Triage Modes: A, B, and L

Mode A is the lightweight path. It applies when a change touches only governance, configuration, or documentation files, adds no new dependencies, and stays well under a couple of hundred lines. Mode A uses the Maker-then-Checker pattern: one agent makes the change, a fresh agent checks it. No full role pipeline is needed because the blast radius is small and fully reversible.

Mode B is the full five-role pipeline. It applies whenever application code is touched, a schema migration is involved, a new dependency is introduced, a user-visible flow changes, or core governance files are modified. Mode B gives you impact analysis, a gated TRD, parallel Devs, independent Test agents, and specialist Reviewer agents — the complete quality discipline. When in doubt, choose Mode B; the cost of an unnecessary review is low, and the cost of skipping a necessary one is high.

Mode L is the autonomous multi-milestone program. It applies when the scope is large enough to span multiple requirements and milestones, the success criteria are machine-checkable, the stack is locked, and the owner has granted full autonomous authority. Mode L runs the EM intake, TL analysis, TRD gating, Dev dispatch, Test, and Review pipeline for each REQ in sequence, pausing only at human-approval gates.

The Mode L loop continues until the program reaches its exit gate or encounters a human-blocked decision. Each milestone retires a set of acceptance criteria. The owner is notified at checkpoints but is not required to intervene between them. At the end, a human approves each merge — the loop never pushes or deploys autonomously.