The payment succeeded. The agent crashed.
What should recovery do? A real Stripe test-mode experiment, its failed assertion, and the limits of the evidence.
The blog
Thinking, in the open.
Case studies, useful methods and short notes from building with AI.
No matching entries. Try another topic or reset the filters.
↗
What counts as evidence that AI work is done?
A practical record for connecting a request, an action and an observable outcome.
Why I kept the failed result
A failed assertion can be useful evidence. Removing it would have made this experiment harder to understand.