fix(metadata): give gpt-5-mini reasoning headroom for tagging/validation succeed

[drozado]

Fixes #249

gpt-5-mini is a reasoning model. Calling it with max_output_tokens=50 (tag) / 500
(validate) and default (medium) reasoning makes it exhaust the budget on reasoning tokens and
return status=incomplete (max_output_tokens) every time, so metadata silently fell back to
category="Other" / valid_question=True.

Changes

• helpers/metadata_llm.py: pass reasoning={"effort": "minimal"} (covers tag + validate).
• tag_questions/main.py: max_output_tokens 50 → 512.
• validate_questions/main.py: max_output_tokens 500 → 1024.
• test_model_request_params.py: assert a reasoning policy is present rather than pinning the
  exact effort value, so future model/reasoning/budget changes don't break the test.

Summary by CodeRabbit

• New Features

  • Improved metadata responses with lighter reasoning guidance for faster, more efficient results.
  • Increased response limits for question tagging and validation to reduce the likelihood of truncated outputs.

• Bug Fixes

  • More reliable question categorization and validation when generating longer model responses.

• Tests

  • Updated automated checks to reflect the new model response settings.

[Copilot]

Pull request overview

This PR fixes metadata tagging/validation failures caused by gpt-5-mini exhausting its token budget on reasoning, which led to systematic fallbacks (category="Other" / valid_question=True) and silent mislabeling in the metadata pipeline under src/metadata/.

Changes:

• Configure metadata LLM calls to use reasoning={"effort": "minimal"} to prevent reasoning-token starvation.
• Increase max_output_tokens for tagging (50 → 512) and validation (500 → 1024) to give responses enough output budget.
• Relax the request-parameter unit test to assert that a reasoning policy is present without pinning an exact effort value.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
src/helpers/metadata_llm.py	Adds a minimal reasoning policy to metadata model requests to avoid incomplete responses.
src/metadata/tag_questions/main.py	Increases tagging output budget to reduce incomplete/empty outputs and fallback categorization.
src/metadata/validate_questions/main.py	Increases validation output budget to reduce incomplete outputs and fallback validity labeling.
src/tests/test_model_request_params.py	Updates test expectations to require a reasoning policy while avoiding brittle effort-value assertions.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[nikbpetrov]

@houtanb This is a good catch for a bug and raises the possibility that not all LLM args/setup is being forwarded/works exactly as expected. Perhaps a smoke test-like cli can run a few more reqests and check outputs for systematic problems/refusals. But that's separate.

LGTM

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 50524307-5ac0-423e-8944-77636d198780

📥 Commits

Reviewing files that changed from the base of the PR and between 6d01eea and 067519a.

📒 Files selected for processing (4)

• src/helpers/metadata_llm.py
• src/metadata/tag_questions/main.py
• src/metadata/validate_questions/main.py
• src/tests/test_model_request_params.py

🚧 Files skipped from review as they are similar to previous changes (4)

• src/helpers/metadata_llm.py
• src/metadata/tag_questions/main.py
• src/metadata/validate_questions/main.py
• src/tests/test_model_request_params.py

---

📝 Walkthrough

Walkthrough

This change adds reasoning={"effort": "minimal"} to the shared metadata model call and raises max_output_tokens in tag_questions and validate_questions, with a matching test update.

Changes

Metadata LLM budget and reasoning fix

Layer / File(s)	Summary
Shared metadata LLM reasoning effort src/helpers/metadata_llm.py, src/tests/test_model_request_params.py	get_metadata_model_response() now passes reasoning={"effort": "minimal"} to get_response(), and the test asserts this kwarg is included in the call.
Caller token budget increases src/metadata/tag_questions/main.py, src/metadata/validate_questions/main.py	max_output_tokens is raised from 50 to 512 in tag_questions and from 500 to 1024 in validate_questions for their respective LLM calls.

Estimated code review effort: 1 (Trivial) | ~5 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly reflects the main fix: increasing reasoning headroom for metadata tagging and validation.
Linked Issues check	✅ Passed	The changes match the issue fix by setting minimal reasoning and raising output token limits for tagging and validation.
Out of Scope Changes check	✅ Passed	All code changes stay within the requested metadata tagging/validation fix and test update.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}