Production Quality

Overview

Production readiness is an escalation ladder — L0 (compiles) through L7 (documented). Most features should reach at least L3 before merge.

Escalation Ladder

Precommit gate (L0-L3 automated):

# mix.exs aliases
precommit: ["compile --warnings-as-errors", "deps.unlock --unused", "format", "credo --strict", "test"]

Common Mistakes

• •Skipping L2 (static analysis): Credo catches naming issues, long functions, and anti-patterns that code review misses
• •Treating L6 (observability) as optional: You can't fix what you can't see — add telemetry before production, not after the first incident
• •Unsafe migrations: Adding indexes or changing column types without concurrently: true or multi-step deployment causes downtime
• •Silent error handling: 92% of catastrophic failures come from incorrect error handling — test every {:error, _} branch
• •Zombie metrics: Metrics nobody acts on are waste. If it can't drive a decision, delete it

Reference Files

• •escalation-ladder.md — Full Production Readiness Ladder (L0-L7 with code examples, gate criteria)
• •testing.md — Testing strategy, TDD, property-based/StreamData, error handling imperative, documentation standards
• •security.md — SQL injection, XSS, CSRF, input validation, secrets management, timing attacks
• •observability.md — Telemetry layers, span conventions, tracing-as-analytics, alerting, gray failures, degraded mode
• •database.md — Safe Ecto migrations, isolation level warnings, dependency SLAs, performance guidelines
• •error-handling.md — Crash early patterns, strict-then-loosen, complexity analysis

Related Skills

• •elixir-patterns: GenServer, Supervisor, OTP patterns
• •cognitive-complexity: Ousterhout principles, deep modules, reducing complexity
• •performance-analyzer: Benchmarking, profiling, latency analysis