I write specs for a living. Claude Code built my NBA forecast

I am a senior business analyst and AI product owner. My job involves determining what a product should do and documenting it for developers to implement. Recently, I created an NBA forecast called Clutch using Claude Code as the developer. This article explains how that process works, the rules I set, and the final output.

Clutch provides a daily NBA game prediction, assigning a percentage chance that the home team will win. The number is published to a public Git repository every night, hash-linked from the previous file to ensure integrity. After the final game, the prediction is graded publicly, comparing it to ESPN's BPI and the market's final price. The rule is simple: pick the home team as the winner.

The system operates entirely on GitHub Actions, running Python scripts on a schedule to fetch game data, injury reports, and market prices. The results are stored in a Postgres database (Supabase), and the information is presented through an iPhone app and a web app. The entire process is transparent, with the app and web app being a single Expo codebase that reads data from the database.

As the analyst, I define Clutch's functionality, write clear acceptance criteria, and review every change in the simulator and browser. I approve or reject any modifications before they go live. Claude Code handles the coding and testing, with over 1,400 commits from the project's four repositories, and around 95% of those commits list Claude as a co-author. Over 5,400 automated tests ensure the codebase is stable before any release.

If you're an analyst wanting to build your own product like this, start by writing down the acceptance criteria, limits, and prohibited words in clear specifications. Let Claude Code handle the coding, and focus on defining what the product should do. For each change, verify the tests, preview, and screenshots. Don't accept any changes without reviewing them thoroughly. Implement guard rails to control production writes, releases, and builds, while allowing Claude Code to handle testing and development.

Measure the product's performance honestly by setting comparison rules before seeing the results. Publish the outcomes, including the graded games alongside ESPN's BPI, market prices, and the home team as the winner. All data is available in plain JSON and CSV formats for transparency. I collaborated with Claude Code to write this post, just as I built the Clutch app itself.