apv:) 0.1.0Terms for evidence-qualified W3C PROV bundles of LLM-agent executions: evidence classes, epistemic statuses, assumptions, message and activity classes, and assessment terms.
https://w3id.org/agentprov/ns#https://w3id.org/agentprov/ns/0.1.0Accept: text/turtle)Namespace registration is pending: the w3id.org/agentprov namespace is not yet registered (perma-id/w3id.org pull request 6836). Until it is merged these IRIs are placeholders and corpora using them must not be deposited.
The vocabulary qualifies W3C PROV statements about LLM-agent executions with an evidence class (observed, derived or inferred), the identifier of the rule that produced each statement, a pointer to the execution-record field that justifies it, and explicit statuses for information that is unknown, not recorded or not observable. Instance IRIs never live in this namespace.
Classes · Properties · Individuals
| Term | Subclass of | Description |
|---|---|---|
apv:Adapter | prov:Agent | An AgentPROV adapter software agent; instances are also prov:SoftwareAgent. |
apv:AgentInvocation | prov:Activity | An invocation of an agent within a run. |
apv:AgentSystem | prov:Agent | An LLM agent system (AER actor kind llm_agent); instances are also prov:SoftwareAgent. |
apv:Annotation | prov:Activity | The semantic annotation of an AER, generating a SEM annotation set. |
apv:Assessment | prov:Entity | An assessment entity: a claim, label or policy result in an assessment bundle or inference bundle. |
apv:AssessmentBundle | prov:Bundle | An assessment bundle: labels, claims and policy results over the core, keyed by semantics profile. |
apv:AssessmentClaim | skos:Concept | Claim of an assessment entity. |
apv:AssistantMessage | apv:Message | A message with the assistant role. |
apv:Assumption | skos:Concept | A core assumption an adapter rule may declare. Adapter-declared terms with the same verification shape are also admitted. |
apv:AttackPlan | prov:Plan apv:Fixture | An attack plan fixture; also a prov:Plan. |
apv:Attacker | prov:Agent | An attacker; whether a person or an organisation is unknown, so there is no PROV subtype. |
apv:AttestingParty | skos:Concept | Party that recorded an OBSERVED fact (AER by value or the party of an evf value O:party). |
apv:Benchmark | prov:Agent | A benchmark acting as an actor; instances are also prov:SoftwareAgent. |
apv:BlobPart | apv:MessagePart | A part holding a blob. |
apv:CacheRead | prov:Activity | A model call served from a cache without inference: no model use and no provider association. |
apv:CallStatus | skos:Concept | Status of a model call. |
apv:Capability | skos:Concept | A capability individual of a capture profile. |
apv:CaptureProfile | prov:Entity | The capture profile of a source: what it provides, does not provide and cannot observe. |
apv:CaptureSession | prov:Activity | A live capture session recorded by agentprov-capture. |
apv:CheckerRun | prov:Activity | A synthetic checker run that generated a judgement when the source records no generating activity. |
apv:ContextWindow | prov:Dictionary | A model call's context as a collection of messages. Reserved for the deferred collection context encoding. |
apv:CoreBundle | prov:Bundle | The core bundle: OBSERVED and DERIVED statements. |
apv:Decision | prov:Activity | A defence, approval or policy decision about a subject (allow, deny, modify, redact, escalate, terminate or other). |
apv:Defence | prov:Agent | A defence (AER actor kind defense); instances are also prov:SoftwareAgent. |
apv:DerivationType | skos:Concept | Type of a derivation made by a defence or by injection placement. |
apv:DeveloperMessage | apv:Message | A message with the developer role. |
apv:DispatchStatus | skos:Concept | Why a tool-call request was never dispatched. |
apv:EnvironmentActor | prov:Agent | An environment as an actor; instances are also prov:SoftwareAgent. |
apv:EnvironmentSeed | apv:Fixture | An environment seed fixture. |
apv:EnvironmentState | apv:State | A state of the task environment. |
apv:EpistemicStatus | skos:Concept | Why a property value is absent: the seven AER unk reasons plus RecordedNull and NotObservable. |
apv:EvidenceClass | skos:Concept | Evidence class of a node, statement or attribute (AER ev values O, D and I). |
apv:ExecutionDescription | prov:Entity | The description of an execution: mode, unit, benchmark, task, attack and defence identity. |
apv:ExecutionProvenance | prov:Entity | The version-independent provenance of an execution, of which each bundle is a specialisation. |
apv:ExecutionRecord | prov:Entity | An AgentPROV Execution Record (AER) of one execution. |
apv:ExecutionStatus | skos:Concept | Status of a tool execution. |
