Skip to content
AIGP

RFC-032: Post-Hoc Evaluation Loop — 12. Burden of the Inquiry

AIGP Specification › RFC-032: Post-Hoc Evaluation Loop › 12. Burden of the Inquiry

← 11. Schema · Section index · 13. Security Considerations →

12. Burden of the Inquiry

This RFC meets the burden of the inquiry — the question of whether governance produces quality — through the following structural guarantees:

12.1 Non-Circularity

The system cannot claim “what happened is what should have happened” because:

  • ANTICIPATE is written before execution. It defines success criteria before the outcome is known.
  • ANTICIPATE is sealed and immutable. It cannot be retroactively adjusted to match the outcome.
  • VERIFY compares against ANTICIPATE, not against RECORD alone. The standard of comparison is the prior expectation, not the observed result.
  • D-DNA signing provides tamper evidence. Any modification to a sealed ANTICIPATE is cryptographically detectable.

12.2 Separation of Concerns

The system separates the four distinct functions that must be independent:

Function Actor Cannot Be
Define expectations Application, Governance Authority, or External Evaluator The executing agent
Execute the work Agent The grader
Grade the outcome CODE/MODEL/HUMAN grader The executing agent
Enforce the verdict Governance Authority The executing agent or the grader

No entity evaluates itself. No entity sets its own success criteria after the fact.

12.3 Empirical Falsifiability

Every ANTICIPATE declaration produces a testable hypothesis: “If execution proceeds under this governance configuration, the outcome will satisfy these criteria.” VERIFY tests that hypothesis. A MISMATCH is a falsification — empirical evidence that governance did not achieve its intended purpose.

This makes governance quality measurable:

Governance Efficacy = MATCH_rate / (MATCH_rate + MISMATCH_rate + VIOLATION_rate)

A governance configuration with low efficacy is not producing quality, regardless of how many rules it enforces. A governance configuration with high efficacy is demonstrably producing anticipated outcomes — the closest a non-causal system can come to proving that governance contributes to quality.

12.4 The Bridge Condition Satisfied

Recall the foundational argument:

Governance + Quality_Definition + Enforcement_of_Definition → Quality

RFC-032 provides:

  • Quality_Definition: ANTICIPATE (explicit, measurable, pre-declared criteria)
  • Enforcement_of_Definition: VERIFY → governance_action (MISMATCH triggers scope narrowing, circuit breaking, or escalation)

With this RFC, AIGP satisfies the bridging condition. Governance alone does not produce quality. But governance armed with explicit anticipations and post-hoc verification enforces quality — not causally, but through the continuous empirical discipline of declaring expectations, measuring outcomes, and adapting governance in response to divergence.

← 11. Schema · Section index · 13. Security Considerations →