This week I noticed one limitation of machine learning when I looked at how Garmin’s PacePro predicts finish times. It spits out a number, but it doesn’t really explain how it got there. My guess is it pulls from logs, past pace, distance, maybe elevation, but it feels like a black box. That’s the same issue in many ML systems: they give an output without showing the reasoning.
Running teaches me what numbers can’t.
The problem is that running performance isn’t only about pace history. A watch doesn’t know if I slept poorly, skipped meals, or felt mentally strong. It ignores training structure, like whether I’ve built a strong base in zone 2 or if I’ve been overreaching. These human factors matter just as much, sometimes more, than raw GPS and heart rate data.
Matters of great concern should be treated lightly.
Yamamoto Tsunetomo
I see this clearly when planning races. A good race plan isn’t just a predicted finish time, it considers course terrain, weather, nutrition, pacing strategy, recovery, and how much zone 2 training built the runner’s aerobic foundation. Today one of my athletes ran the Gubat Half Marathon 12k in Sorsogon and matched his predicted and actual finish time almost perfectly. That worked because the plan blended data with my context, not because the watch alone dictated it.
What I learned is this: numbers are useful, but they aren’t the whole story. The real value comes when we combine them with insight and experience.
The real mastery comes from running smart, training steady, and not being bound by one rigid number.
OGRap 