feat(api): update API spec from langfuse/langfuse a7fa476

[langfuse-bot]

Greptile Summary

The PR regenerates the unstable evaluation API from the latest upstream specification.

• Adds multi-evaluator assignments to evaluation rules, including create, update, and response models.
• Adds default evaluator mappings and optional bounds for numeric evaluator outputs.
• Updates evaluator identity and lifecycle contracts, removing evaluator scope and allowing assignment cleanup during deletion.
• Updates synchronous and asynchronous generated clients and package exports consistently.

Confidence Score: 5/5

The PR appears safe to merge with no concrete blocking or independently actionable non-blocking issues identified.

The generated request serialization, response models, exports, and synchronous and asynchronous client surfaces are internally consistent with the updated multi-evaluator API contract.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Client["SDK caller"] --> RuleRequest["Evaluation-rule request"]
  RuleRequest --> Assignments["One or more evaluator assignments"]
  Assignments --> Evaluator["Evaluator family"]
  Assignments --> Mapping["Default or rule-specific mapping"]
  Evaluator --> Latest["Latest evaluator version"]
  Mapping --> Target["Observation or experiment target"]

Loading

_{Reviews (1): Last reviewed commit: "feat(api): update API spec from langfuse..." | Re-trigger Greptile}

Context used:

• Knowledge Base — API Client (langfuse/api/)

[github-actions]

@claude review

[claude]

LGTM — this is an auto-generated Fern spec-sync PR; the flagged issues are docstring-only nits with no functional impact.

What was reviewed: verified the new/changed generated types (EvaluationRuleReadMapping, EvaluationRuleEvaluatorAssignment, CreateEvaluationRuleWithEvaluatorsRequest, PublicNumericEvaluatorOutputScoreDefinition, etc.) import cleanly and are internally consistent with the removal of EvaluatorScope; checked the __init__.py lazy-import tables (_dynamic_imports/__all__) stay in sync with the added/removed symbols; confirmed the evaluator/mapping deprecated-alias fields were consistently made optional alongside the new evaluators array across client, raw_client, and type files.

Extended reasoning...

Overview

This PR is a mechanical Fern-generated sync of the Python SDK's unstable API client from an upstream API definition (langfuse/langfuse@a7fa476). It removes the EvaluatorScope concept (project vs. managed), introduces multi-evaluator assignments on evaluation rules (evaluators array replacing the single evaluator/mapping pair), adds a stable evaluator id semantic across versions, adds bounded numeric score definitions, and raises the active-rule limit from 50 to 500. All 36 changed files are under langfuse/api/unstable/, i.e. generated client/model code, not hand-written business logic.

Security risks

None identified. There is no auth, crypto, or permission logic in this diff — it's pydantic model definitions and generated HTTP client wrappers for an already-unstable, versioned API surface.

Level of scrutiny

Low-to-medium is appropriate here: this is routine, repeatable spec-sync tooling output (the repo already has a prior commit with an identical PR title pattern), the actual behavior is delegated to the server-side API, and correctness is primarily about type/import consistency, which I verified directly by importing the changed modules. The two issues surfaced by the bug hunt are both docs-only (stale docstring prose referencing deleted EvaluatorScope semantics, and a pre-existing wrong import path convention in examples) with explicitly no runtime effect.

Other factors

No CODEOWNERS restriction applies to these paths. No prior claude[bot] review exists on this PR to reconcile against. I ran a syntax/AST check and a live import of all newly added and structurally changed types (assignment types, read-mapping type, evaluator types) and confirmed no import errors or circular-import issues from the EvaluatorScope removal.