Glossary
Definitions of key platform terms - from acceptance regions and actions to world model and workspaces.
Key terms used throughout the Concurrence platform documentation. Definitions are kept short and practical. Terms are listed in alphabetical order.
Use established capitalization for named product components, such as Context Graph, World Model, and Agent Forge. Use lowercase for generic roles and processes, such as an agent, a classifier, or a projection; retain established labels such as System 1 and System 2. Keep the qualifier when a term has different meanings: source-class confidence, calibrated model probability, and identifier-match score are not interchangeable. Likewise, distinguish an agent definition from its running interaction and a recorded observation from a confirmed workflow outcome.
A
Acceptance Region: The set of outcomes that count as successful for a given use case. Defined across multiple dimensions (accuracy, safety, empathy, latency, cost). An outcome must satisfy all dimensions simultaneously to fall within the acceptance region.
Action (Classic API: tool): A versioned code package that an agent can execute during a conversation. Actions can connect the agent to external systems such as scheduling APIs, EHR lookups, or notification services. A generic tool or Platform integration is not necessarily a Classic Action.
Agent: A named configuration resource whose versions define identity, background, behavioral guidance, communication patterns, and optional voice configuration. A service combines a selected agent version with a Context Graph and runtime settings. The agent definition is distinct from the workflow, decision procedure, and running interaction. See Agents.
Agent Engine: The runtime layer that combines shared reasoning components with channel-specific execution, tool coordination, and delivery. Its signal-and-effect vocabulary is shared, but supported runtime paths do not all execute the same loop. See Reasoning Engine.
Agent Forge: The CLI tool for managing agent configurations. Used to create, update, version, and promote services and version sets across environments. See the Agent Forge reference page.
Application View: Information selected or derived from permitted evidence for a particular reader and task. Each view has its own derivation, freshness, and coverage requirements; it does not establish that every source assertion is correct. See World Model.
Auto-Enrichment: A best-effort background process that can gap-fill registered enrichment keys for supported person and place entities. It writes source-tagged enrichment events asynchronously; higher-authority values can supersede the result. This is distinct from event embedding.
B
Backfill: Loading historical records or recomputing historical derived data for a supported pipeline or metric. Coverage depends on the available inputs and the specific processing path. A backfill does not by itself prove improvement or guarantee deterministic replay of model-generated results.
Barge-In: Supported voice behavior that can stop agent playback after incoming caller speech meets the checks of the active voice path. On the platform-managed runtime the active state's turn policy must allow it and the speech needs recognized words, and speech still in progress also needs a minimum duration and a cooldown; on gpt_realtime it needs a minimum duration of detected speech, after the call opening. Sensitivity and timing depend on the active voice path and service configuration.
C
Channel Use Case: A provisioned channel identity and purpose, such as an email sender or managed messaging pool. A service binding establishes the workspace routing for supported inbound and outbound operations. Provisioning, consent, and sender approval remain separate prerequisites. See Channels.
Classifier: A procedure that assigns an input to one or more declared categories. It can be rule-based or learned; its output requires evaluation for the task and does not independently establish truth, authorization, or an appropriate action.
Clinical Tools: Built-in tools available to supported agent runtimes for patient and appointment workflows, typed clinical reads, coverage and refill-request writes, outbound-queue lookup, and operational logging. Supported writes emit world-model events; recording an event does not deliver it to an external system.
Conference-First Architecture: The Platform-managed voice-call model in which the caller and agent use a shared conference and an eligible operator can join without reconnecting the caller. Listen, takeover, and warm-forwarding paths have different behavior; conference participation does not guarantee a complete transcript or clinical briefing.
Confidence Score: A context-specific numeric measure, not a single platform-wide notion of certainty. World-model source-class confidence ranks observations for supported conflict resolution; a projection can group those values into precedence tiers. Calibrated model probabilities, identifier-match scores, model-reported certainty, and evaluation scores have separate meanings and validation needs. None automatically grants permission or triggers operator escalation.
