[lalith0407]

HUMAN

While scoring with numpy-based evaluators, I noticed that since #2006 a numpy.bool_ score is sent as JSON true and a Decimal as the string "<Decimal>", because the new hand-built request only converts bool and int. I restored the old behavior of converting any numeric value to float, and extended the existing parity test with Decimal, Fraction and numpy cases.

What does this PR do?

Fixes #2057. Before #2006 every non-str score value went through CreateScoreRequest, which coerced it to float. The hand-built body only coerces bool/int, so numpy.bool_ is sent as true (rejected by the API's string | number schema) and Decimal as "<Decimal>" (stored as a categorical string). Now any non-str, non-float value that float() accepts is converted. Anything else passes through unchanged, as on main.

Type of change

• Bug fix

Verification

value	main	this branch
numpy.bool_(True)	true	1.0
Decimal("0.75")	"<Decimal>"	0.75
numpy.int64(3)	3	3.0 (matches v5.0.1)
True, 0.5, "good"	unchanged	unchanged

• test_create_score_payload_matches_request_model_serialization now includes Decimal and Fraction cases. It fails on main and passes here.
• New test_create_score_sends_numpy_values_as_numbers (skips if numpy isn't installed). It fails on main and passes here.
• pytest tests/unit: 955 passed. The 18 test_prompt.py errors also happen on unchanged main without LANGFUSE_* env vars. ruff check, ruff format --check, and mypy langfuse are clean.

Confidence Score: 4/5

The fix appears safe to merge, with a non-blocking gap in NumPy test coverage.

Summary

create_score now converts values such as Decimal, Fraction, and NumPy scalars to floats while leaving strings and existing floats unchanged.

• Score creation turns convertible numeric values into floats.

_{Reviews (1) · Last reviewed commit: "fix(scores): coerce numpy and Decimal sc..." · Reviewed by Greptile}

[claude]

Claude Code Review

This pull request is from a fork — automated review is disabled. A repository maintainer can comment @claude review to run a one-time review.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

CI skips the NumPy test

The new test uses pytest.importorskip("numpy"), but numpy is absent from the declared dependencies and uv.lock. CI installs those locked dependencies, so it never checks the NumPy cases that motivated this fix. Install NumPy in at least one test job and run this test there.

Prompt To Fix With AI

This is a comment left during a code review.
Path: tests/unit/test_score_consumer.py
Line: 817

Comment:
**CI skips the NumPy test**

The new test uses `pytest.importorskip("numpy")`, but `numpy` is absent from the declared dependencies and `uv.lock`. CI installs those locked dependencies, so it never checks the NumPy cases that motivated this fix. Install NumPy in at least one test job and run this test there.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

[chrikrah]

@lalith0407 From my perspective this can be merged at 0cda483. #2006 promised the queued values would match CreateScoreRequest, and its parity test says so by name, but the merged code converted only bool and int. This closes that gap for span.score, score_trace and the batch evaluator as well, since they all call create_score.

$ pytest tests/unit/test_score_consumer.py -q -p no:cacheprovider   # Python 3.12, numpy 2.5.4 installed by hand
38 passed, 1 skipped in 4.87s                 # 0cda483 (skip: gevent not installed)
2 failed, 36 passed, 1 skipped in 4.79s       # client.py from ab861f1, tests kept

non-blocking: I agree with the bot's note on line 817. numpy isn't in uv.lock, so CI skips the numpy test. Should numpy go in the dev group so CI runs it?