Read
Live
Morning EditionMonday, July 6, 2026Barcelona
01 · Teardown — Field Note

It was never about speed

I built a stock-analysis tool. Everyone assumes it's about "speed" — but speed was never the problem. The real problem was comparison: a raw ratio says nothing without the right peer set.

nocashflow.net · Finance Engineering ~5 min · TR/EN

Pulling a P/E in two seconds isn't a skill. One API call, one line of code. The tool's value wasn't there. The value was being able to say what that number means — and a ratio is meaningless until you know what it's being compared against.

The real problem

A 25× P/E, on its own, is a crumb of data. Ordinary at a software company, expensive at a utility, meaningless at a bank. The tool's job wasn't fast lookup; it was to normalize a name against its own sector peers and show where it sits. The output wasn't a number — it was a position.

Same ratio, three readingspeer set = meaning
P/EPeer setReading
25×SaaS peers (~35× median)cheap
25×Utility peers (~16× median)expensive
25×Wrong peer setnoise

What I built

Claude Code as the implementation agent, yfinance as the data source. Pull the fundamentals, normalize against a sector peer set, surface where the name sits relative to its comparables. The logic isn't complex; what's valuable is that it puts a raw ratio into a context.

The tool didn't fail at speed — it succeeded at speed, and speed was never the point. Both the value and the risk live in judgment.

Where it breaks

Without honesty the piece is worthless. It breaks in two places:

Breakpoint 01

Peer-set selection is still human judgment

The model can suggest a sector bucket. But deciding the "right" comparison set is the analyst's job: is Circle a payments company, crypto infrastructure, or a money-market proxy? That choice determines the result up front — and the tool can't own it.

Breakpoint 02

yfinance is not an official API

It's an unofficial scraper. Schemas change, rate limits hit, gaps appear silently — it's fragile. Fine for a personal tool; not defensible in a commercial context you sell or stake a client deliverable on. I use it knowing exactly that.

Principle

Knowing a tool's limits isn't the cost of using it; it's the license to use it. Choose the right comparison first, know what the data rests on underneath — the rest is just speed.

Code: github.com/<repo> — assumptions and limits written plainly in the README.

This is a financial/technical educational note, not investment advice. The ratios and thresholds mentioned are illustrative.

← Finance Engineering