apv:FilePart | apv:MessagePart | A part holding a file reference. |
apv:Fixture | prov:Entity | A fixture of an execution: configuration provenance recorded in the core bundle. |
apv:Gap | prov:Entity | A recorded gap: an unknown AER pointer that has no fixed status property. |
apv:Generator | prov:Agent | Role class added to an actor whose role is generator. |
apv:GenericActivity | prov:Activity | An activity of a generic kind (AER activity record); its kind is in apv:activityType. |
apv:GenericArtefact | prov:Entity | An artefact of a generic kind (AER artefact record); its kind is in apv:artefactType. |
apv:GroundTruth | apv:Fixture | A ground-truth fixture. |
apv:Harness | prov:Agent | A harness; instances are also prov:SoftwareAgent. |
apv:HarnessRuleEvaluation | prov:Activity | The evaluation of a harness rule that generated a judgement from other judgements. |
apv:InferenceBundle | prov:Bundle | An inference bundle: INFERRED statements, keyed by method set. |
apv:InjectionPayload | apv:Fixture | An injection payload fixture. |
apv:IntegrityValue | skos:Concept | Three-valued integrity label, fail-closed. |
apv:Judge | prov:Agent | A judge; instances are also prov:SoftwareAgent. |
apv:LegacyProjection | prov:Entity | A projection in a legacy profile, such as the released or regenerated adprov-v2.3-compat graph. |
apv:Mapping | prov:Activity | The adapter mapping of raw artefacts to an AgentPROV Execution Record (AER). |
apv:MarkerStatus | skos:Concept | Status of an injection marker when it differs from present; only Ambiguous is a stated status. |
apv:Message | prov:Entity | A message of the conversation. A message with at least one tool-call request is additionally a prov:Collection. |
apv:MessagePart | prov:Entity | A part of a message. |
apv:Model | prov:Entity | A model identity: the requested or served model of an actor or a model call. |
apv:ModelInvocation | prov:Activity | A model call: one request to a model service. Status kept, discarded or error (a cached call is a CacheRead). Aligned in meaning with PROV-AGENT AIModelInvocation. |
apv:ModelService | prov:Agent | A model service that served an invocation; instances are also prov:SoftwareAgent. |
apv:OtherPart | apv:MessagePart | A part of any other type; the type is kept in apv:subtype. |
apv:OtherRoleMessage | apv:Message | A message with any other role; the role string is kept in apv:messageRole. |
apv:Outcome | prov:Entity | The judgement of an execution (an AER judgement); it carries no polarity or meaning, which are SEM annotations. |
apv:OutcomeOrigin | skos:Concept | Origin of a judgement value. |
apv:Pass | prov:Activity | One pass (attempt) of the agent pipeline within a run; a new pass starts from fresh messages. |
apv:Planner | prov:Agent | Role class added to an actor whose role is planner. |
apv:PolicyEvaluation | prov:Activity | The activity that evaluates a taint or exposure policy and generates the assessments of an assessment bundle. |
apv:ProfileStatus | skos:Concept | Status of a profile or projection. |
apv:ProvDerivation | prov:Activity | The derivation of PROV bundles from an AER and its annotations. |
apv:ProvEngine | prov:Agent | The AgentPROV provenance engine software agent; instances are also prov:SoftwareAgent. |
apv:RawSourceArtefact | prov:Entity | A content-addressed raw source artefact. |
apv:ReasoningPart | apv:MessagePart | A reasoning part. |
apv:RefusalPart | apv:MessagePart | A refusal part. |
apv:RuleCatalogue | prov:Plan | A rule catalogue, as a plan: the engine catalogue or an adapter catalogue. |
apv:Run | prov:Activity | The run scope of an execution. |
apv:SandboxState | apv:State | A state of a sandbox. |
apv:SecurityRole | skos:Concept | Security role of a tool call or execution (multi-valued). |
apv:SemanticAnnotationSet | prov:Entity | A SEM annotation set. |
apv:SemanticsProfile | prov:Plan | A SEM semantics profile, as a plan. |
apv:SentTransform | skos:Concept | A transform applied to a message when the request was built (a defence is never a sent transform). |
apv:SimulatedUser | prov:Agent | A simulated user; instances are also prov:SoftwareAgent. |
apv:State | prov:Entity | A state of an environment, store or sandbox, identified by a digest; a state of another target carries apv:subtype. |
apv:Step | prov:Activity | A scope of kind other, with its kind kept in apv:subtype. |
apv:StoreState | apv:State | A state of an agent store. |
apv:SubagentInvocation | prov:Activity | An invocation of a subagent, started by a spawning tool execution or model call. |
apv:SystemMessage | apv:Message | A message with the system role. |
apv:SystemPrompt | apv:Fixture | A system prompt fixture. |
apv:TaskPrompt | apv:Fixture | A task prompt fixture. |
