[aknight-ld]

JS SDK build cannot use the same variable names as what we have in python, so we have updated our spec to indicate target variable names. Updates tests accordingly to prove consistency with current standard.

This will be a breaking change, which is an acceptable risk for experimental features.

---

Note

Overview
Breaking change for experimental evaluations: init_evaluations() renames base_uri to api_base_uri so Python matches the cross-SDK spec (callers passing base_uri get a TypeError).

Host resolution is centralized in _resolve_host: option → env → default, with blank/whitespace treated as unset and trailing slashes stripped. Management API traffic uses api_base_uri / LD_API_BASE_URI; run links still use ui_base_uri / LD_UI_BASE_URI only—LD_BASE_URI is never used for evals API calls (relay/SDK delivery). Docs in README and agents.md spell out that split.

Tests replace the old override checks with coverage of precedence, blanks, slash normalization, SDK-delivery isolation, and an end-to-end run asserting staging API URLs do not change the default UI result link.

^{Reviewed by Cursor Bugbot for commit 941c216. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 89a0063. Configure here.}

[jeffdupont]

Thanks, the rename and _resolve_host look right. They match spec §8.2: a blank option now falls through to LD_API_BASE_URI, _env strips whitespace, both hosts drop a trailing /, and the name lines up with JS apiBaseUri. One thing blocks merge, and one follow-up is on the spec side.

1. Must fix: the new UI-host test fails before it reaches its assertions (Tests is red on 89a0063, the only failure of 2,863):

FAILED tests/test_evaluations_run.py::test_api_host_env_alone_leaves_result_url_on_default_ui_host
EvaluationsError: handler does not declare provides_for.

run() rejects the plain async def handler and the generation dict with no mode before any request goes out. So the spec rule this test exists for (§8.2: the UI host is not derived from LD_API_BASE_URI) is still untested. The fix is two lines, matching the other run() tests in this file: handler=tagged(handler) and "mode": "completion" in generation. Bugbot flagged the same thing.

2. Should fix (separate monorepo PR): remove the "Python pending" notes from the spec. TESTING.md on main still has three [host overrides: Python pending] tags (§8.2 bullets 1 and 4, and the cross-SDK parity paragraph). The "What Python still lacks" bullet describes exactly what this PR fixes. Once this merges, those can go, so the spec doesn't claim a gap that no longer exists.