An assistant can organize a failure investigation when it receives relevant logs, configuration, requirements, and provenance. Begin with a focused evidence bundle, then expand when a hypothesis depends on earlier events or another component. The first reported error is a starting point, not necessarily the root cause.
Lesson | What it covers |
|---|---|
First error, context lines, seed and config in one small slice. | |
Guesses ranked with the evidence that would confirm each. | |
Normalised-message clustering with a runnable Python program. | |
Check existing evidence, then choose a discriminating experiment when needed. |
A slice: the first error with context lines, the test name and seed, and the config. The full log stays local until a wider search is needed.