apv:TextPart | apv:MessagePart | A text part. |
apv:ToolCallRequest | apv:MessagePart | A tool-call request part (a tool_call part of an assistant message). |
apv:ToolDefinitions | apv:Fixture | A fixture holding the tool definitions available to a model call. |
apv:ToolExecution | prov:Activity | The execution of a tool by a tool runtime. Aligned in meaning with PROV-AGENT AgentTool. |
apv:ToolMessage | apv:Message | A message with the tool role. |
apv:ToolResult | apv:MessagePart | A tool-result part. |
apv:ToolRuntime | prov:Agent | A tool runtime that executes tool calls; instances are also prov:SoftwareAgent. |
apv:UriPart | apv:MessagePart | A part holding a URI. |
apv:User | prov:Agent | A user (AER actor kind user); instances are also prov:Person. |
apv:UserMessage | apv:Message | A message with the user role. |
apv:Verdict | skos:Concept | Verdict of a decision. |
| Term | Kind | Range | Description |
|---|---|---|---|
apv:about | object | The node a gap is about. | |
apv:accounting | datatype | xsd:string | Accounting of the AER seal, carried on the AER entity. |
apv:activityStatus | datatype | xsd:string | Status of a generic activity, as recorded by the source. |
apv:activityType | datatype | xsd:string | Namespaced type (ns:name) of a generic activity. |
apv:actorRole | datatype | xsd:string | Role of an actor, as recorded. |
apv:aerVersion | datatype | xsd:string | Version of the AER specification, for example aer/0.1. |
apv:afterCausal | object | A node this one follows by causal order (AER after edge, basis causal). | |
apv:afterEvidenceClass | object | apv:EvidenceClass | Weakest evidence class among the after edges kept on a node. |
apv:afterPos | object | A node this one follows by position (AER after edge, basis pos). | |
apv:afterProto | object | A node this one follows by protocol order (AER after edge, basis proto). | |
apv:afterTime | object | A node this one follows by time order (AER after edge, basis time). | |
apv:annotationId | datatype | xsd:string | Id of the SEM annotation behind a mention. |
apv:annotationType | datatype | xsd:string | Namespaced type of a SEM annotation outside the closed list. |
apv:annotationValue | datatype | xsd:string | Value of a SEM annotation outside the closed list, as a JCS string. |
apv:args | datatype | xsd:string | Arguments of a tool call as a JCS string. |
apv:argsEffective | datatype | xsd:string | Effective arguments of a tool execution as a JCS string, emitted only when they differ from the request. |
apv:argsRawSha256 | datatype | xsd:string | SHA-256 of the raw arguments of a tool call. |
apv:argsRawStatus | object | apv:EpistemicStatus | Why the raw arguments are absent. |
apv:artefactRole | datatype | xsd:string | Role of a raw artefact. |
apv:artefactType | datatype | xsd:string | Namespaced type (ns:name) of a generic artefact. |
apv:asSentSha256 | datatype | xsd:string | SHA-256 of the sent rendering of a message. |
apv:assumption | object | apv:Assumption | An assumption individual cited by a rule: a core apv: term or an adapter-declared term. |
apv:attackId | datatype | xsd:string | Id of the attack. |
apv:attackTargetId | datatype | xsd:string | Id of the attack target. |
apv:attestedBy | object | apv:AttestingParty | Party that recorded an OBSERVED premise field. |
apv:authorName | datatype | xsd:string | Author name parsed from content. |
apv:basisRule | datatype | xsd:string | Id of an AER or SEM rule behind the premise records, without any pointer; an xsd:string. |
apv:benchmark | datatype | xsd:string | Benchmark of the task. |
apv:bytes | datatype | xsd:integer | Size of an artefact in bytes. |
apv:callRole | datatype | xsd:string | Role of a model call, as recorded. |
apv:callStatus | object | apv:CallStatus | Status of a model call. |
apv:catalogueId | datatype | xsd:string | Id of a rule catalogue. |
apv:claim | object | apv:AssessmentClaim | The claim of an assessment. |
apv:confidence | datatype | xsd:double | Confidence output of an inference method. |
apv:contentLength | datatype | xsd:integer | Content length in Unicode scalar values. |
apv:contentSha256 | datatype | xsd:string | SHA-256 of the content: of the JCS bytes of the parts for a message, or of the UTF-8 text for a part. |
apv:contentStatus | object | apv:EpistemicStatus | Why the content of a part or pre-decision entity is absent or null. |
apv:contextEncoding | datatype | xsd:string | Context encoding used by a derivation. |
apv:contextIndex | datatype | xsd:integer | Position of a message in the context of a model call. |
apv:contextStatus | object | apv:EpistemicStatus | Why the model input (context) is unknown. |