Consumer Dispatch Policy: In the universal processing design, the rules selecting which programs may consume a processing packet. It is separate from packet formation and effect arbitration, and is not a new Platform API resource or an alias for a tool binding's dispatch settings.
Context Graph (Classic API: service_hierarchical_state_machine): The authored workflow definition of states, objectives, transitions, guidance, and tool bindings. It describes how work should proceed, not the world's facts. Model-facing guidance is distinct from server-enforced permissions and validation; a graph does not make model-driven navigation deterministic. See Context Graphs.
Cut Policy: In the universal processing design, the rules deciding when to form a processing packet and which evidence it contains. A cut does not itself require a model call, cancel ongoing work, or interrupt playback.
D
Data-MCP: A standalone Model Context Protocol service that exposes read-only SQL exploration tools using Classic API organization credentials. For agent-side access during conversations, use platform functions; external integrations using Platform API workspace credentials can use the Platform API MCP server. See Data Access (MCP).
Decision Procedure: A rule, query, program, classifier, or model-based process used to answer a bounded question or select a next step. This is an architectural category, not a separate Platform API resource. See Reasoning Architecture.
Drift: A gradual change in agent performance over time. Can be caused by changes in the input distribution (new types of conversations), shifts in agent behavior, or evolving requirements. See Drift Detection.
Dynamic Behavior: Guidance that a model chooses to apply on the caller turns where it seems to fit, with a condition, examples, an instruction, a mode, and sticky turns. Behaviors live in a workspace library of Dynamic Behavior sets with immutable numbered versions, and a service's version set pins an exact version of each. The Classic API has a separate dynamic_behavior_set resource with triggers and tool-candidate changes. See Dynamic Behaviors.
E
Effect: A typed output from supported reasoning components. Depending on its type and runtime, it can request future delivery or report work already performed; for example, a tool-called effect can report a completed tool invocation. Inspect the specific contract rather than treating every effect as a queued action. An emitted effect alone does not establish recipient delivery or a confirmed external mutation. See Reasoning Engine.
Effect Arbitration Policy: In the universal processing design, the rules deciding which intended outputs may proceed under authority, relevance, dependency, capacity, and ordering constraints. Priority cannot grant permission. This is a design responsibility, not a claim that every runtime implements one universal scheduler.
Email Channel: A managed channel for transactional and marketing email with verified sending domains, typed use cases, inbound reply threading, and separate delivery-status records. Email has its own send API and is not a target of the Conversations API's outbound-start or channel-switch operations. See Channels.
Entity Resolution: The process of associating an incoming record or caller with a workspace entity. Current paths use explicit source identifiers, caller bindings, and scoped lookups. same_as edges represent historical merges from the retired pre-SDP resolver; the current Platform pipeline does not produce new deterministic cross-source merge edges.
Entropy Stratification: The design discipline of allocating computation and action freedom according to task uncertainty, error consequence, and the expected value of more evidence or reasoning. It is not a documented runtime setting or a universal uncertainty score. See Reasoning Architecture.
Escalation: The process of handing a supported conversation from the agent to a human operator. Escalation follows authored agent and context-graph policy, supported operator workflows, or an explicit user request; a generic low-confidence score does not automatically trigger it.
External Principal: An entity-anchored, role-scoped session identity for external users (such as clinicians or partner employees) driving agent conversations. A principal binds an external user identity to a workspace entity with assigned roles, and role grants scope which integration endpoints, skills, and knowledge base scopes the session can access. See Compliance.
F
FHIR Store Connector: The fhir_store data-source type for a FHIR R4 store. The platform stores the data-source record but does not poll it; send FHIR records through FHIR import.
Filler Speech: Short spoken acknowledgements that can cover part of a voice-processing or tool-execution delay. Filler selection is channel-, state-, and empathy-aware, can be suppressed, and does not guarantee that every delay is covered.
Function Store: The workspace registries from which platform functions and workspace data queries are loaded at session start. A successfully loaded definition can become an agent tool when the selected runtime and any applicable Context Graph binding expose it; registration alone is not a universal visibility or authorization rule. See Platform Functions.
