OBEY Code Complete by Steve McConnell

When to use

Use as a compact always-on construction discipline for implementation, review, debugging, refactoring, and tuning.

Working code is not enough. Construction must lower defect risk, control complexity, and make human inspection cheap.

Clarify requirements, architecture fit, risks, conventions, and major construction decisions before coding from a solution idea.
Choose clarity, locality, explicitness, simple control flow, and consistent style over clever compactness or fashionable idioms.
Keep routines, classes, and modules cohesive, precisely named, small at the interface, encapsulated, and hard to misuse.
Make data meaning visible with names, constants, stronger types, closed states, deliberate initialization, units, and ownership.
Validate trust boundaries; use assertions, invariants, and contracts for programmer assumptions; keep error handling explicit and diagnosable.
Keep branches, loops, exits, exceptions, and table-driven logic simple enough to verify.
Build, test, review, debug, refactor, integrate, and tune in small evidence-based loops: root cause before fixes, behavior protection before refactoring, measurement before optimization.
Use comments, documentation, tools, and standards to reduce reader or manual effort, never to hide poor structure.

When a solution appears before the problem is clear, restate the requirement, constraints, and construction risks.
When readers must decode names, flags, primitives, units, states, or layout, model the meaning explicitly.
When a routine mixes phases or has a hard-to-use interface, split concerns or change the data model.
When input crosses a trust boundary, decide validation, rejection, recovery, assertion, and diagnostics.
When control flow, loops, exceptions, or branching tables are hard to inspect, simplify before adding logic.
When tests only prove happy paths, add boundary, invalid-input, defensive-check, contract, and data-driven cases.
When debugging, refactoring, or performance work starts from a guess, get evidence first.
When comments repeat obvious mechanics, rewrite the code or delete the comment; keep comments for intent and constraints.