# Governance Modonome treats autonomy as something to be earned, one rung at a time. ## Maker, checker, merger Every change needs three distinct identities: - A maker that creates the diff. - A checker that did not create the diff. The checker runs the gates first, then reviews the diff and the rationale. A persuasive rationale is evidence, not a verdict. - A merge authority that is not the author. If only one identity or model is available, autonomous checking and merging are disabled and the change waits for human review. ## Risk tiers - Tier 1: mechanical, small, test-fenced, no protected paths. Local maker and checker. - Tier 2: multi-file but bounded, no public contracts. Stronger checker, full gates. - Tier 3: public API, schema, migration, dependency, security, CI, auth, secrets, protected docs. Owner or frontier review. - Tier 4: new product claims, architecture, policy, legal, market positioning, autonomous merge enablement, cross-repo changes. Owner decision only. Tier 3 and Tier 4 changes route to the same human review and security processes a team already runs, through CODEOWNERS and owner decision. Modonome extends those controls and works within them, rather than adding a parallel approval path. ## The activation ladder 1. Disabled. Read state and produce plans, read-only. 2. Dry-run. Project actions and record metrics, read-only. 3. Shadow. Run against live state read-only, compare decisions against human outcomes. 4. Armed. Allowed only when every gate below passes. Each rung is earned by a clean record on the rung before it. ## Arming gates Armed mode requires all of: branch protection or equivalent enforcement, required CI checks, code-owner review, a verified trusted author, protected-path rules, a separate merge identity, a daily merge cap above zero, a remote budget cap, and a rollback path. The arming levers are read from the environment or CI. ## Owner decisions Open questions live in `.modonome/DECISIONS.md`. Each carries a recommendation and a default of hold. Unanswered means no action. ## Learning Lessons are captured only on a real correction signal, staged in `.modonome/LEARNINGS.md`, bounded and dated and evidence-backed, and promoted into canonical rules only by an owner. Promotion deletes the staged entry and, where it fits, adds a deterministic gate.