← Glossary
Quality gates
Quality gates are the smallest checks that prevent predictable damage. They are not bureaucracy. They are risk controls—especially for high-stakes flows, regulated contexts, and agentic automation.
Definition
- A quality gate is a pre-defined release criterion applied to a specific scope (a flow, a pattern, an integration).
- Good gates are risk-based, testable, and tied to measurable acceptance criteria.
- Gates should increase speed by reducing late-stage reversals—not slow delivery.
- A gate is effective only if it has a clear pass/fail signal and a pragmatic remediation path.
Why it matters
- Most production incidents come from known categories: missing states, missing recovery, missing instrumentation, unsafe autonomy.
- Quality gates protect trust-sensitive journeys where failure cost is real.
- Gates enable distributed teams: a shared definition of “done” that holds across org boundaries.
- In practice, this is where many digital programs fail: the concept is understood, but the operating discipline is missing.
Common failure modes
- Gate sprawl: too many checks applied to everything, so teams route around them.
- Opinion gates: reviews that generate debate but no measurable criteria.
- Late gates: checks applied at the end, when it’s too expensive to fix.
- No enforcement: gates exist on paper, but releases ignore them.
- Gates without learning: criteria never update after incidents.
- Gates that ignore operator reality: criteria pass in theory, but support cannot handle the live flow when it breaks.
How I design it
- Start with a short set for critical journeys: state model completeness, recoverability, accessibility, auditability, instrumentation.
- Attach owners and evidence: what proves the gate is met (tests, screenshots, logs, memos).
- Move gates upstream: review states and recovery during definition, not after build.
- Make exceptions explicit: if you bypass a gate, record who accepted the risk and why.
- Update gates after incidents: convert failures into new criteria or better patterns.
- Keep gates visible: dashboards or checklists that teams can self-serve without waiting for approval.
- Treat it as a repeatable pattern: define it, test it in production, measure it, and evolve it with evidence.
Related work
Proof map claims
Case studies
See also
Contact
Let’s discuss a leadership role, advisory work, or a complex product challenge.