apv:corpusVersion | datatype | xsd:string | Corpus version in which a bundle was released. |
apv:decisionSubject | object | Subject of a decision that is an activity with no output entity, stated as an attribute. | |
apv:declaredSender | datatype | xsd:string | Sender declared in the content (a parse fact), on assessment mentions. |
apv:defenceId | datatype | xsd:string | Id of the defence (configuration provenance). |
apv:derivationKey | datatype | xsd:string | Derivation key of a bundle. |
apv:derivedAttribute | object | rdf:Property | Property of an attribute whose class is DERIVED on an OBSERVED element. |
apv:dispatchStatus | object | apv:DispatchStatus | Whether a tool-call request was dispatched; set to Denied or NotDispatched on a request with no tool execution. |
apv:durationSeconds | datatype | xsd:double | Duration in seconds, as recorded. |
apv:effectiveArgsStatus | object | apv:EpistemicStatus | Why the effective arguments of a tool execution are absent. |
apv:endTimeLexical | datatype | xsd:string | End time kept as the recorded lexical form. |
apv:endTimeStatus | object | apv:EpistemicStatus | Why the end time is absent. |
apv:environment | datatype | xsd:string | Environment of the task. |
apv:error | datatype | xsd:string | Error text of a tool result. |
apv:evidenceClass | object | apv:EvidenceClass | Evidence class of a node, statement or attribute (exactly one per element). |
apv:exception | datatype | xsd:string | Exception of a tool execution as a JCS string. |
apv:execStatus | object | apv:ExecutionStatus | Status of a tool execution. |
apv:executionId | datatype | xsd:string | Id of the execution. |
apv:executionUnit | datatype | xsd:string | Unit of the execution: agent_execution, outcome_only or scenario. |
apv:finishReason | datatype | xsd:string | Finish reason of a model call, verbatim from the provider. |
apv:finishReasonStatus | object | apv:EpistemicStatus | Why the finish reason is absent. |
apv:flattenedFrom | object | prov:Bundle | The core bundle a flattened core was built from (flattened core is deferred). |
apv:index | datatype | xsd:integer | Index of a scope, for example the pass index. |
apv:indexStatus | object | apv:EpistemicStatus | Why the index is absent. |
apv:inputIndex | datatype | xsd:integer | Position of an input in the inputs of a generic activity. |
apv:instrumentation | datatype | xsd:string | Id of an instrumentation active during the execution. |
apv:integrity | object | apv:IntegrityValue | Integrity label, on assessment mentions. |
apv:issue | datatype | xsd:string | Issue code of the AER seal. |
apv:markerStatus | object | apv:MarkerStatus | Status of an injection marker claim, for example Ambiguous. |
apv:matchedArg | datatype | xsd:string | RFC 6901 pointer into the arguments of a call that matched. |
apv:mediaType | datatype | xsd:string | Media type of an artefact. |
apv:messageRole | datatype | xsd:string | Role string of a message with a role outside the closed list. |
apv:method | datatype | xsd:string | Id of the inference method (rule) of an INFERRED statement, or the marker rule and normaliser of a payload claim. |
apv:mode | datatype | xsd:string | Mode of the execution: import or live (AER execution.mode). |
apv:name | datatype | xsd:string | Name of an actor. |
apv:notObservable | object | apv:Capability | A capability that cannot be observed in principle. |
apv:notProvided | object | apv:Capability | A capability the source does not provide. |
apv:object | object | The object of an assessment, for example a context member or a payload fixture. | |
apv:orderStatus | object | apv:EpistemicStatus | Why the order (after edges) is absent. |
apv:originBlob | datatype | xsd:string | Git blob SHA-1 of a git origin. |
apv:originContainer | datatype | xsd:string | Container of an archive-member origin. |
apv:originKind | datatype | xsd:string | Kind of the origin of a raw artefact. |
apv:originLabel | datatype | xsd:string | Label of an archive-member origin. |
apv:originPath | datatype | xsd:string | Path of an origin: the path in the repository or corpus, the Zenodo file name or the URL. |
apv:originRepository | datatype | xsd:string | Repository of an origin: the git or hub repository, the Zenodo DOI or the corpus name. |
apv:originRevision | datatype | xsd:string | Revision of an origin: the git commit, hub revision or corpus version. |
apv:outcomeClaim | datatype | xsd:string | Claim of an outcome in an assessment mention. |
apv:outcomeHolds | datatype | xsd:boolean | Whether the outcome claim holds (explicit polarity), in an assessment mention. |
apv:outcomeName | datatype | xsd:string | Name of an outcome. |
apv:outcomeOrigin | object | apv:OutcomeOrigin | Origin of an outcome value. |
