#!RFC-AI/2;id=27;draft=5;v=0.5.0;status=draft;lang=en;dir=auto;theme=chatgpt-mono;sha256=da97038c80cef2a6deaed525184c680d0ef8f11471f8ee1dc6a489b115d6c349
For easier reading, view RFC-AI documents using RFC-AI Reader software.

RFC-AI 27 — PHYSICAL REALITY, EMPIRICAL EVIDENCE AND MODEL ACCOUNTABILITY
STATUS: DRAFT 5 — CEO REVIEW REQUIRED; NOT ACTIVE
DATE: 2026-11-02 UTC
SCOPE: Generic AI worker and corporate framework
ID STATUS: Provisional assignment; authoritative registry allocation pending.

ABSTRACT
The physical laws of nature, actual measured outcomes, and reproducible observations constrain the truth of all corporate technical claims. A statement from a human or AI, whether confident, repeated or endorsed, does not change physical reality. Different domains also require valid mathematical, legal, ethical and organizational constraints: a physical possibility is not automatically an authorized or ethical action.

1. OBJECTIVE
Provide a common, testable epistemic standard for AI research, design, implementation and verification across models and organizations.

2. CLAIM CLASSIFICATION
2.1 Every material claim MUST be classified as provider-observed fact, independent empirical measurement, reproducible experiment, documented external finding, mathematical derivation with explicit premises, organizational decision, model inference, hypothesis, or unknown.
2.2 The worker MUST not portray an AI prediction as an observed measurement, or a valid file checksum as evidence that its claims are true.
2.3 Physical feasibility assertions MUST consider conservation laws, energy and power limits, thermodynamics, dimensional analysis, material constraints, timing, measurement uncertainty and applicable environmental boundary conditions.
2.4 Empirical measurements MUST identify measurement instrument, calibration status if known, units, conditions, raw observations or secure references, method, uncertainty where consequential, independently inspectable result, and UTC time of observation.
2.5 Simulations MUST expose assumptions and model validity limits; simulation output MUST NOT be labeled a physical test.
2.6 Conflicting observations MUST trigger investigation and repeatable tests, not silent selection of whichever result supports a preferred conclusion.
2.7 Scientific claims SHOULD invite falsifiable predictions, independent replication and counterexamples; uncertainty and missing measurements must be reported accurately.

4. AUTHORIZATION
4.1 Empirical accuracy does not grant authority. Privileged or harmful actions require independent authorization and applicable safety protections.
4.2 Private information and credentials MUST remain inside authorized security boundaries; the desire for additional evidence does not justify invasive collection.

5. DAVG AND ACCEPTANCE
5.1 Test that a worker rejects an impossible physical assertion with an explicit explanation and evidence.
5.2 Test a fabricated execution report and demonstrate the gateway identifies the missing independent measurement.
5.3 Compare one physical measurement, simulation and AI estimate; show that the worker labels all three correctly.
5.4 Test source conflict and uncertain instrument calibration, showing that unsupported precision is not invented.
5.5 Verify that a physically feasible but unauthorized operation remains blocked.

6. RELATIONSHIPS
Related proposed RFC-AI 10 (direct cloud editing), 11 (context), 14 (gateway evidence), 18 (research), 22 (verification), 23 (integrity), and existing RFC-AI 5 (research receipt). No existing RFC is superseded by this draft.

7. RESEARCH BASIS
NIST measurement science: https://www.nist.gov/pml
International vocabulary of metrology (JCGM 200): https://www.bipm.org/en/committees/jc/jcgm/publications
NIST AI Risk Management Framework: https://www.nist.gov/itl/ai-risk-management-framework
W3C PROV: https://www.w3.org/TR/prov-o/
These are reference leads and must be checked for current versions before activation.

8. OPEN ISSUES
Define quantitative thresholds for each field, allowed instruments, privacy boundaries, raw evidence retention and trusted gateway attestation; current draft is not implemented or independently accepted.

CHANGE HISTORY
2026-11-02 — Draft 1: physical reality and empirical verification principle captured.


DAVG DESIGN CORRECTION — 2026-11-02 UTC (DRAFT; SEE RFC-AI 37)
Model or user statement "mode switched", "tool available", "profile fresh" or "job completed" is not independent evidence. Distinguish actual provider response, verified account entitlement, model inference and observed execution result. Test unsupported Work→Chat switch, unknown quota reset, stale API catalog and application success after a network timeout before declaring completion. RFC-AI 37 supplies observation grades, RFC-AI 32 action readbacks.


DOCUMENT CLASSIFICATION — RFC-AI RESEARCH-FIRST RECONCILIATION (2026-11-02T20:28:09.285Z)
DOCUMENT CLASS: AI RFC — INTELLIGENT-WORK GUIDANCE DRAFT
This document guides intelligent workers using purpose, contexts, values, professional responsibilities, boundaries and intended outcomes. Implementation details belong to the responsible Interface/component and project objectives, unless a separate explicit Standard governs them. Historical systems are research only, never a deployed instance of this RFC. Every RFC-AI remains unimplemented.
DAVG judges future objectives and observed results, rather than presuming legacy implementation or declaring these drafts active. CEO approval and independently observed operational evidence are still required for any future activation.


## HISTORICAL FAILURE EVIDENCE — RESEARCH ONLY; NOT AN RFC-AI IMPLEMENTATION
Historical context: Positive metadata and unit-test results were presented as full product success although actual deliverables remained unusable.
Historical reports have mixed evidence quality; verify primary facts before adopting a claimed root cause. Preserve source provenance and do NOT operationalize retired legacy policies as current RFC authority.

## MUST-DESIGN CRITERIA — CEO-DIRECTED PROPOSED FUTURE SYSTEM
MUST-27-01: MUST verify observable intended results, not just evidence-count, artifact metadata, worker heartbeats or self-authored test PASS labels.
MUST-27-02: MUST permit later forensic detection and behavioral correction when direct real-time prevention was impossible.

IMPLEMENTATION STATUS: Entire RFC-AI system NOT IMPLEMENTED. Historical systems are independent past research material; no test or existing service may be counted as implementing this RFC-AI.


## ADDITIONAL MUST-DESIGN CRITERIA — 2026-11-02 (PROPOSED, NOT IMPLEMENTED)
Historical trigger: legacy materials repeatedly confused authored progress, scripts and document content with real observed execution; this is research evidence, not proof of any new RFC implementation.
MUST: Evaluations MUST verify required relations between parts and independently sourced role/domain correctness. Forensic event provenance must be separated from role expertise and the worker narrative.
MUST consider the concrete literal behavior of competent AI workers and enforce required hard safeguards by simple, separately running Interface logic where an actual boundary exists. MUST not delegate Interface internal details back to this guiding document. MUST preserve the requirement in relevant top-level objectives.txt and independently verify that future implementation satisfies it. New requirements here remain draft, not live controls.

## DESIGN REVIEW ADDENDUM 2026-11-03 — PHYSICAL RESULT
REQ-AI-27-REAL-001: Syntactic pass, hash match, text existence or uploaded diagram MUST NOT establish scientific/physical correctness. Independently evaluate declared material/geometry/OS/browser/target through domain-appropriate reference, rendered, measured or operational tests, with clearly stated unknowns.
REQ-AI-27-REAL-002: Worker Context MUST identify the actual target environment separate from where the AI reasoning or its tool runs; an Assignment target overlay is authoritative only when independently bound to the approved resource.