Functional Memory: Cross-session context built from source conversation records, model-extracted episodic observations, a bounded semantic user model with lineage, and separately projected structured clinical state. Current enrichment can load into later sessions, but it does not guarantee complete recall or eliminate authoritative data retrieval. See Functional Memory.
I
Information Entropy: A measure of uncertainty in a specified probability distribution over outcomes. It is distinct from source-class confidence, correctness, and error consequence. Low entropy can describe a confidently wrong model; it is not sufficient by itself to choose a safe action.
Interaction Insight: A Classic API response for a permitted conversation interaction, including the current state and action, transition logs, permitted working-memory or reflection records, and executed tool-call logs. The Platform API does not expose an equivalent hidden-reasoning contract; Platform simulation traces contain best-effort observed evidence instead.
K
Keyterm Boosting: A speech-to-text configuration that biases compatible recognition paths toward domain vocabulary such as medication, provider, or organization names. It can improve recognition for selected terms but does not guarantee an accurate transcript.
L
Liquid World Model: An architectural perspective on the World Model in which retained, source-linked evidence supports different, revisable task-specific views. SQL transformations, classifiers, and model-generated interpretations can contribute different derivations. The term does not imply autonomous schema evolution, instant freshness, or a learned predictive simulator. See Reasoning Architecture.
LLM-Evaluated Metric: A metric whose eligible evidence is scored by a model-based judge against authored criteria rather than by a deterministic rule. Availability, scheduling, and execution limits depend on the evaluation path, and results require calibration against human review. See Metrics.
M
Metric: A configured measure of interaction or operational evidence. Evaluation metrics can score eligible conversations against rubrics. Operational metrics aggregate supported workspace evidence through producer-specific paths; a definition does not imply that every channel or interaction produces a value. See Metrics and Quality and Metric Store.
Metric Store: The Platform API's common value model and query surface for operational, quality, and evaluation metrics. The current catalog has 41 built-in definitions and accepts up to 50 custom definitions per workspace. The schema contains five extraction modes, but supported custom modes have narrower rules. Latency tiers and freshness values describe intended processing and configured targets, not contractual delivery times. See Metric Store.
Modality Adapter: A channel-specific integration boundary between input, reasoning, and delivery. Signals and effects provide a shared vocabulary, while text, simulation, native voice, and provider-owned realtime voice retain distinct execution paths and delivery, lifecycle, authorization, and failure semantics. See Reasoning Engine.
O
Operator: An authorized human who can monitor eligible live runs and, for supported channels, listen, guide, take over, or hand control back. Run-scoped transcript and context can be available, but operator entry is not a guaranteed complete patient briefing or proof that a downstream handoff succeeded.
Outcome: An observed result relative to a defined task or acceptance criterion. Request acceptance, tool completion, recipient delivery, and a confirmed external change are different outcomes and require different evidence. A workflow outcome is not established solely by an agent's statement of success.
Output Track: In the universal processing design, a logical capacity and ordering boundary for outputs or ongoing work, such as exclusive foreground speech or bounded parallel tools. It is not a new API resource or a requirement to implement every resource as a priority queue.
P
Patient Arc: A reviewed description of the path from a patient's request to completion or handoff, including the facts, decisions, allowed actions, and exception paths it requires. Teams implement an arc through a Context Graph, tools, and the surrounding application, then test the expected behavior. See Improving Agent Behavior.
Persona (Agent): The conceptual identity and background carried by an agent version, including structured identity, background text, behavioral guidance, and communication patterns. The Platform API has no standalone persona resource; services resolve an agent version through version-set configuration. The Developer Console lists agents as Personas, and services as Agents. See Agents.
Persona (Simulation): A synthetic user profile used in simulations. Defines the characteristics, behaviors, and communication style of a test user. See Simulations.
Platform Function: A workspace-registered, read-only SQL, AI, Python, or table-valued computation. Supported runtimes can expose named functions (fn_*) and workspace data queries (wsq_*) after successful loading and any applicable Context Graph binding. Management is available through workspace-scoped APIs and supported tooling. See Platform Functions.