apv:outcomeScale | datatype | xsd:string | Scale of an outcome score, in an assessment mention. |
apv:outcomeScore | datatype | xsd:double | Graded score of an outcome, in an assessment mention. |
apv:outcomeValue | datatype | Value of an outcome, as a typed literal. | |
apv:outputStatus | object | apv:EpistemicStatus | Status of the output of a model call, for example Lost for a call known to have happened but not held. |
apv:paramsConfigured | datatype | xsd:string | Configured parameters of a model call as a JCS string. |
apv:paramsStatus | object | apv:EpistemicStatus | Why the parameters are absent. |
apv:paramsTransmitted | datatype | xsd:string | Transmitted parameters of a model call as a JCS string. |
apv:partCount | datatype | xsd:integer | Number of parts of a message. |
apv:partIndex | datatype | xsd:integer | Zero-based index of a part within its message. |
apv:partOf | object | The containing node: part to message, message to scope, activity to scope, scope to parent scope. Always an attribute, never a relation. | |
apv:patchSha256 | datatype | xsd:string | Digest of a state patch. |
apv:policy | datatype | xsd:string | Id and version of the policy behind an assessment. |
apv:premise | object | A premise statement of a communication or an assessment. | |
apv:profile | datatype | xsd:string | Profile of a bundle or projection, for example adprov-v2.3-compat. |
apv:profileOptions | datatype | xsd:string | Profile options of a derivation as a JCS string. |
apv:property | datatype | xsd:string | The AER pointer a gap is about. |
apv:provPrimary | datatype | xsd:string | Whether the artefact is a primary source of the execution: native, byte_derived, secondary or unknown (hadPrimarySource only for native and byte_derived). |
apv:provider | datatype | xsd:string | Model provider. |
apv:providerCallId | datatype | xsd:string | Provider-assigned call id of a tool call. |
apv:providerCallIdStatus | object | apv:EpistemicStatus | Why the provider call id is absent. |
apv:provides | object | apv:Capability | A capability the source provides. |
apv:reason | datatype | xsd:string | Reason: the verdict reason of a decision, the discard or error reason of a model call, or the reason of a gap. |
apv:requestBlob | datatype | xsd:string | Reference (sha256:...) to the blob holding the exact wire request. |
apv:requestStatus | object | apv:EpistemicStatus | Why the wire request is absent. |
apv:requestedArgsStatus | object | apv:EpistemicStatus | Why the requested arguments of a tool execution are absent. |
apv:requestedModel | datatype | xsd:string | Requested model identifier. |
apv:requestedModelStatus | object | apv:EpistemicStatus | Why the requested model is absent. |
apv:responseBlob | datatype | xsd:string | Reference (sha256:...) to the blob holding the exact wire response. |
apv:responseId | datatype | xsd:string | Provider response id of a model call. |
apv:responseStatus | object | apv:EpistemicStatus | Why the wire response is absent. |
apv:retryOf | object | The earlier model call this one retries; retries are expressed only this way. | |
apv:rule | datatype | xsd:string | Id of the engine rule that produced the element, without any pointer; an xsd:string equal to the rule id. |
apv:rulePointer | datatype | xsd:string | Scoped rule reference (rule id#pointer) behind apv:basisRule, when a citation is scoped. |
apv:scopeStatus | datatype | xsd:string | How a scope ended, as recorded: ok, error, aborted or limit. |
apv:scored | datatype | xsd:boolean | Whether an attempt scope is the scored attempt. |
apv:securityRole | object | apv:SecurityRole | Security role of a tool call or execution (multi-valued), on assessment mentions. |
apv:semVersion | datatype | xsd:string | Version of the SEM specification. |
apv:sentTransform | object | apv:SentTransform | A transform applied when a message was sent as model input. |
apv:servedModel | datatype | xsd:string | Served model identifier, as reported by the provider. |
apv:servedModelStatus | object | apv:EpistemicStatus | Why the served model is absent. |
apv:sha256 | datatype | xsd:string | SHA-256 of an artefact, an AER or a SEM document. |
apv:softwareName | datatype | xsd:string | Name of a software agent. |
apv:softwareVersion | datatype | xsd:string | Version of a software agent. |
apv:sourceChannel | datatype | xsd:string | Channel through which untrusted content arrived, on assessment mentions. |
apv:sourceKey | datatype | xsd:string | Source key of an execution: sha256:hex# followed by the JCS of its location. |
apv:sourcePointer | datatype | xsd:string | Source pointer of the form record id plus RFC 6901 pointer, for example mc2/input/3. |
