[donei003]

Promotes Evaluations from Code from launchdarkly_ai_server.experimental.evaluations to the package root. This follows the Agent Skills promotion in #210.

Draft. This PR releases together with launchdarkly/js-ai-sdk#181. A Jira ticket is not linked yet.

Migration

Import Evaluations from Code names from launchdarkly_ai_server instead of launchdarkly_ai_server.experimental.evaluations. The names, signatures and behavior do not change.

# before
from launchdarkly_ai_server.experimental.evaluations import EvalTool, init_evaluations
# after
from launchdarkly_ai_server import EvalTool, init_evaluations

This is a breaking change (!). It bumps the minor version while the package is 0.x.

What changed

• Root exports. The 14 names that experimental/evaluations.py exported are now in launchdarkly_ai_server.__all__, with the same spelling. No name collides with a current root export. launchdarkly-ai-python gets the names through its existing import *.
• Experimental module removed. experimental/evaluations.py is deleted. experimental/__init__.py stays as a docstring only, so a future experimental feature has a place to go.
• Usage ID. init_evaluations now reports client.initEvaluations, not experimental.initEvaluations. Queries that count the old ID must also count the new one.
• API reports. Regenerated with make api-report. The names are in launchdarkly_ai_server.txt, and launchdarkly_ai_server.experimental.evaluations.txt is deleted.
• Changelog. The three evaluations rows are removed from changelog-sections, so feat(evaluations): lands under Features. The AGENTS.md scope examples now use a placeholder scope.
• Docs. The client README, the launchdarkly-ai-python README, packages/client/agents.md and AGENTS.md no longer call Evaluations from Code experimental. The README heading anchor changed from #run-an-evaluation-from-code-experimental to #run-an-evaluation-from-code.

Tests

• TestPackageExports checks that every evaluations name is on the root and that experimental.evaluations no longer imports.
• test_public_surface.py adds the 14 names to the root snapshot.
• test_reexports.py checks that the barrel re-exports the names.
• test_core_does_not_import_experimental_code now checks only launchdarkly_ai_server.experimental, because the root loads evaluations. A new test checks that lifecycle imports no evaluations module.
• test_sdk_usage.py expects client.initEvaluations.

Import time

import launchdarkly_ai_server now loads the evaluations package. Measured with python -X importtime, it adds about 33 ms to a root import of about 130 ms. The largest part is evaluations.api.

Checks

• make lint, make format-check, make typecheck, make api-report-check: pass
• python -m scripts.api_breaking pr --base origin/main: no breaking changes to the public API
• make test: 3640 passed, 18 skipped
• Run one evaluation end to end against staging with the root import

Conflicts

The open evaluations PRs #228, #227, #143, #135 and #92 edit the same test files. Merge them first, or rebase them after this PR.

Related PRs

Releases together with:

• feat(evaluations)!: promote Evaluations from Code to core js-ai-sdk#181

🤖 Generated with Claude Code

---

Note

Overview
Evaluations from Code moves from launchdarkly_ai_server.experimental.evaluations to the launchdarkly_ai_server package root (and through launchdarkly-ai-python re-exports). The experimental shim is removed; callers should import init_evaluations, EvalTool, Judge, Scorer, and the rest from the root instead.

The public API contract shifts accordingly: generated launchdarkly_ai_server.txt now lists the evaluations surface, and launchdarkly_ai_server.experimental.evaluations.txt is deleted. init_evaluations usage telemetry changes from experimental.initEvaluations to client.initEvaluations.

Release/changelog policy is updated: evaluations is no longer an Experimental scope in release-please-config.json, and AGENTS.md / READMEs describe Evaluations (and Skills) as core, with docs and examples using root imports and a non-experimental README anchor.

Tests and snapshots (test_public_surface, TestPackageExports, test_reexports, lifecycle import guards) enforce the new export layout and that experimental.evaluations cannot be imported.

^{Reviewed by Cursor Bugbot for commit 3ee3612. Bugbot is set up for automated code reviews on this repo. Configure here.}

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.