Bob writes the change. PROOF tries to break it.

PROOF — Production Runtime Observation & Outcome Forecasting — is an AI-assisted developer workflow for finding reliability problems that traditional code review may miss. Its core principle is simple: “Bob writes the change. PROOF tries to break it.” PROOF analyzes a repository using three layers: deterministic static checks, controlled runtime experiments, and AI-assisted code review. Runtime scenarios include dependency outages and malformed service responses. The system records concrete evidence from these experiments and uses that evidence to determine whether a change should be BLOCKED or can PASS. AI findings are explicitly treated as hypotheses rather than proof. Only verified static or observed runtime evidence can produce a blocking verdict. This prevents the AI from inventing failures or making unsupported claims about application behavior. When verified problems are found, PROOF can use an AI Fix Agent to propose a code change. The repaired application is then restarted and tested again, demonstrating the complete workflow: Vulnerable → Detect → BLOCK → AI Fix → Apply → Verify → PASS The project uses IBM Bob IDE as a core development component, with Bob agent tasks used to analyze the architecture and extend PROOF with additional runtime failure scenarios and automated behavioral tests. PROOF is demonstrated entirely with synthetic, locally generated payment and fraud-service data. No production, personal, confidential, or customer data is required.