# Arize Phoenix

CURRENT_STATUS: UNKNOWN
OBSERVATION: SPEC_ONLY
FRESHNESS: UNKNOWN
PAGE_TYPE: reference
CANONICAL_URL: /tools/ea-171-arize-phoenix
CLAIM_IDS: none
EVIDENCE_IDS: EA-171-R11-RESEARCH, EA-171-R11-WEB-READBACK
UNKNOWN_IDS: EA-171-UNKNOWN-01, EA-171-UNKNOWN-CURRENTNESS
OPS_IDS: none

## Summary
Arize Phoenix is documented as an open-source AI observability and evaluation platform built on OpenTelemetry and OpenInference. Its documented surface includes traces for model calls, retrieval, tool use and custom logic, together with evaluation, versioned datasets and experiments. Phoenix OSS and the managed Arize AX workflow are separate operational and contractual boundaries.

## Current judgment
Phoenix documentation describes OpenTelemetry/OpenInference-based instrumentation and OTLP ingestion for AI application tracing. Documentation and repository availability establish capabilities, not complete instrumentation, secure deployment, evaluator validity, or enterprise SLO.

## Scope
Arize Phoenix

## Confirmed
- Phoenix documentation describes OpenTelemetry/OpenInference-based instrumentation and OTLP ingestion for AI application tracing. Documentation and repository availability establish capabilities, not complete instrumentation, secure deployment, evaluator validity, or enterprise SLO.

## UNKNOWN
- EA-171-UNKNOWN-01: Current component releases; exact ELv2 scope for the intended use; instrumentation matrix; OpenTelemetry/OpenInference parity; trace loss and overhead; evaluator behavior; self-host security defaults; managed feature, region, retention and SLO behavior; public incident completeness.
- EA-171-UNKNOWN-CURRENTNESS: 外部根拠の再確認日がcanonical body/control-planeで確定していないため、currentnessはUNKNOWNとして保持する。日付を推測補完しない。

## Evidence
- EA-171-R11-RESEARCH (RESEARCH): EA-171 監査済みresearch / canonical revision binding
- EA-171-R11-WEB-READBACK (IMPL): EA-171 Web body identity / canonical readback

## Failure / limits
Do not treat Phoenix's OSS label, OpenTelemetry foundation, integration list, or UI trace as proof of complete observability, safe privacy, correct evaluation, or Arize AX equivalence.
## Practice
Phoenix can be used as a locally controlled observability/evaluation layer, but an OSS label, OpenTelemetry support, integration listing, or visible trace is not proof of complete visibility, privacy-safe capture, correct evaluation, production readiness, or parity with Arize AX.
## Practice
Pin and verify the Phoenix release/commit and package versions, Phoenix OSS versus Arize AX boundary, ELv2 obligations, OTLP collector/exporter/backend, supported framework/provider/language, auto/manual instrumentation, sampling and redaction, authentication/RBAC/tenant controls, storage/backup/retention/region, dataset/evaluator versions, deployment/upgrade path, cost and SLO.
## Practice
Evidence-only desk research. No Phoenix server, UI, SDK, OTLP exporter, evaluation, dataset, or Arize AX tenant was exercised by the source packet.
## Practice
Traces can contain prompts, outputs, retrievals, tool calls and identities. Redaction and access controls must be applied before export; collector, backend, UI and backups require protection. Self-hosting transfers compute, storage, backup, upgrade, patching and incident-response obligations to the deployer.
## PUBLIC_CLAIM
Arize Phoenix is documented as an open-source AI observability and evaluation platform built on OpenTelemetry and OpenInference. Its documented surface includes traces for model calls, retrieval, tool use and custom logic, together with evaluation, versioned datasets and experiments. Phoenix OSS and the managed Arize AX workflow are separate operational and contractual boundaries.

## ESTABLISHED
Phoenix documentation describes OpenTelemetry/OpenInference-based instrumentation and OTLP ingestion for AI application tracing. Documentation and repository availability establish capabilities, not complete instrumentation, secure deployment, evaluator validity, or enterprise SLO.

## INFERENCE
Phoenix can be used as a locally controlled observability/evaluation layer, but an OSS label, OpenTelemetry support, integration listing, or visible trace is not proof of complete visibility, privacy-safe capture, correct evaluation, production readiness, or parity with Arize AX.

## AVAILABILITY_GATES
Pin and verify the Phoenix release/commit and package versions, Phoenix OSS versus Arize AX boundary, ELv2 obligations, OTLP collector/exporter/backend, supported framework/provider/language, auto/manual instrumentation, sampling and redaction, authentication/RBAC/tenant controls, storage/backup/retention/region, dataset/evaluator versions, deployment/upgrade path, cost and SLO.

## IMPLEMENTATION_STATE
Evidence-only desk research. No Phoenix server, UI, SDK, OTLP exporter, evaluation, dataset, or Arize AX tenant was exercised by the source packet.

## UNKNOWN
Current component releases; exact ELv2 scope for the intended use; instrumentation matrix; OpenTelemetry/OpenInference parity; trace loss and overhead; evaluator behavior; self-host security defaults; managed feature, region, retention and SLO behavior; public incident completeness.

## SECURITY_PRIVACY_AUDIT_OPERABILITY
Traces can contain prompts, outputs, retrievals, tool calls and identities. Redaction and access controls must be applied before export; collector, backend, UI and backups require protection. Self-hosting transfers compute, storage, backup, upgrade, patching and incident-response obligations to the deployer.

## STOP_CONDITION
Do not treat Phoenix's OSS label, OpenTelemetry foundation, integration list, or UI trace as proof of complete observability, safe privacy, correct evaluation, or Arize AX equivalence.

NEXT: /agent/start.json
STOP: This is a knowledge projection, not an execution order.