Skip to content
Back to Reference

GARDEN glossary

Terms below are used with exactly the meaning given here throughout this documentation set. Each entry links to the principle section in GARDEN principles where the term is normative.

agent — An LLM-driven program that reads, writes, and reviews code via tools. Normative across all six principles; see GARDEN principles.

canonical name — The single name a concept has across the codebase; synonyms break grep-first discovery. Normative in G — Grep-first Discoverability.

context file — A short hand-maintained instruction file for agents (for example, AGENTS.md or CLAUDE.md) loaded at session start. Normative in N — Navigable Knowledge.

context rot — Degradation of agent performance as a session grows; primarily a state-tracking failure rather than a reasoning failure. Motivates N — Navigable Knowledge.

context window — The finite token budget an agent can attend to at once. Bounds the sizing rules in A — Atomic Vertical Slices.

deterministic gate — A check (type check, lint rule, test, or CI job) that passes or fails identically on every run, independent of any model. Normative in D — Deterministic Verification.

hop distance — The number of link or directory traversals between an edit site and the knowledge required to edit it safely; the GARDEN target is at most one hop. Normative in N — Navigable Knowledge.

indirection tax — The extra search-and-read cost every layer of indirection (deep inheritance, dynamic dispatch, dependency-injection magic) imposes on an agent; it also raises hallucination risk. Motivates G — Grep-first Discoverability and the rejections listed in R — Regenerable Components.

magic value — A literal whose meaning is not discoverable at its use site. Prohibited under E — Explicit Everything.

managed duplication — Tolerating duplicate code until a clone-detection signal in CI justifies extracting an abstraction, instead of abstracting preemptively. Normative in A — Atomic Vertical Slices.

one-context task — A task whose requisite context fits comfortably in a single agent context window. The sizing target for A — Atomic Vertical Slices.

premature abstraction — An abstraction created before at least three concrete usages exist; it costs agents more than the duplication it removes. Prohibited under A — Atomic Vertical Slices.

progressive disclosure — Layering knowledge so an agent loads a short summary first and follows links only when needed. Normative in N — Navigable Knowledge.

regenerability — The property that a component can be rewritten from scratch against its contract without ripple effects. Defines R — Regenerable Components.

requisite context — The minimum set of files and facts an agent must load to perform a task correctly. Sizing basis for A — Atomic Vertical Slices and N — Navigable Knowledge.

self-certification — An agent declaring its own output correct without a deterministic gate; forbidden under D — Deterministic Verification.

spec drift — Divergence between a spec or contract and the code that claims to implement it. Addressed in R — Regenerable Components.

temporal coupling — A hidden requirement that operations happen in a particular order, not expressed in types or signatures. Prohibited under E — Explicit Everything.

vertical slice — A module containing everything one capability needs (entry point, logic, data access, tests) rather than a horizontal layer. Defines A — Atomic Vertical Slices.