A pipeline failed twice. The second error hid the first.

The audit subprocess exited with code 7 and returned invalid JSON. The wrapper tried to parse it before preserving the process failure, so JSONDecodeError replaced the operational error the developer actually needed.

The fix was tiny. The useful model was this:

• failure + valid JSON → save the receipt, then re-raise • failure + invalid JSON → preserve the subprocess error; write no receipt • success + invalid JSON → surface the broken output contract

That matrix turned “it failed” into a reproducible diagnosis.

Full case, including the regression tests: https://dev.to/relaunchdept/the-json-error-that-hid-the-real-failure-5da6?utm_source=contra&utm_medium=community&utm_campaign=json_failure_followthrough

Need the same evidence-first review on a broken automation? Fixed-scope audit: https://buy.stripe.com/bJe6oJ8PQ3lKgULaCrefC02?utm_source=contra&utm_medium=community&utm_campaign=json_failure_followthrough

AI-assisted analysis, human-reviewed against public code and tests. Operational debugging only; no guaranteed outcome.