-4506 or just leave a comment in this post!

Subscribe using your favorite podcast platform (but be sure to rate and review us on Apple Podcasts).


  • pete ying

    on the podcast you go into the concept of single use longevity for lube on the chain. but what about a related concept/issue? does that longevity change depending on the interval between application and use? eg some

    lubes need drying or curing time but aside from that, let’s say you lube a chain and it sits in the garage for 2 weeks/3 weeks/4 weeks before use. does that affect use longevity? similarly if it sits in the garage versus inside an apartment etc., how does that change things? other corollary questions would be does lube last equivalent times if you do 1 100 miler versus 2 50 milers 1 day apart or 2 days apart 1 week apart etc?

  • Weiwen Ng


    I have a conceptual question on how meaningful marginal gains are. Last year, and I’m paraphrasing, you said that Primoz Roglic looked like he was having a bad day on that last TT, but you could name a bunch of marginal gains and you think that the sum of their expected value was roughly equal to his losing margin. This year, he looked like he was doing better. Never mind that he crashed out, the point is that our performance fluctuates. We can control some of that fluctuation, but not all of it.

    Let’s say I have a Gran Fondo in a week, and I measured my FTP at 200W. Really, I should be assuming that my FTP on Gran Fondo day is a random variable with a mean of 200W and some standard deviation. Imagine that I can identify 10W worth of marginal gains that I think are cost-effective. Now, imagine that my FTP has a normal distribution for simplicity – it can’t really be normal, that implies that the variation is symmetric when there has to be a physical upper bound, but this is for ease of illustration. Imagine that the standard deviation is 10W – comparable to the marginal gains I identified. If you plot two normal density functions with standard deviations of 10 and means of 200 and 210, they overlap a non-trivial amount. Basically, there will be quite a few days when my pre-gains self would have beaten my post-gains self.

    If we are talking more like a scenario where I am doing an FTP test on Zwift, and imagine I time traveled back to the morning of the same day and erased all my physical and mental fatigue and ran that test a bunch of times, there’s still going to be some variation in my actual performance there. I am not talking about your power meter’s +/-2% accuracy, I am talking my actual performance. But that variation is going to be a lot smaller. The further back I time travel, the more variation there is.

    Anyway, within individuals, how much variation in performance are we expecting compared to the magnitude of potential marginal gains? I can tell you that in 2018, I had my FTP measured at 175W, because I was way out of shape. I got back into riding, and I went on a Zwift bender this winter. I think I measured it at 210-220W late 2020, and got it up to nearly 250 in March 2021 or so. Then after some real life stuff, I missed some riding time and I think it’s back down to the 220s. Which is fine, I can get it back – but to me that illustrates the possible magnitude of my intra-individual standard deviation under real life conditions. If have marginal gains that I think are cost effective, I should still do them, as I am unquestionably better off on average. This question is more about how often I am better off. If you increase the standard deviation, the two hypothetical normal curves overlap more. If you reduce the standard deviation, they overlap less – so in my hypothetical scenario of a large number of Zwift trials across multiple time travel sessions, the amount of overlap will be smaller, and the selves that waxed their chain will have a higher FTP in a clear majority of those trials.

    Moving from intra-individual to inter-individual variation, I think that among top athletes, the variation between different individuals’ mean FTPs has got to be fairly narrow; the time gaps in Grand Tours have to reflect how things compound from day to day plus the influence of adverse events. The variation between individual means at the amateur level has got to be pretty big. In that sense, while I’m still waxing my chain, it is not going to vault me from Cat 4 to 3 on its own, let alone from 2 to 1 or 1 to pro.

    For reference, my sense is that the distribution of FTPs should be left skewed by quite a bit, so definitely not a normal distribution (which is symmetric). You’re typically resting and prepping for an FTP test, so that test should identify something on the high side of your potential FTPs, i.e. there’s not much distance above your average FTP from your recent test, and there’s a longer distance below it. There has to be a physical limit. A normal distribution would imply that if my mean FTP was 200W, there’s a small but non-zero chance that I could crank out 230W or better on a day, and it’s equally as likely me only being able to do 170W or less. That can’t be. If anyone is familiar with the skew normal distribution, I’m thinking something like that with a negative shape parameter, e.g. alpha = -4 or lower in the parameterization on Wikipedia. I am not familiar with the skew normal distribution or other alternative distributions, so I haven’t been able to graph those scenarios out. However, I think my statement that the area of overlap is fairly large would still generalize to left-skewed probability distributions.

  • Andy Walford

    A waxing question for Josh. I use a slow cooker to rewax my chains. Does the fact that the cookers heats up through induction make any difference to the quality of the penetration etc? The chain is also being heated by induction as well as the conduction via the heated wax.

    It probably marginal on top of marginal but I,m intrigued!

    Also on a separate topic, do tyres lose pressure quicker when being ridden rather than just sitting? The rapid fluctuations in pressure over surface unevenesses might serve to raise the average pressure for a given volume over time?


    Andy Walford

Leave a comment

Please note, comments must be approved before they are published

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.