Glossary

41 terms

AI-ready webDiscovery

A public-site preparation model that keeps pages human-first while exposing reliable discovery, structured content, capability boundaries, provenance, and review evidence for AI systems.

Answer Engine OptimizationDiscovery

Answer-first publishing that gives a direct, citable, source-backed answer near a stable heading and routes readers back to the canonical public page.

Generative Engine OptimizationDiscovery

Source preparation that helps generative systems retrieve, ground, summarize, cite, and route back to public evidence without hidden bot-only claims.

Canonical URLSEO

The preferred stable URL for a page or route, used to consolidate discovery, citations, metadata, and machine-readable catalogs.

JSON-LDStructured data

A JSON-based linked-data format used to publish structured page facts such as article metadata, breadcrumbs, organizations, and defined terms when those facts match visible content.

Readiness profileDiscovery

A maturity label describing which AI-ready capabilities are supported with evidence, which are only package-ready, and which are not claimed.

Route inventoryDiscovery

A machine-readable list of canonical public routes with titles, status, reviewed dates, purposes, related pages, and deployment notes.

Structured dataSEO

Machine-readable metadata that clarifies public page entities, relationships, dates, and breadcrumbs while remaining consistent with visible page content.

Agent runtimeArchitecture

The execution environment that hosts an agent loop, preserves task state, mediates tools, and handles lifecycle events.

Approval gateGovernance

A runtime state that pauses a consequential action until an authorized human or service approves, rejects, or modifies it.

CheckpointRecovery

A durable snapshot of relevant runtime state created before a risky step or at a defined lifecycle boundary.

CompensationRecovery

A corrective action that semantically reverses or offsets a completed side effect when direct rollback is impossible.

Control planeArchitecture

The management layer for identity, policy, provisioning, routing, configuration, lifecycle, and observability.

Evidence eventEvidence

A typed, timestamped record of a request, context decision, tool action, approval, artifact, error, or unresolved uncertainty.

Evidence ledgerEvidence

An ordered collection of evidence events designed for review, incident analysis, and user-facing accountability.

Human-in-the-loopGovernance

A control pattern in which a person reviews or authorizes selected decisions rather than supervising every model token.

Inference runtimeModel serving

Software that loads model weights, schedules compute, manages memory, decodes outputs, and returns model results.

Local-firstDeployment

An architecture in which the user-visible control plane, policy, and critical data placement remain locally governed even when some compute is remote.

Machine Intelligence RuntimeDefinition

An emerging execution-time control layer that coordinates agents, tools, memory, policy, approvals, recovery, telemetry, and evidence around model inference.

Memory scopeMemory

A named boundary defining who or what can read, write, retain, correct, export, or delete a class of remembered information.

Model adapterArchitecture

A boundary that normalizes provider-specific requests, streaming, budgets, tool schemas, and error behavior.

Policy decisionGovernance

A typed runtime result—allow, deny, require approval, transform, or escalate—derived from identity, context, action, and risk.

Prompt injectionSecurity

An attack or failure mode in which instructions in user or retrieved content manipulate model behavior contrary to intended policy.

Recovery ratePerformance

The share of failed or degraded runs that return to a valid state and complete under defined constraints.

Runtime intelligenceResearch

The allocation and control of inference-time computation, search, verification, refinement, memory, and interventions while a task executes.

Runtime planeArchitecture

The isolated execution environment for an active agent session, including workspace, tools, model adapters, memory, and artifacts.

Side effectTools

A change outside the model response, such as writing a file, sending a message, changing an account, or deploying software.

Tool contractTools

A typed definition of tool inputs, outputs, permissions, side effects, idempotency, failure modes, and evidence requirements.

Trust boundarySecurity

A point where data, authority, identity, or execution crosses between components with different assumptions or privileges.

Active inferenceResearch

A computational framework in which action and belief update are derived from a generative model; it is distinct from teleodynamics and should not be treated as a synonym.

AutopoiesisResearch

A biological and systems-theory concept describing a network that produces and maintains its organization; software use is an analogy unless explicitly formalized.

Control objectiveGovernance

A bounded runtime condition such as budget, evidence completeness, or checkpoint freshness that supports an external objective without replacing it.

Goal hierarchyGovernance

A typed relationship among an external objective, success criteria, derived subgoals, runtime control objectives, immutable policies, and terminal conditions.

HomeostasisResearch

The regulation of variables within viable ranges; in runtime engineering those ranges are explicitly designed operating thresholds.

No-opRecovery

An explicit decision to make no structural or operational change because evidence, benefit, authority, or resources do not justify one.

Reflective criticResearch

A bounded evaluator that diagnoses runtime evidence and proposes recovery or verification without authorizing its own high-risk actions.

Resource gatingArchitecture

A control that permits an action or structural edit only when expected benefit, cost, budget, policy, and maintenance requirements are satisfied.

Structural adaptationResearch

A change to components, routes, rules, plans, or topology rather than a parameter update inside a fixed structure.

TeleodynamicsResearch

A theory of end-directed organization arising from mutually constraining processes; its runtime use remains an editorial research translation unless formally implemented.

Teleodynamic runtime profileResearch

An optional research-only extension for bounded goals, viability metrics, reflection policy, structural-edit proposals, and immutable constraints.

Viability metricOperations

A measured operating-condition value such as budget remaining, context pressure, tool reliability, evidence completeness, or checkpoint freshness.

No glossary terms match that search.