Evidence before reputation.

Research / Selected work

Examples and research candidates are kept in their proper epistemic state. An illustrated mechanism is not the same thing as a publicly reproducible result.

Research method Evidence & Scope

SW01 · Synthetic explanation / current technical preview

Guardian · a decision before consequence

A proposed customer-data export lacks established destination authority and exceeds stated scope. The illustrated controller requests repair instead of treating the model proposal as permission.

Conceptual synthetic Guardian decision sequence from proposal through evidence and repair — Synthetic Guardian decision sequence. Quantitative claims bounded to the AgentDojo-compatible budget validation package cited on Evidence & Scope (Guardian Runtime · e3-guardian-harness-2026).

Required for public empirical classification: exact Guardian version, fixture, policy/evidence inputs, resulting trace, provenance, checksum, host enforcement limits and reviewed disclosures.

Evidence & Scope · Explore Guardian · Guardian technical preview

SW02 · Proposed mathematical protocol

Computation and representational invariance

Question: does a chosen measure remain stable under an explicitly defined class of transformations while discriminating meaningful differences between systems?

Specify

Define state space, mapping class, measure, domains and allowable assumptions before running a comparison.

Compare

Use trivial and adversarial re-encodings, rival metrics, constructed counterexamples and resource measurements.

Falsify

Report violating transformations or indistinguishable states, including negative cases and model dependence.

This is a proposed protocol. It does not report a theorem, experimental result or universally valid model of reality.

Conceptual comparison of computational substrates across formal and physical realizations — Computational substrates across formal and physical realizations.

SW03 · Consciousness-relevant research · status under evidence review

Observable self-monitoring is not subjective experience.

A system's ability to represent and correct its own errors can be studied functionally. The move from functional findings to phenomenological claims requires independent justified theory and discriminating evidence. Consciousness is a scientific question at LUXION, not a product claim.

Illustrative competing research hypotheses for a system producing a self-error report — Competing research interpretations. MoC7 venue and author status require verified clearance before any stronger public claim.

Consciousness-relevant research programme · Research method

Provenance before promotion

Concept diagram connecting a claim, artifact, experimental conditions, and publication approval — Evidence provenance schema. Publication uses fail-closed publishedClaim() filtering.