Skip to content
AIGP

RFC-026: Human Feedback Signal — 1. Problem Statement

AIGP Specification › RFC-026: Human Feedback Signal › 1. Problem Statement

Section index · 2. Proposed Extension: FEEDBACK Message Type →

RFC-026: Human Feedback Signal

PRIVATE AND PROPRIETARY — NOT A PUBLIC RFC. Owned by Kanjani AI Research & Causum. See NOTICE.md.

Status: DRAFT

Authors: Kanjani AI Research & Causum

Date: 2026-06-17

1. Problem Statement

The AIGP protocol defines REQUEST (pre-check), RECORD (post-invocation telemetry), and TRACE (execution spans). These capture what the AI did — but not how the user perceived it.

User feedback is the only signal that measures:

  • Whether the AI response was helpful, accurate, or harmful
  • Whether governance constraints (moderation, guardrails) were too aggressive or too lenient
  • Whether model quality is degrading from the user’s perspective
  • Alignment between AI output and user intent

Without a protocol-level feedback mechanism:

  • Governance decisions cannot be calibrated against real user experience
  • Evidence chains lack the human-in-the-loop signal required by NIST AI RMF (MAP 3.2, MEASURE 2.6)
  • Reinforcement loops have no ground truth to train against

Section index · 2. Proposed Extension: FEEDBACK Message Type →