Concept Map

codeArbiter is easiest to understand as a chain:

1. A request enters a lane. The lane defines the work, its required evidence, and its stops.
2. Material choices are explicit. The user decides hard gates; SMARTS records bounded autonomous choices; ADRs preserve architectural choices.
3. The repository carries the context. Provenance and just-in-time injection keep agents aligned with the checked-in rules that govern the files they touch.
4. Independent reviewers challenge the result. Checkpoints and lane-specific reviewers produce findings without silently rewriting the author’s work.
5. The audit record explains what happened. Commits, decisions, overrides, questions, sprint choices, and reports can be assembled for a range after the work is done.

Start with the problem you are trying to understand

01Work routingThe gated-lane modelWhy must each kind of work follow a different path?02AutonomySMARTSHow does an autonomous sprint choose without asking every time?03ProtectionEnforcement & SecurityWhich protections are deterministic and which are advisory?04DecisionsADRs and the decision logHow is a long-lived architecture choice recorded?05Repository truthProvenance and context driftHow do repository facts stay current as source changes?06ContextJust-in-time context injectionHow does the right rule reach an agent at file-read time?07SeparationThe persona-register splitWhy are authors, reviewers, and routing kept separate?08Review cadenceCheckpointsWhat catches drift outside one feature lane?09EvidenceAuditabilityWhat evidence can be reconstructed later?10Design historySelected hardening notesWhy does a particular hardening control exist?

A concrete example

Suppose a feature adds an authentication dependency and changes a security entry point. The feature lane provides the specification, tests, and author/reviewer sequence. The dependency lane vets the new package before installation. Reading the security file injects the controlling security document. An unresolved trust-boundary choice becomes an ADR instead of an implementation guess. The commit gate requires the applicable evidence, and the final audit can show the decision, dependency review, security findings, override history, and commit together.

No single concept is the product. The value comes from their composition: the lane limits what may happen, repository context says what must remain true, reviewers challenge the result, and the audit trail makes the path inspectable.

Put the model to work

• To complete a task, choose a page under Workflows and follow its host-native command.
• To investigate a block, start with Troubleshooting and the Hook Gates reference.
• To inspect exact command, skill, or agent behavior, use All Reference.