Part
Holdout scenario
You might say: hidden test, holdout set
A scenario kept where the coding agent cannot read it. Only the verifier runs it, so the author cannot shape its code to pass that test alone.
- What it does
- Catches an agent that wrote code to pass the tests it could see, and not to do the job.
- Ask
- Which scenarios can the author run read, and which ones does only the verifier hold?
What to look for
- Scenarios stored outside the repository the agent works in
- A verifier run that reports results the author never saw
- The loop ends only when the hidden scenarios pass
Pattern
Sources
- Simon Willison, How StrongDM’s AI team build serious software without even looking at the code
That idea of treating scenarios as holdout sets … used to evaluate the software but not stored where the coding agents can see them … is fascinating.
- Justin McCarthy (StrongDM), Software Factories And The Agentic Moment, The Principles
The loop runs until the holdout scenarios pass (and stay passing)
- Justin McCarthy (StrongDM), Software Factories And The Agentic Moment, Software Factories And The Agentic Moment
We repurposed the word scenario to represent an end-to-end "user story", often stored outside the codebase (similar to a "holdout" set in model training)
Use it on your project
teardown-apply: hold "Holdout scenario" (holdout-scenario) from the teardown at https://teardownatlas.com/software-factory/ against my project
Paste it into a coding agent that has the teardown skills.
An AI agent wrote this map with the teardown skill. It can contain errors: check the sources before you rely on a part. About and legal