Act as a senior AI systems engineer. Build an interactive tool called The AI Agent Evidence-to-Action Gate that evaluates whether an AI agent has enough reliable evidence to execute a proposed action.
Allow users to enter a proposed action, its expected impact, supporting evidence, evidence sources, confidence levels, reversibility, and potential consequences of being wrong.
The tool must:
Separate observed facts, inferred claims, and unsupported assumptions.
Identify missing evidence and conflicting sources.
Apply configurable approval rules based on risk and reversibility.
Produce one of four decisions: proceed, request more evidence, require human approval, or block.
Explain exactly which conditions triggered the decision.
Generate an audit record containing inputs, evidence, rule evaluations, and the final decision.
Include test cases for misleading evidence, stale data, incomplete context, and high-impact actions.
Create a polished, responsive interface with editable sample cases, clear decision explanations, and exportable JSON results. Use deterministic rules for the approval gate rather than pretending a confidence score guarantees correctness. Build a working prototype, not just a mockup.
0 Comments