name: Arize Phoenix Vocabulary
description: Domain vocabulary for the Arize Phoenix AI observability and evaluation
  platform, covering LLM tracing, evaluation, dataset management, and experiment concepts.
url: https://arize.com/docs/phoenix
version: '1.0'
terms:
- term: Span
  definition: A unit of work or operation captured during LLM execution, representing
    a single traced activity with timing, inputs, outputs, and metadata.
  relatedTerms: []
- term: Trace
  definition: A collection of spans that represent a complete end-to-end request through
    an LLM application, linked by a trace ID.
  relatedTerms: []
- term: Annotation
  definition: A human or automated label attached to a span or trace, used to evaluate
    quality, correctness, or other dimensions of LLM output.
  relatedTerms: []
- term: Dataset
  definition: A named collection of data records used as input for LLM experiments,
    evaluations, or fine-tuning.
  relatedTerms: []
- term: DatasetVersion
  definition: A snapshot of a dataset at a specific point in time, enabling reproducible
    experiments.
  relatedTerms: []
- term: Experiment
  definition: A structured evaluation run that applies a task or prompt to a dataset
    and records the resulting outputs and metrics.
  relatedTerms: []
- term: ExperimentRun
  definition: A single execution within an experiment, pairing one dataset record
    with its corresponding LLM output.
  relatedTerms: []
- term: Project
  definition: A logical grouping of traces, spans, and experiments in Phoenix, typically
    corresponding to a single LLM application or use case.
  relatedTerms: []
- term: AnnotationConfig
  definition: A schema definition for an annotation, specifying the label name, type
    (categorical or continuous), and allowed values.
  relatedTerms: []
- term: EvaluationResult
  definition: The outcome of an evaluation step, including a score, label, and optional
    explanation produced by a judge or human reviewer.
  relatedTerms: []
- term: InferencesSpan
  definition: A span specifically representing an LLM model inference call, capturing
    the input prompt, output, token counts, and model name.
  relatedTerms: []
- term: SpanAnnotation
  definition: An annotation applied to a specific span, enabling evaluation at the
    span level within a trace.
  relatedTerms: []
- term: Session
  definition: A sequence of related traces representing a multi-turn conversation
    or user session with an LLM application.
  relatedTerms: []
- term: PromptVersion
  definition: A versioned snapshot of a prompt template, enabling comparison and reproducibility
    across experiments.
  relatedTerms: []
- term: OpenTelemetry
  definition: An open-source observability framework used by Phoenix for collecting,
    processing, and exporting telemetry data (traces, metrics, logs) from LLM applications.
  relatedTerms: []
- term: OTLP
  definition: OpenTelemetry Protocol — the wire protocol used to ingest span and trace
    data into Phoenix from instrumented LLM applications.
  relatedTerms: []
- term: AutoInstrumentation
  definition: Automatic injection of tracing code into LLM framework calls (LangChain,
    LlamaIndex, OpenAI, etc.) without manual span creation.
  relatedTerms: []
- term: LLMOps
  definition: Practices and tooling for operationalizing LLM applications in production,
    including monitoring, evaluation, debugging, and iteration.
  relatedTerms: []
- term: BearerToken
  definition: The authentication mechanism for the Phoenix REST API, passed as an
    Authorization header value.
  relatedTerms: []
- term: CursorPagination
  definition: Phoenix pagination strategy using opaque cursor tokens to navigate through
    large result sets.
  relatedTerms: []