▶ YouTube Summaries

• YT :: https://www.youtube.com/watch?v=9ANJjK06A54
• Original title :: Your Tests Pass. Your Software Still Sucks.

A green test suite with high coverage and hundreds of passing cases can still fail to answer the only question that matters: can the system change without hurting customers, revenue, operations, or developer judgment? The host calls this "confidence without evidence"

• a test suite that looks like an instrument panel with correctly

lit gauges while measuring the wrong failure modes entirely.

More tests create more signal, not automatically more quality; someone still has to separate real signal from theater. Chasing project-wide coverage numbers above roughly 90% is usually not worth the cost, while per-change coverage (stopping new untested code from sneaking in) is more useful than a global percentage. Heavy mocking makes this worse by testing implementation choreography instead of behavior, so 50 tests can explode on a harmless refactor because they were guarding yesterday's design, not today's risk.

Flaky tests are singled out as a quiet cultural poison: a cited survey found 59% of developers deal with flaky tests monthly, weekly, or daily, and once a team starts treating red builds as "probably unrelated, just rerun it," the suite stops functioning as a safety system and becomes a tax with a progress bar.

The fix isn't fewer tests - most teams are still under-tested - but tying every test to a real, nameable risk: what decision does this test support (a release, a regression, an interface contract, a performance budget)? Not every risk belongs in a unit test; some belong in synthetic monitoring, canary releases, feature flags, observability, or manual review for rare high-impact changes. Practical hygiene: track flaky tests as debt with owners and expiration dates, budget suite duration so people don't route around it, delete low-value implementation-detail tests, and after every incident ask what signal - test, monitor, or rollout control - could have caught it earlier.