204: Instruments That Cannot Fail
How do we know our checks are checking? At project scale a check is software in its own right: a guard in a commit hook, a gate that computes done, a fence over a schema, a step in a release pipeline. When an agent builds one, it can report a confident verdict over rows it never opened. 104 taught one person to see a check go red; this chapter turns the same question on the instruments themselves. A control has to be able to fail, and an instrument has to say what it did not look at. A silent success, whose output matches the run that was meant, is found by sweeping the source for its class, and a fence keeps a class closed by asking the system what exists. A verdict names the tree and the artefact it measured, and a release is checked as the user receives it.
Materials
This Chapter's Insights
- ACI-038: Controls That Can Fail
- ACI-039: An Instrument Must Say What It Did Not Look At
- ACI-040: Silent Success
- ACI-041: The Fence
- ACI-042: A Verdict Names Its Tree
- ACI-043: Landed Is Not Working
Browse every insight on the course map.