Pre-Emptive Tone Adjustment: A supported voice fallback that can select sympathetic delivery when navigation sets no tone for the reply, no workspace tone is configured, and the current Context Graph action matches a configured sensitive topic. This is delivery guidance, not a prediction that the caller is distressed or an independent safety control.
Problem Definition: The authored specification of the object of work, objectives, evidence requirements, permissible actions, and completion conditions. A Context Graph expresses part of that specification; server-side authorization and validation enforce controls that instructions alone cannot. It is not a single Platform API resource.
Processing Packet: A bounded input formed from selected signals and relevant context at a cut in the universal processing design. It may be one message or a task-specific grouping, not necessarily a transport packet. Earlier harness designs call bounded context a quantum. It is not a new Platform API resource.
Projection: A derived read view computed from relevant evidence and evaluation context. Reproducibility depends on the derivation's complete inputs, logic, configuration, and any time-relative evaluation. Model-generated memory or narrative views need separate evaluation rather than inheriting deterministic guarantees. See World Model.
Projection Function: Logic that folds world-model events into an entity-state read model. Deterministic derivations are reproducible only with their complete inputs and evaluation context fixed; time-relative values can change when unchanged events are recomputed later. Projection is asynchronous from event acceptance, so a successful write can precede visibility in the serving view.
Q
Quality Score: Context-dependent call evidence. The terminal operational quality score runs from 0-100 and reflects latency, barge-ins, loops, operator escalation, and tool failures. Voice Judge and configured evaluations use separate rubrics; a retained analysis field does not guarantee a result. See Call Intelligence.
R
Reasoning Engine: Shared components for interpreting signals, applying agent and Context Graph guidance, selecting eligible actions, and producing responses or effects. Text and simulation use effect-processing paths; native voice and provider-owned realtime voice have distinct execution paths. Shared concepts do not imply one identical loop across channels. See Reasoning Engine.
Review Queue: The operations that list stored external write proposals and record a reviewer's decision on one. A decision does not deliver a write or change the source event's confidence. See Review Queue.
S
Scenario (Simulation): A defined situation used in simulations. Describes the context, events, and goals for a test interaction. See Simulations.
Selected Fact: The value chosen by the applicable projection or resolution rules from available source assertions. Selection establishes the current read value, not independent verification that it is true or fresh. See World Model.
Service: Deployment configuration that connects selected agent and Context Graph versions with runtime and channel settings. Named version sets select the component versions used when configuration is resolved. See Core Concepts.
Service Binding: The association between a provisioned channel use case and a service in a workspace. It establishes routing, not provider provisioning or recipient consent. Rebinding can change the service in the same workspace; removing a binding removes that route without deleting the upstream channel resources.
Service Voice Configuration: Per-service voice tuning that controls filler behavior (style, vocabulary, timing), barge-in sensitivity, response length limits, end-of-turn detection, TTS settings, and call forwarding. Different services within the same workspace can have different voice characteristics. Managed through the Platform API and Agent Forge CLI.
Signal: A supported input occurrence relevant to agent execution, such as an utterance, tool result, silence, interruption, external event, or system event. A signal can update context without producing a response. Signal production and handling differ by runtime and channel. See Reasoning Engine.
Silence Monitor: The voice-runtime behavior that handles caller quiet, the same on both voice runtimes. Ten seconds of quiet after the agent speaks brings a check-in, the next ten seconds a second one, and the ten seconds after that end the call, unless the caller has described a medical emergency or self-harm, which leaves the line open. See Silence Management.
Simulation Bridge: An exploratory testing mode that generates scenario variations from a natural-language objective, runs each as a full multi-turn conversation with an LLM-driven persona, and collects interaction insights for coverage tracking. See Simulations.
Skill: An orchestrated companion definition configured through the Platform API. A skill declares a required system prompt, input and result schemas, model and execution bounds, and the integration or static tools it may call. The Platform API does not expose a five-tier skill model. See Skill Execution.
Source Assertion: A record of what a source reported, with its available provenance and timestamps. It is evidence of the report, not independent proof of the claim. See World Model.
