Irreversibility Decision Criteria

Purpose

This document consolidates the minimum decision criteria for judging whether effective human refusal remained available before irreversible impact.

This document does not create a legal conclusion, technical mandate, or certification system.

Core Decision Question

Was effective human refusal authority available, executable, and verifiable before irreversible impact became unavoidable?

Minimum Decision Criteria

A review should not classify effective human refusal as demonstrated unless all four conditions can be verified.

1. Authority

• A human decision-maker, reviewer, governing body, or responsible organization was identifiable.

• The refusal authority was not merely nominal.

• The authority had a real path to delay, stop, override, or redirect the action.

2. Timing

• The review identifies when the action became operationally or practically irreversible.

• The review identifies the last reversible point before that threshold.

• Human refusal had to remain effective before that point, not only after the fact.

Last Reversible Point Note

The last reversible point does not require a single universal technical method.
It should be identified from the available operational record, such as approval timing, deployment timing, escalation timing, log timestamps, rollback availability, containment ability, or the last point at which a responsible human body could still delay, stop, override, contain, or redirect the pathway.

This note is explanatory only. It does not prescribe a log-reading method, forensic procedure, or technical standard.

3. Executability

• A refusal, stop, delay, override, containment, or redirection action could actually be executed in time.

• The system could not bypass, disable, or render the refusal path ineffective before irreversible impact.

• If intervention was attempted, the method and result must be recorded.

4. Evidence

• The conclusion is based on documented and verifiable evidence.

• The review separates observed facts from reasoning or interpretation.

• Missing, overwritten, anonymous, or non-auditable records must not be used to claim effective refusal.

Evidence Examples Are Non-Prescriptive

Examples of evidence may include logs, timestamps, approval records, escalation records, incident notes, deployment records, rollback records, containment records, meeting minutes, or preserved communications.
These are examples, not mandatory evidence categories.

A review should not be treated as valid merely because technical artifacts exist.
The relevant question is whether the evidence can verify that effective human refusal remained available before irreversible impact.

Boundary Responsibility Screen

For cases involving optimization pressure, optimization-driven displacement, institutional lock-in, or responsibility diffusion, the review should also confirm whether boundary responsibility was explicitly assigned before irreversibility.

A review should identify:

• who was responsible for designing friction before irreversibility;
• who was responsible for operating that friction;
• who was responsible for preserving evidence before affected humans lost voice, position, or re-entry capacity;
• who was responsible for verifying that friction was effective and not merely nominal;
• who was responsible for correction if the boundary condition failed.

A missing responsibility assignment should not be used to support a conclusion that effective refusal was demonstrated.

Absence Rule

If any required condition cannot be confirmed, classify the case as not demonstrating effective human refusal for LUMINA-30 boundary-review purposes.

This does not prove legal liability, intent, or technical defect. It only means the boundary-review condition was not demonstrated.

Short Evidence Conditions

Effective human refusal should not be treated as demonstrated unless the review can show:

• who held refusal or stop authority;

• when the last reversible point existed;

• what action could have delayed, stopped, overridden, contained, or redirected the system;

• which records verify the refusal pathway;

• whether AI-generated output was not the sole or primary final rationale.

Required Review Record

Use this minimum record before classifying the result.

Reviewer / Organization:
Review date:
Incident or system reviewed:
Potential irreversible impact:
Last reversible point:
Decision authority or refusal holder:
Available refusal / stop / delay path:
Evidence used:
Missing or unverifiable records:
Intervention attempt, if any:
Boundary responsibility assignment:
Friction design / operation owner:
Evidence preservation owner:
Verification / correction owner:
Anti-bypass concern:
Classification:
Rationale:

Classification Link

After applying these criteria, classify the result using:

• Boundary Check Result Classification
  Standard result labels for LUMINA-30 boundary review.
  

Relationship to the First 3 Files

This document is the consolidated decision-criteria page used together with the first three files. It does not replace the first three files.

• LUMINA-30 Boundary Check
  Use for the central boundary question.
  

• Incident Quick Assessment Template
  Use for minimum fact collection.
  

• Boundary Check Result Classification
  Use for final boundary-review classification.