test(decisions): add hosted_vllm /v1/decisions and provider-400 translation cases

[devin-ai-integration]

TLDR

Problem this solves:

• feat(decisions): add hosted_vllm provider #45501 added hosted_vllm to Decisions but its only TranslationTestCase covers /v1/systemone
• /v1/decisions, the route with the actual translation logic, had no exact-literal case
• vLLM's 400 for non-choice questions had no case either, since the runner always replied 200

How it solves it:

• TranslationTestCase gets mock_provider_status_code (default 200), assert_translation passes it to Reply
• QWEN3_0_6B_DECISIONS_TEST_CASE: OpenAI choice body in, exact vLLM /v1/systemone body out, OpenAI answer back
• QWEN3_0_6B_PREDICATE_REJECTED_TEST_CASE: OpenAI predicate goes out as noul, vLLM's 400 comes back as LiteLLM's 400
• Both provider replies captured from a real vLLM (CPU build of vllm main 702ce313f, Qwen/Qwen3-0.6B) with the exact bodies the cases pin

Linear ticket

Related to LIT-9367

Pre-Submission checklist

• I have added meaningful tests
• The handful of test files covering my change pass locally, e.g. uv run pytest tests/unit/<your_test_file>.py -v. Leave the suites (make test-unit-*, make test-unit) to CI: it finishes in ~15 minutes where a laptop takes an hour or more
• My PR passes all required CI/CD checks (e.g., lint, schema.d.ts sync check, etc.)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Screenshots / Proof of Fix

Test-only PR, so the proof is the cases going red under mutation and green on the shipped code. Local integration rig (tests/integration/proxy_config.yaml, fake provider on :8191), tests/integration/translation/decisions/basic/test_decisions_basic_hosted_vllm.py

Before (05166e2, runner mutated)

Mutation 1: assert_translation ignores mock_provider_status_code and replies 200

1. python tests/integration/run.py providers tests/integration/translation/decisions/basic/test_decisions_basic_hosted_vllm.py
2. hosted_vllm/Qwen/Qwen3-0.6B-predicate_rejected fails: assert 500 == 400, LiteLLM turns the 200 error body into InternalServerError ... 1 validation error for DecisionsResponse. 1 failed, 2 passed

Mutation 2: BaseDecisionsConfig sends choice criteria under the key options

1. Same command after restarting the proxy
2. basic and decisions fail on the provider body: {'is_defect': {'type': 'choice', ..., 'options': {'yes': None, 'no': None}}} == {... 'criteria': ...}. 2 failed, 1 passed

After (37d3bab)

1. Same command, mutations reverted
2. 3 passed, and a second run against the same Redis also 3 passed (the no-cache flag holds)
3. All six decisions bases plus tests/integration/providers/test_decisions_wire.py still pass with the runner change

Provider captures, exact expected_provider_request bodies sent to the real vLLM on 2026-10-09:

1. choice: POST /v1/systemone -> 200 {"id":"decision-91ce41627dbf016f","object":"structured_decision",...,"answers":{"is_defect":{"type":"choice","choice":"yes","probabilities":{"yes":0.8670357477770336,"no":0.13296425222296632},"confidence":0.8669148648509282}},"usage":{"input_tokens":43,"output_tokens":1},...}
2. noul: POST /v1/systemone -> 400 {"error":{"message":"unknown question type 'noul'; supported: ['choice']","type":"BadRequestError","param":null,"code":400}}

Type

✅ Test

Caveats (if any)

Low

• The 400 case pins LiteLLM's wrapped error text, including model group 'hosted_vllm/Qwen/Qwen3-0.6B', so a change to proxy error wording fails it on purpose
• No other decisions provider has a /v1/decisions case yet; same pattern applies if wanted

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

Link to Devin session: https://app.devin.ai/sessions/3d0a4b725c1d4d65968372a35704cdd2
Open in Devin Desktop: https://app.devin.ai/desktop/session/3d0a4b725c1d4d65968372a35704cdd2?variant=devin
Requested by: @kerry-berri

---

Note

Devin errored when opening this Pull Request as kerry-berri.
As a fallback, Devin opened this PR as itself.

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[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.

---

kerry seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[greptile-apps]

Confidence Score: 5/5

[Low impact] This test-only PR appears safe to merge

Summary

Adds two hosted_vllm translation cases and lets the fake provider return a chosen HTTP status

• Hosted vLLM choice questions now have an exact /v1/decisions translation case.
• Hosted vLLM predicate tests now check its 400 error translation.

devin-ai-integration[bot] explicitly acknowledged that the exact wrapped error text is pinned intentionally, so changes to that wording should fail the case

_{Reviews (1) · Last reviewed commit: "test(decisions): add hosted_vllm /v1/dec..." · Reviewed by Greptile}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!