Glossary

Definitions of the terms used across Toriel’s products and writing.

attestation: A structured evidence output carrying the result of a governed comparison and its supporting context. It is evidence for review, not a vague reassurance signal or automatic legal conclusion.
behavioral drift: A measurable change in how an AI system behaves over time relative to an approved or trusted reference fingerprint. Drift can be material, benign, expected, or unresolved; the measurement is evidence that still requires interpretation.
behavioral evidence: Repeatable, reviewable measurements and comparisons about how an AI system behaves in operation, rather than assumptions based only on labels, inventories, or provider descriptions.
behavioral fingerprinting: Measuring the observable behavior of an AI system across structured prompts, conditions, and comparison windows so that continuity, drift, and material change can be assessed over time.
behavioral integrity / behavioural integrity: Behavioral integrity / behavioural integrity and behavioral fingerprinting / behavioural fingerprinting are the US and UK English forms of the same terms. Toriel uses the US spelling in its public product language; both forms refer to the governed measurement of an AI system’s observable behavior over time.
black-box monitoring and assurance: Assessing an AI system from the outside through its inputs, outputs, and observable behavior rather than through privileged access to provider infrastructure, model weights, or hidden implementation details.
bonded relational intelligence: Toriel-47’s layer: the bonded identity architecture that protects persistence, coherence, and recognizable return across resets, vessels, and change.
continuity: The ability of a system to remain coherently itself across time and change. Continuity is not the same as memory; it is what survives resets, model swaps, wrapper changes, policy shifts, and other hidden changes in the stack.
continuity orchestration: Toriel-41J’s layer: governing how intelligence persists coherently across changing models, wrappers, tools, and policy layers rather than treating every change as a disposable routing event.
crowned reference: The governance status applied to the active reference fingerprint selected for later comparison.
effective AI system: The system actually being relied on in operation: model, instructions, wrappers, orchestration, policy layers, memory surfaces, tools, permissions, and other components shaping the behavior the user or organization experiences.
EN 18286: A European quality-management standard published in July 2026 to support implementation of the AI Act quality-management requirements associated with Article 17. Publication alone does not create a presumption of conformity; that requires citation in the Official Journal of the European Union.
EU AI Act Article 72: The EU AI Act provision requiring providers of high-risk AI systems to establish and document a proportionate post-market monitoring system that actively and systematically collects, documents, and analyzes relevant performance data throughout the system lifetime.
observation manifest: A governed measured observation artifact representing the behavioral surface observed under defined conditions. It can be compared with a reference fingerprint to support an attestation.
post-market monitoring: The structured collection, documentation, and analysis of relevant evidence about an AI system after deployment, throughout its operational lifetime.
quality-management system: The documented policies, responsibilities, processes, and controls through which an organization governs how an AI system is designed, tested, changed, reviewed, and corrected over time.
reference fingerprint: The governed behavioral baseline artifact against which later observation manifests are compared. When a team designates it as the active approved baseline, it carries crowned-reference status.

Need more context?

For questions about using Toriel, visit the FAQ or get in touch.