Safety Model

VibePro's safety model is based on bounded authority, current-head evidence, independent inspection, and fail-closed release operations.

Authority Boundaries

• Human: product intent, material trade-offs, waivers, and final release authority.
• Story / Architecture / Spec: outcome, structural boundary, and testable contract.
• Code and runtime: actual behavior; generated narratives cannot override them.
• Verification: observed outcomes tied to a commit and durable artifact.
• Independent reviewer / adjudicator: inspection and judgment from a separate execution identity.
• Gate DAG: readiness synthesis; it reports missing proof but does not invent it.

Brainbase may supply upstream context. Graphify, codebase-memory, Journey packs, external design prompts, and generated screenshots are supporting evidence. None becomes implementation truth merely by being available.

Fail-Closed States

• Missing or stale evidence remains needs_evidence.
• Required inspection that has not occurred remains needs_review.
• A violated condition remains blocked until fixed or handled by an explicit, attributable decision.
• A scanner that found no eligible targets is inconclusive, not proof of absence.
• Review records must include the correct stage, role, status (pass, needs_changes, or block), agent identity, inspection inputs, and a closed lifecycle.

Decisions and Waivers

vibepro decision record . \
  --id <story-id> \
  --type waiver \
  --summary "<accepted residual risk>" \
  --reason "<why this is acceptable>" \
  --artifact <evidence-path> \
  --reviewer <identity> \
  --status accepted

A waiver is visible debt, not a passing test. Keep the source gate/finding, reason, evidence, owner, and status explicit.

Release Boundary

guard check, pr prepare, pr create, and execute merge are the standard release path. Raw GitHub PR or merge commands bypass VibePro's current-head and waiver audit and should not be the normal path.