Most products start with a market gap. Mine started with a methods section. Here's what happened when I tried to turn three years of exercise physiology research into something cyclists could actually use.
The research first
During my master's programme, I spent a lot of time on a concept called durability — the ability of an athlete to maintain their physiological profile over a prolonged bout of exercise. Put simply: how much does your cycling performance drop off after two hours at moderate intensity?
This turns out to matter enormously. A cyclist with a high critical power (CP) but poor durability will fade badly in the back half of a long race. A rider with moderate CP but exceptional durability will outlast them. The numbers that most training tools focus on — FTP, power zones, VO2max — don't capture this.
We published papers on it. The first looked at whether carbohydrate ingestion during prolonged exercise helps maintain durability (it does). The second showed that durability can be predicted from markers of physiological decoupling you can already measure. Good science. Peer-reviewed. Published in the European Journal of Applied Physiology.
Then I started trying to explain it to cyclists, and realised we had a communication problem.
The gap between paper and practice
The papers are written for researchers. The methods sections describe laboratory protocols that require metabolic carts, venous blood draws, and carefully calibrated ergometers. The analysis involves statistical models that assume familiarity with concepts like ventilatory threshold identification and oxygen uptake kinetics.
None of that is accessible to an athlete sitting at home with a smart trainer and a Garmin file.
What they need is: upload your FIT file, get a number, understand what it means. A tiered classification — like a credit score, but for how well your physiology holds up over long efforts. Something actionable.
The Durability Analyzer is my attempt to build that bridge.
The product decisions the paper didn't make for me
Building the app forced me to make decisions the research doesn't specify. The hardest ones:
Where to set the tier boundaries. The science gives you a continuous distribution, not discrete categories. I had to decide what "Elite" means in numbers, and those thresholds are inherently somewhat arbitrary. I based them on published population data and my own judgment from working with athletes — but every threshold is a design decision, not a scientific finding.
How to explain it without losing people. The full explanation of durability involves CP/kg, VT1 drift, decoupling coefficients. Explaining all of that loses most users by paragraph two. I ended up building a tiered explanation — a headline score and tier badge visible immediately, with the full methodology available for those who want it.
Privacy. Cycling data is more personal than most people realise. A FIT file contains precise GPS location, heart rate throughout a ride, power output second by second. I made the decision to process everything client-side — the file never leaves the browser, no server sees it, no data is stored. This was a constraint that shaped the entire architecture.
The research answers "does this phenomenon exist and can we measure it?" The product answers "can a real person use this on a Tuesday morning?" These are different questions.
What the research background actually helped with
A lot, actually. The domain knowledge meant I didn't have to reverse-engineer the science from first principles — I knew what the meaningful variables were, what the literature said about normal ranges, and where the common misconceptions were.
More importantly, I had calibrated uncertainty. I knew which parts of the scoring model were well-supported by evidence and which parts were reasonable heuristics that needed validation. That distinction matters when you're deciding what to communicate confidently and where to hedge.
The research also gave me a clearer picture of the user. I've worked with cyclists. I've watched them interact with data. I know that most training apps overwhelm with numbers while underdelivering on meaning.
What came out of it
The app parses real FIT files — Garmin, Wahoo, anything that exports the standard format. It calculates your CP/kg across multiple effort windows, runs a durability score, classifies you into a tier with sex-specific thresholds, and gives you a plain-English summary of what that means for your training.
It also has a demo mode with synthetic ride data from a real-world athlete profile, so you can explore it before uploading anything.
Is it perfect? No. There are features I want to add — trend tracking over multiple rides, personalised intervention suggestions, a community benchmark dataset. These require solving some harder problems around data collection and privacy.
But it works. Real cyclists can upload real files and get real signal back. For a v1, that's the bar.
The lesson
Building a product from research you understand deeply is different from building from a domain you've just read about. The shortcuts I took were informed shortcuts. The decisions I made were grounded in something more than intuition.
But the product thinking — what to cut, what to explain, how to make it trustworthy without being overwhelming — that came from somewhere else. From thinking about the person on the other side of the screen. From shipping enough things to know what real-world friction looks like.
I want to keep doing both. The research and the building. They make each other better.