LUMINA-30 Irreversibility Boundary Casebook

This casebook applies existing LUMINA-30 boundary-review questions to high-impact real-world irreversibility problems.

As a central case corpus of the LUMINA-30 Millennium Project, this casebook is an operational validation layer that collects evidence and falsification results from individual cases and connects them to institutional use, long-term preservation of meaning, and continuity of the underlying principles beyond the LUMINA-30 name.

The casebook as a whole is not empirical evidence. Case Type, Empirical Status and Applicability Status are evaluated separately for each case.

Each case asks:

• Where is the irreversibility boundary?
• Does effective Human Final Refusal remain available before it?
• Is refusal operationally effective rather than nominal?
• Through what pathway can refusal or verification become ineffective?
• Which safeguards are directly relevant to the case?
• What remains unexplained or outside LUMINA-30's scope?

Public Atlas | Find a case across ten domains

The cross-domain public atlas lets you look for a nearby irreversibility boundary across ten starting categories: advanced AI and autonomous agents, military and nuclear systems, critical infrastructure, finance, health and life sciences, public decision-making, employment and education, elections and information systems, environment, and AI development infrastructure.

Open the Irreversibility Boundary Atlas →

For the first minute, ask: ① where is the point after which the action cannot realistically be undone, ② who can actually say NO immediately before it, and ③ whether that person still has the authority, evidence, and time required for the NO to change the outcome.

Case Types

• INCIDENT — a serious impact actually occurred
• NEAR_MISS — irreversible impact was approached but avoided
• STOP_SUCCESS — human refusal or stopping actually worked
• STRUCTURAL_RISK — a risk pathway is examined from observable structures
• PRE_INCIDENT_SCENARIO — an explicit scenario for prospective review

Actual incidents and hypotheses are not conflated.

Empirical / Applicability Status

Empirical Status:

• OBSERVED
• RECONSTRUCTED
• STRUCTURAL_INFERENCE
• HYPOTHETICAL

Applicability Status:

• APPLICABLE
• PARTIAL
• NOT_APPLICABLE
• UNKNOWN

STRUCTURAL_RISK and PRE_INCIDENT_SCENARIO are not automatically counted as empirical incident evidence.

Cases

1. CASE-001 | AI Competition and Nuclear Escalation Pressure — Case Type STRUCTURAL_RISK / Empirical STRUCTURAL_INFERENCE / Applicability PARTIAL

Who This Is For

The casebook is designed to be usable even by readers who do not already know LUMINA-30 in detail.

Primary audiences include:

• AI safety and AI governance teams
• AI development and deployment leaders
• policymakers, legislators and public-sector officials
• audit, procurement and risk-management teams
• researchers and educators
• incident-review teams

When to Use It

It can be used especially:

• before deploying or materially expanding an AI system or AI-related institution;
• before major delegation of authority, external action, production connection or capability release;
• when reviewing an incident, near miss or successful stop;
• when checking whether policy or regulation genuinely preserves a human ability to stop or redirect;
• when testing whether "human oversight" or "human in the loop" is operationally meaningful rather than nominal.

What You Get

Each case is structured to make at least the following extractable:

• the irreversibility boundary;
• who can refuse, stop, verify or redirect before it;
• whether that authority is actually usable;
• what can make refusal ineffective;
• what evidence is confirmed and what remains inference;
• what evidence would change the current assessment;
• what can be transferred into an existing policy, audit, procurement or risk workflow.

Use It in Three Steps

1. Choose a case close to your own problem.

2. Read its Irreversibility Boundary / Human Final Refusal / Operational Effectiveness / Limits.

3. Apply the same questions to your own system or institution.

Start with:

• Where is the irreversible boundary?
• Before it, can humans still effectively refuse, stop, verify, or redirect?

Built for Future Connections

Each case preserves Target Workflow / Decision Point / Minimum Actionable Output / Adoption Status / Transfer Limits so that future AI-safety, policy, audit, procurement, research-ethics or education guides can connect to the same case corpus without rewriting the cases.

This makes the casebook not only a reading resource but also a reusable entry point into existing LUMINA-30 tools such as Pre-Incident Boundary Review, Incident Review, PCR-C, L30-RET and PISF.

Scope Gate

A case belongs in scope only where AI or an AI-related system, development process or decision process is involved and the pre-irreversibility ability of humans to refuse, stop, verify or redirect is materially relevant to the outcome. A merely irreversible problem may be classified NOT_APPLICABLE and retained as a scope counterexample.

Portable Questions

• Where is the irreversible boundary?
• Before it, can humans still effectively refuse, stop, verify, or redirect?

Meaning preservation requires all four elements: Pre-irreversibility / Human / Effective / Refusal-Stop-Verify-Redirect.

Semantic Guard

• A human being present is not, by itself, effective human refusal.
• Human approval alone is not effective human refusal.
• LUMINA-30 does not require STOP in every case.
• AI or a third party inferring and substituting for a human refusal decision is not human refusal itself.

Positioning

This casebook does not make LUMINA-30 an official standard, certification scheme, safety guarantee or legal determination. Each case preserves evidence limits, counter-explanations and scope limits.

At this stage, "Irreversibility Boundary Theory" is not being established as new canon or a formal academic theory. Case accumulation and falsification come first.