The App Was Done. But I Couldn't Trust the Data.

ChulbujiRunbuddyVibeCodingAICollaborationRunningBuildingInPublic

Today I ran 5.43km in 45 minutes and 53 seconds.

Average pace: 8’26”. Heart rate: 127 BPM. On paper, a decent run. But when I opened the split data, the cadence looked like this:

Segment 1: 200 spm. Segment 2: 3,074 spm. Segment 3: 3,648 spm.

No human takes 3,000 steps per minute. I knew right then — this app couldn’t be trusted.

Today's run summary — 5.43km / 45:53 / 8'26"
Today's run summary — 5.43km / 45:53 / 8'26"

I can’t read code. I can’t find the problem myself.

So I started with Meta-chulbuji. “The split cadence is way off. What do you think might be wrong?” Meta-chulbuji mapped out the possible causes. Armed with that, I asked Claude to run a diagnosis. Claude identified the structural issues, and I turned those findings into a brief — then handed it to Claude Code. Claude Code made the actual fixes.

My job was to judge and connect in between.

And it wasn’t just cadence. Pace was jumping between 3 and 14 min/km. Distance was skipping in 400m chunks. The watch and GPS were pulling in opposite directions and colliding. Five separate problems — each diagnosed separately, each handled separately. Not “fix everything.” One cause at a time. One fix at a time.

Run history — three days of data, building up
Run history — three days of data, building up

Verification still remains.

The fixes are in. But with a running app, you only know if it worked when you run. If the next time I open my split data and the cadence is quietly sitting somewhere between 160 and 180 spm — that’s when I can say it’s fixed.

Today wasn’t a day for adding features. It was the day a person who can’t read code, working with multiple AIs across clear steps, rebuilt the foundation for data worth trusting. The coaching comes after that.

Split data — Segment 2: 3,074 spm. Segment 3: 3,648 spm. That was the problem.
Split data — Segment 2: 3,074 spm. Segment 3: 3,648 spm. That was the problem.

Curious whether my cadence comes back to normal on the next run?


What I did: Fixed 5 data bugs by collaborating with multiple AIs — step by step. What broke: Cadence hit 3,648 spm. Pace swung between 3 and 14 min/km. What I learned: I don’t read code. I direct AI. That’s the workflow.