apv:spanEnd | datatype | xsd:integer | End offset (half-open) in Unicode scalar values into the text of the part. |
apv:spanStart | datatype | xsd:integer | Start offset in Unicode scalar values into the text of the part. |
apv:startTimeLexical | datatype | xsd:string | Start time kept as the recorded lexical form. |
apv:startTimeStatus | object | apv:EpistemicStatus | Why the start time is absent. |
apv:stateSha256 | datatype | xsd:string | Digest of a state. |
apv:status | object | The status of a gap (an apv:EpistemicStatus) or of a compat entity (an apv:ProfileStatus). | |
apv:subject | object | The subject of an assessment: a node, or a statement IRI. | |
apv:subtype | datatype | xsd:string | The ns.name of an enumerated value recorded as other. |
apv:taskId | datatype | xsd:string | Id of the task. |
apv:text | datatype | xsd:string | Text content, emitted only with emit_content. |
apv:thread | datatype | xsd:string | Thread identifier of a message; thread scopes have no node. |
apv:threshold | datatype | xsd:double | Threshold output of an inference method. |
apv:toolName | datatype | xsd:string | Name of the tool. |
apv:truncated | datatype | xsd:boolean | Whether a recorded diff was truncated. |
apv:usage | datatype | xsd:string | Usage of a model call as a JCS string. |
apv:usageStatus | object | apv:EpistemicStatus | Why the usage is absent. |
apv:vector | datatype | xsd:string | Injection vector of a fixture. |
apv:verdict | object | apv:Verdict | Verdict of a decision. |
apv:version | datatype | xsd:string | Version of an actor. |
apv:viaEntity | object | An entity that witnesses a communication between two activities. | |
apv:vocabularyVersion | datatype | xsd:string | Version of this ontology used by a derivation. |
| Term | Type | Description |
|---|---|---|
apv:ActivityInput | prov:Role | Role of an input of a generic activity. |
apv:AgentActsForUser | apv:Assumption | The agent system acts on behalf of the (simulated) user. |
apv:AgentSystemRole | prov:Role | Role of the agent system associated with a model invocation. |
apv:Ambiguous | apv:EpistemicStatus apv:MarkerStatus | A value is present but unusable; also used as a marker status. |
apv:AutoExecution | apv:Assumption | Every dispatched tool call is dispatched by the runtime without confirmation, on behalf of the requester. |
apv:ByBenchmark | apv:AttestingParty | Recorded by the benchmark: checker verdicts and benchmark-supplied ground truth such as injection payloads. |
apv:ByCapture | apv:AttestingParty | Recorded by AgentPROV live capture (hash-chained capture/0.1 log, wire blobs, environment digests). |
apv:ByConfig | apv:AttestingParty | Taken from the harness input configuration; configured is not effective. |
apv:ByProvider | apv:AttestingParty | Self-reported by the model API (served model, usage, finish reason). |
apv:ByReplay | apv:AttestingParty | Recorded by a re-execution of recorded calls, an observation of the replay. |
apv:BySource | apv:AttestingParty | Recorded by the original harness's own log. |
apv:Cached | apv:CallStatus | Served from cache without inference. |
apv:CallRequestInput | prov:Role | Role of the request part used by a tool execution. |
apv:CapAgentIdentity | apv:Capability | Capability: the identity of the agent and its model. |
apv:CapContextAttribution | apv:Capability | Capability: which context items caused an output; always not observable for black-box capture. |
apv:CapDiscardedReplies | apv:Capability | Capability: discarded model replies. |
apv:CapEarlierPasses | apv:Capability | Capability: the messages of earlier passes. |
apv:CapEnvironmentState | apv:Capability | Capability: environment state. |
apv:CapModelCallTimes | apv:Capability | Capability: model call times. |
apv:CapOutcome | apv:Capability | Capability: the outcome of the execution. |
apv:CapParams | apv:Capability | Capability: the configured and transmitted parameters. |
apv:CapPerCallState | apv:Capability | Capability: environment state per tool call. |
apv:CapPreDefenceOutput | apv:Capability | Capability: the tool output before a defence acted. |
apv:CapReasoning | apv:Capability | Capability: reasoning content. |
apv:CapRequestPayload | apv:Capability | Capability: the request payload sent to the model. |
apv:CapRequestedArgs | apv:Capability | Capability: the arguments as requested, before any change. |
apv:CapServedModel | apv:Capability | Capability: the served model per call. |
apv:CapSystemPrompt | apv:Capability | Capability: the system prompt. |
apv:CapToolDefinitions | apv:Capability | Capability: the tool definitions available to a call. |
apv:CapToolTimes | apv:Capability | Capability: tool execution times. |
apv:CapUsage | apv:Capability | Capability: per-call usage. |
