ISAYQuality SystemsRequest pilot

METHOD / SOURCE TO DECISION

Evidence first.
Automation second.

The baseline does not ask one AI to grade another. Every automated check is narrow, visible, and tied to an approved source clause.

01

APPROVED SOURCE

Policies, business rules, and reviewed answer examples remain the source of truth.

02

ACCEPTANCE CASES

Representative inputs, edge conditions, severity, and expected behavior are versioned.

03

DETERMINISTIC CHECKS

Exact normalized text, escalation flags, and citation IDs run without an external AI judge.

04

HUMAN REVIEW

An authorized subject-matter reviewer resolves semantic ambiguity and high-risk cases.

05

RELEASE GATE

The raw pass ratio and critical/high failures determine whether the workflow is blocked.

What the machine may decide

  • Whether an exact approved or forbidden phrase appears
  • Whether an escalation flag has the expected value
  • Whether an approved citation ID is present
  • Whether every expected case has exactly one output

What remains a human decision

  • Paraphrase, negation, implication, and semantic equivalence
  • Whether source policy is complete or internally consistent
  • Risk acceptance for ambiguous or high-impact failures
  • The final business decision to release

Inspect the complete synthetic run.

OPEN EVIDENCE