Execution is solved. The decision feeding it is still written for an average day.

A peak sale. Tens of thousands of transactions in a single afternoon. One PSP quietly degrades. Nothing is broken. Nothing throws an error. The routing rule keeps sending traffic exactly where it was told to, months earlier, by someone who could not have known this hour would look like this.

By the time anyone reads the dashboard, the sale is over.

That is not a vendor failure. The orchestration executed perfectly. It did what it was configured to do. The failure is structural: the decision feeding that execution was written in advance, for an average day, and the day was not average.

Payments is never average. Payments is all edges.

I did not want to build another system that executes well. I wanted to build the one that decides what should be executed — for this transaction, right now, from what is true this minute.

So I built it. Pillar8 is that layer: it sits above the orchestrator, returns a decision before execution, and learns from every outcome. Its checkout decisioning ran live for real shoppers at a Shopify Plus merchant, measured against a control group.