dogfood

To smoke test a new script, run an evidence-based fix-through loop that repeatedly exercises the target, diagnoses in-cone failures, and applies fixes until the pipeline converges or triggers a replan.

Dogfooding is running your own newly built pipelines or scripts through a smoke-driven validation loop to capture evidence and force iterative fixes until the end-to-end goal succeeds, preventing drift into false success.

Smoke tests pass but pipelines fail when validation only logs issues instead of forcing a binary fix-through outcome; you need a loop that mandates per-finding classification and iterates on failures until convergence.

To run idempotent re-runs during post-build validation, enforce a pre-flight guardrail that ensures repeated invocations of the target produce machine-parseable progress artifacts and explicit per-pass mode artifacts for convergence tracking.

You should switch from iterative fixing to replanning when the smoke-driven fix-through loop exhausts tiered coverage options, including narrow, broad, and shakedown passes, without achieving convergence on the invocation target.

You can control exercised surfaces during pipeline smoke testing by selecting tiered coverage modes—narrow, broad, or shakedown—which dictate the scope of validation and when PM check-ins occur.