142M tokens. One day. 10 projects.

OpenAI Pro gives you a weekly Codex quota. I ran mine dry in 24 hours.
~30 autonomous coding tasks in parallel. One claude writing prompts. 644 tests at the end of the day.
Not toy projects. A full PM intelligence stack that feeds live data into itself.
What it does
I built 6 independent systems that each look at Polymarket from a different angle.
One scores 222 whale wallets and ranks them S through F. One generates daily prediction cards with CLOB momentum data. One runs geopolitical scenario trees through Claude and maps them to live markets. One scans for calibration mispricings. One tracks my own 490-trade history and tells me where I leak money. One watches what S-tier wallets are buying right now.
Then I built a seventh system that connects the other six.
The signal bus
Each system writes signals to a shared database. The bus reads them all and asks one question: do multiple independent systems agree on the same market?
First live run. Two systems converged on Kostyantynivka.
The geo simulator built a Ukraine scenario tree and concluded the market was 14pp overpriced. The prediction card engine — completely separate codebase, different methodology, no shared data flagged the same market as overpriced on the same side.
Neither system knew the other existed. That is not a coincidence I designed. That is emergent behavior from infrastructure.
Why this matters
One system calling a market overpriced is an opinion. Two independent systems calling the same market overpriced on the same side is a signal.
The more systems you build, the more convergences you find. The stack compounds. Every new data source makes every other system more useful.
The boring part nobody talks about
Auto-resolvers that check if past calls were right. Accuracy dashboards that track whether your predictions actually work. Tier history databases that notice when a whale gets promoted or demoted. Weekly cron jobs that refresh everything while you sleep.
None of this is exciting. All of it is what separates a tweet from a track record.
142M tokens. 644 tests. The infra is done. Now it runs.