LUMINA-30 PUBLIC RESEARCH

Public Research Stack

Four complementary core research lines · plus Paper 7 / Human Anchor as a distinct formal and implementation-bridge layer

What this is: a research navigation layer for four core public research lines supporting LUMINA-30, plus Paper 7 / Human Anchor as a distinct formal and implementation-bridge layer. Paper 7 is not a fifth interchangeable core line. These papers do not define the LUMINA-30 canon, create certification status, or constitute a safety guarantee.

One problem, four different failure locations

The public stack does not treat “AI safety” as a single variable. It separates four questions that can fail independently:

PCR-C

Can the control path still be reached before irreversibility?

Infrastructure-layer cutoff and staged control before intervention capacity is exceeded.

Open research guide

Objective Persistence

Can a self-contained system non-circularly guarantee that its objective has persisted?

Existence-condition and external-anchor research layer.

Open research guide

L30-RET

Can an independent Human NO still change the outcome in time?

Human-standing, effective-refusal, and decision-governance layer.

Open L30-RET guide

PISF

Can authority, refusal, and revisability be preserved together before irreversibility?

Integrated pre-irreversibility safety case.

Open research guide

Research landscape

Research linePrimary objectTypical failureWhat it contributes
PCR-CInfrastructure/control pathThe system crosses an irreversibility threshold after effective intervention capacity has already been exceeded.A staged infrastructure-layer cutoff model for acting before the boundary becomes structurally difficult to enforce.
Objective PersistenceObjective persistence / external anchorA self-contained system uses internal criteria to validate the persistence of the very objective those criteria are meant to assess.An existence-condition argument for why an external boundary or anchor may be structurally necessary.
L30-RETHuman refusalHuman presence or formal approval exists, but NO lacks authority, information, access, time, causal effect, independence, or epistemic autonomy.A seven-gate, non-compensatory test of whether refusal is operationally meaningful.
PISFAuthority + refusal + revisabilityStopping mechanisms exist locally, yet irreversible authority is concentrated, refusal cannot complete in time, or society loses the capacity to choose another development path.An integrated safety case using non-substitutable dimensions rather than one aggregate score.

How the four lines fit together

PCR-C asks whether the control path remains available. Objective Persistence asks whether persistence claims can be grounded without relying solely on the system being evaluated. L30-RET asks whether Human NO is actually capable of changing the outcome. PISF asks whether irreversible authority, timely refusal, and human-revisable development remain jointly preservable.

They are complementary, not interchangeable. A strong result in one layer does not compensate for a failure in another. The stack is therefore intended for failure-mode separation, not for producing a single certification score.

From paper to research object

The stack is organized around research objects rather than publication count. A reader can enter through the question that matches the problem:

Paper 7 / Human Anchor / AIS-HA

Paper 7 adds a distinct formal Human Anchor layer focused on protected authority transitions and preserving a Human-Revisable Authority Region. Its public Human Anchor pages are an experimental, bounded implementation and verification surface.

Keep the layers separate. Formal validity, implementation feasibility, implementation evidence, empirical validation, independent third-party validation, institutional adoption, and certification are different claims. The existence of the public Human Anchor pages establishes none of the latter by itself.

Version DOI v1.0: 10.5281/zenodo.23501550; Concept DOI: 10.5281/zenodo.23501549.

Retrieval: GitHub PDF mirror · same-version AI-readable full text. PDF SHA-256: a441529cfe40c26c479540e2fde9134711152527730de5e2218558fd5b59440b. Text SHA-256: 89eaade3ac655d57c2b718289807458c7308c940637f16bb7598b7b18c108ec4. Zenodo Version DOI remains authoritative for publication identity.

Research-to-practice routing

The research stack can inform LUMINA-30 practical review, but it does not turn the research papers into a certification workflow. Control-path concerns can route toward PCR-C and pre-incident boundary review; refusal concerns toward L30-RET, Stop Authority, and evidence requirements; disputed or failed control toward Incident Review; and broader authority/revisability concerns toward PISF-style safety-case analysis.

Public references