What does the machine claim mean?
Semantic identity and preservation make meaning explicit before another system relies on it.
Independent research · formal semantics · verification systems
Research for consequential machine action.
RISU studies what must remain true as machine-readable meaning crosses evidence, interfaces, and action boundaries. The work turns those distinctions into bounded definitions, executable checks, and inspectable public records.
Scope consequential machine actionMethod formal + executableRecords public technical notes + DOI archivesOrganization independent research initiative
Research architecture
Three bounded questions, one consequence layer.
What justifies reliance?
Evidence qualification and transport separate an observation from the stronger premise a downstream action may consume.
What is fixed at the action cut?
Closure and operativity ask whether the distinctions that matter to the consequence have actually been established and still govern the effect.
Cross-cutting assuranceProjection Assurance tests whether consequence-relevant semantics survive translation into agent-facing interfaces. RISU Verify carries that result into reproducible software and browser-consumable records.Understand Projection Assurance →
Selected research
A research program built around failure at boundaries.
Each project names a narrow failure mode, states a bounded claim, and keeps the formal, empirical, and operational layers separate.
Projection assurance
Projection Assurance
Did the consequence-defining behavior survive when a source operation became an agent-facing tool?
Correspondence, discrimination, operativity, exact realization, and coverage are evaluated separately so surface similarity cannot stand in for consequence preservation.
Consequence assurance
Consequence Closure
At the moment a machine acts, is the specified consequence already determined?
Relational counterexamples identify what remains material. Minimal semantic obligations remain distinct from the mechanisms that can establish them.
Evidence-qualified reliance
Reliance Before Closure
When can one machine rely on a claim before the broader workflow has finished?
Provider-specific qualification, a verified stability certificate, and a separately bounded proposed action feed a provider-blind relying function without pretending the broader workflow has already closed.
Flagship software
RISU Verify
A local-first assurance surface for consequence preservation across agent-facing projections.
Consequence assurance · research software · August 2026
Check consequences, not appearances.
An interface can keep the same name, inputs, and shape while losing a safeguard that changes what the action actually does.
RISU Verify evaluates the declared consequence, carries the producer-side result through exact handoff artifacts, and lets the browser independently verify what it can from the bytes it receives.
CLI / CI → .risu record → browser consumer
R
Local browser consumerRISU VerifyAssurance Workbench
PRESERVEDBROWSER CONSISTENCY PASS
Safeguard preserved
The projected mechanism reproduces the declared source consequence in the admitted worlds.
CorrespondenceC1
DiscriminationD1
OperativityO1
Handoff integrityBrowser verified
Certificate bindingBrowser verified
Frozen core identityBrowser matched
Public research record
Citable work, linked to the evidence it rests on.
RISU Technical Notes state the claim boundary, archival record, and supporting software or evidence where applicable.
2026
Projection Assurance: Consequence-Relevant Semantics Across Agent-Facing Interfaces
RISU Technical Note 2026-04 · Moon Lee
Consequence Closure: Semantic Assurance at the Machine Action Boundary
RISU Technical Note 2026-03 · Moon Lee
Reliance Before Closure: Evidence-Qualified Stability for Machine Decisions Under Unresolved Effects
RISU Technical Note 2026-02 · Moon Lee
From Revocation to Closure: Verifying Attributable Consequences in AI Agent Decommissioning
RISU Technical Note 2026-01 · Moon Lee