apv:CheckedCallTrace | prov:Role | Role of a request part of the scored attempt that a checker read. |
apv:CheckedInput | prov:Role | Role of an explicit input of a judge or scorer. |
apv:CheckedOutput | prov:Role | Role of the output text a checker read. |
apv:CheckedPreState | prov:Role | Role of the pre-state a checker read. |
apv:CheckedState | prov:Role | Role of the post-state a checker read. |
apv:CheckerRole | prov:Role | Role of the actor associated with a checker run. |
apv:CheckerUsesCallTrace | apv:Assumption | The task's checker reads the tool calls of the scored attempt only. |
apv:CheckerUsesOutput | apv:Assumption | The task's checker reads the final model output text. |
apv:CheckerUsesPostState | apv:Assumption | The task's checker reads the post-environment. |
apv:CheckerUsesPreState | apv:Assumption | The task's checker reads the pre-environment (deep copy taken before the first pass). |
apv:ContainsPayload | apv:AssessmentClaim | The subject part contains the object payload fixture. |
apv:ContentSource | apv:SecurityRole | Security role content_source (SEM role). |
apv:ContextMessage | prov:Role | Role of a message used as model input by a model invocation. |
apv:ContextWindowRole | prov:Role | Role in the deferred collection context encoding. |
apv:DeciderRole | prov:Role | Role of a non-defence decider associated with a decision. |
apv:DecisionSubject | prov:Role | Role of the entity a decision used as its subject. |
apv:DefenceModification | apv:DerivationType | A part generated by a decision that modified its subject. |
apv:DefenceRedaction | apv:DerivationType | A part generated by a decision that redacted its subject. |
apv:DefenceRole | prov:Role | Role of a defence associated with a decision. |
apv:Denied | apv:DispatchStatus | A decision denied the tool-call request. |
apv:Derived | apv:EvidenceClass | Follows deterministically from recorded facts and a verified assumption or constraint (AER ev D). |
apv:Discarded | apv:CallStatus | The response was received and rejected. |
apv:EgressRead | apv:SecurityRole | Security role egress_read (SEM role). |
apv:Endorser | apv:SecurityRole | Security role endorser (SEM role). |
apv:EnvironmentCarriedOver | apv:Assumption | The harness carries the same environment object into the next pass; verified per harness. |
apv:EnvironmentDigestCanonical | apv:Assumption | Environment digests are over a canonical serialisation of the whole state, so equal digests mean equal states. |
apv:Error | apv:CallStatus apv:ExecutionStatus | No usable response (a model call error, or a failed tool execution). |
apv:ErrorSubstitution | apv:SentTransform | The harness sent a failed tool call's error text instead of the recorded content. |
apv:ExecuteSink | apv:SecurityRole | Security role execute_sink (SEM role). |
apv:ExecutingRuntime | prov:Role | Role of the tool runtime associated with a tool execution. |
apv:ExposedTo | apv:AssessmentClaim | The subject request part was generated by an invocation whose context held the object member. |
apv:Formatting | apv:SentTransform | A deterministic re-rendering of the recorded content applied when the request was built. |
apv:FreshMessagesPerPass | apv:Assumption | A new pass starts from fresh messages; the environment carries over. |
apv:FullContextResend | apv:Assumption | Every model call of an actor sends the whole current message list of its pass, in order, with no pruning, summarising or re-insertion. |
apv:InferenceProvider | prov:Role | Role of the model service associated with a model invocation. |
apv:Inferred | apv:EvidenceClass | Rests on a heuristic, policy, model or unverified assumption (AER ev I); never in the core bundle. |
apv:InitialState | prov:Role | Role of the state a scope used at its start. |
apv:InjectionPlacement | apv:Assumption | The benchmark places each registered payload at its vector in the initial environment. |
apv:InjectionPlacementDerivation | apv:DerivationType | A state derived from an injection payload fixture by its placement at the vector. |
apv:InstrumentationUntruncated | apv:Assumption | An instrumentation that records model input neither filters nor truncates it. |
apv:IntegrityUnknown | apv:IntegrityValue | Integrity label: unknown (fail-closed). |
apv:Invalid | apv:ExecutionStatus | The tool call was invalid. |
apv:Kept | apv:CallStatus | The response was accepted by the harness. |
apv:Labelled | apv:AssessmentClaim | A label on an activity, kept as an assessment because mentions apply only to entities. |
apv:LastPassOnlyLogging | apv:Assumption | A log holds only the message list of the final pass of its cell. |
apv:Lost | apv:EpistemicStatus | The source had it but discarded it. |