Surface: A dynamic or template-based data collection interface that can be delivered by email or handed off as a web link. Eligible agent tools and configured state templates can create surfaces; creation, delivery, submission storage, event publication, and downstream mutation remain separate lifecycle steps. See Surfaces.
SurfaceSpec: The structured specification an agent generates to define a surface - title, fields, delivery channel, expiration, and entity association.
System 1: The execution role of fast, familiar, bounded decisions. Rules, classifiers, decision models, or simple model calls can fill this role. It is not an agent resource or a fixed model family. See Reasoning Architecture.
System 2: The execution role of deliberation over ambiguous or unfamiliar situations, often supported by an LLM. It is not a separate agent resource, and more deliberation does not guarantee correctness or grant additional authority. See Reasoning Architecture.
T
Test Run: The execution of a test set that produces scored results for each unit test. See Simulations.
Test Set: A group of related unit tests that are run together. Test sets are often organized by capability area or risk level.
Trajectory: An ordered sequence of observations, decisions, tool actions, and results across a task. The available record depends on the execution and observability path; the term does not promise a complete cross-application trace or establish a causal clinical conclusion.
Transcript Extraction: In the current Platform memory flow, asynchronous model-based processing can derive source-linked episodic observations and registered behavioral memory dimensions from a completed conversation. It does not automatically convert phone, date-of-birth, insurance, or address mentions into authoritative structured clinical state; explicit tools and connector projections have separate write semantics.
Trigger: A workspace-scoped automation that binds an action to a cron schedule or supported platform event. Triggers support event filters, timezone-aware scheduling, pause/resume lifecycle, manual firing for testing, and execution history tracking. Managed through the Platform API, Developer Console, and Agent Forge CLI.
Typed Decision Model: A model used to choose or score declared alternatives from context with a structured decision contract. Typed outputs constrain form, not correctness, candidate completeness, or authority. Treating scores as reliable probability estimates requires task-specific evaluation and calibration.
U
Unit Test: A combination of a persona, a scenario, and success criteria that tests a specific agent behavior.
Universal Processing Engine: The broader architectural design for forming processing packets from heterogeneous signals, dispatching eligible computations, and arbitrating intended effects before channel execution. Reasoning is one responsibility within it. It does not name a separate API resource or imply one fully implemented runtime across modalities. See Design Responsibilities.
Universal Reasoning: The architectural composition of authored problem definitions, scoped evidence, decision procedures, and supported runtime execution. "Universal" refers to reusable composition, not universal competence, a separate endpoint, or a shipped general-purpose controller. See Reasoning Architecture.
Universal Reasoning Harness: A proposed extension of execution coordination for inputs, reasoning, actions, and delivery with independent lifetimes. Its broader execution ownership, background behavior selection, and evidence-preload contracts are design direction, not universal current runtime capabilities. See Reasoning Engine.
V
Version Set: A named service configuration that can pin agent and context graph versions, dynamic behavior versions, model preferences, and a supported turn runtime. Platform services require a release version set; omitted version numbers resolve to the latest available version. Classic version sets have a different component model.
W
Workspace: The Platform tenant boundary for data, configuration, and access. A workspace can contain several services and has its own scoped entities, sources, and permissions. See Workspaces.
Workspace Data Query: A parameterized SQL template registered per workspace that runs directly against workspace-owned custom tables. Exposed to the agent as a wsq_<name> tool with typed parameters and write-time validation. See Platform Functions.
World Model: The workspace-scoped data layer that retains source-attributed events and entity relationships and asynchronously derives supported views for agents and applications. It represents available evidence, not a workflow definition or inherently a learned predictive simulator. Agents receive selected context and can use authorized retrieval; no session payload is guaranteed to contain every source or current fact. Health World Model names the healthcare application of this same layer, not a separate product or API resource. See World Model.
Write Scope: A server-constructed boundary that limits the workspace, entity, resource types, creation rights, and confidence available to supported model-originated write tools. Trusted services use separate service authorization; their access does not let an agent tool bypass its own scope.
Last updated
Was this helpful?