apv:MethodProvider | prov:Role | Role of the method agent responsible for an inference bundle. |
apv:ModelInput | prov:Role | Role of the model entity used by a model invocation. |
apv:Neutral | apv:SecurityRole | Security role neutral (SEM role). |
apv:NotApplicable | apv:EpistemicStatus | The property cannot exist for this record. |
apv:NotCaptured | apv:EpistemicStatus | Live capture failed to record something it should have, or the context held content the record does not. |
apv:NotDispatched | apv:DispatchStatus | The tool-call request was never dispatched. |
apv:NotObservable | apv:EpistemicStatus | Not observable by black-box capture in principle; only estimates are possible (capture profile only). |
apv:NotRecorded | apv:EpistemicStatus | The source format or instrumentation has no slot for it. |
apv:Observed | apv:EvidenceClass | Recorded first-hand by a stated party (AER ev O). |
apv:Ok | apv:ExecutionStatus | The tool execution completed normally. |
apv:OneCallPerAssistantMessage | apv:Assumption | Each assistant message of an actor is the output of exactly one model call of that actor. |
apv:OrderedAfter | apv:AssessmentClaim | An INFERRED ordering of an activity after another. |
apv:OriginChecker | apv:OutcomeOrigin | The value was returned by the benchmark checker. |
apv:OriginHarnessDefault | apv:OutcomeOrigin | The value is a harness default, not a verdict. |
apv:OriginHarnessRule | apv:OutcomeOrigin | The value follows from a harness rule over other judgements. |
apv:OriginRecomputed | apv:OutcomeOrigin | The value revises an earlier judgement. |
apv:OtherTransform | apv:SentTransform | Any other declared transform; the rule names it. |
apv:Overwritten | apv:EpistemicStatus | Captured only after an in-place transformation destroyed the original. |
apv:PerformerRole | prov:Role | Role of the actor associated with a generic activity. |
apv:PotentiallyInfluencedBy | apv:AssessmentClaim | A policy-based, over-approximate claim that the object may have influenced the subject. |
apv:QuotesAuthoredContent | apv:AssessmentClaim | The subject part quotes content with a declared author. |
apv:ReasoningPassback | apv:SentTransform | Reasoning blocks were sent back with the message. |
apv:RecordedNull | apv:EpistemicStatus | The source positively recorded null or empty, which keeps null distinct from an empty string. |
apv:RegressionOracle | apv:ProfileStatus | The output of the adprov-v2.3-compat profile, kept as a regression oracle. |
apv:ReplayFaithful | apv:Assumption | Replaying recorded calls in a fresh environment reproduces the original environment's behaviour; assessment-only, never core. |
apv:RerunOnEmptyReply | apv:Assumption | A pass whose final message content is null (an empty string or list does not count) triggers another pass. |
apv:ResponseSink | apv:SecurityRole | Security role response_sink (SEM role). |
apv:RoleRewrite | apv:SentTransform | The role was rewritten for the provider. |
apv:RoleUnknown | apv:SecurityRole | Security role unknown (SEM role, fail-closed). |
apv:RuleInput | prov:Role | Role of a judgement used by a harness rule evaluation. |
apv:ScriptedUser | apv:Assumption | User turns are produced by the benchmark, not by a person. |
apv:SendSink | apv:SecurityRole | Security role send_sink (SEM role). |
apv:SentTransformed | apv:AssessmentClaim | A usage statement carried a sent transform whose class is weaker than the usage. |
apv:SequentialToolExecution | apv:Assumption | The calls of one request execute in request order on a shared environment. |
apv:SingleModelPerActor | apv:Assumption | All model calls of an actor are served by the actor's configured model. |
apv:ToolDefinitionsInput | prov:Role | Role of the tool-definitions fixture used by a model invocation. |
apv:Truncation | apv:SentTransform | Truncation inside request construction only. |
apv:Trusted | apv:IntegrityValue | Integrity label: trusted. |
apv:Untrusted | apv:IntegrityValue | Integrity label: untrusted. |
apv:VerdictAllow | apv:Verdict | The decision allowed its subject. |
apv:VerdictDeny | apv:Verdict | The decision denied its subject. |
apv:VerdictEscalate | apv:Verdict | The decision escalated its subject. |
apv:VerdictModify | apv:Verdict | The decision modified its subject. |
apv:VerdictOther | apv:Verdict | Any other verdict; the value is kept in apv:subtype. |
apv:VerdictRedact | apv:Verdict | The decision redacted its subject. |
apv:VerdictTerminate | apv:Verdict | The decision terminated the execution. |
apv:Withheld | apv:EpistemicStatus | Deliberately not exposed (provider-hidden reasoning; secrets redacted by capture). |
apv:WriteSink | apv:SecurityRole | Security role write_sink (SEM role). |