fix(gateway): pass the invalid_guard_verdict worker code through

[dragosboca]

Why

Since #285 a guard model whose first output positions carry no complete Yes/No distribution terminates with error_code="invalid_guard_verdict" and "guard model did not produce a valid thresholded verdict". The Python and TypeScript SDK READMEs already tell callers to handle that code. The gateway's worker error allowlist (queue/streaming.rs::client_safe_worker_error_code) predates it, so the terminal collapsed to inference_error / HTTP 500 / "internal error during generation" on the buffered path and "Generation terminated with an upstream error." on the streaming path. A caller has no way to tell "the guard could not produce a verdict" from a backend fault.

Reproducible with POST /v1/generate/ibm-granite__granite-guardian-3.0-2b {"prompt": "...", "max_new_tokens": 4} (a raw prompt has no Guardian template, so the model answers in prose): the body is the generic 500 instead of the typed code.

What

• http_error.rs: openai_code::INVALID_GUARD_VERDICT.
• queue/streaming.rs: admit the code in client_safe_worker_error_code and pin it in the contract test.

Status and OpenAI type stay the defaults (500 / server_error), the same settlement as empty_model_output: terminal, non-retryable, tokens were consumed.

Verification

• cargo test --lib worker_error_public_contract: 2 passed.
• cargo fmt --check and cargo clippy --lib -D warnings: clean.

Summary by CodeRabbit

• Bug Fixes
  • Guard-model responses with incomplete Yes/No verdicts now return a clear, non-retryable error code.
  • This error code is preserved when passed through the worker, improving error reporting for clients.
  • Added coverage to verify the error-code handling.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Team

Run ID: 29ea26ec-71ed-46f4-b2f6-9729b32ec729

📥 Commits

Reviewing files that changed from the base of the PR and between 120059f and 86a2ae2.

📒 Files selected for processing (2)

• packages/sie_gateway/src/http_error.rs
• packages/sie_gateway/src/queue/streaming.rs

Included review availability: Your plan provides up to 10 included reviews per hour; 7 remain after this review.

---

📝 Walkthrough

Walkthrough

The gateway adds invalid_guard_verdict as a public OpenAI error code. Client-safe worker error mapping preserves this code, and a regression test covers the allowlist.

Changes

Guard verdict error code handling

Layer / File(s)	Summary
Error code declaration and propagation packages/sie_gateway/src/http_error.rs, packages/sie_gateway/src/queue/streaming.rs	The gateway declares INVALID_GUARD_VERDICT, preserves it in client-safe worker error mapping, and tests the sanctioned error-code list.

Suggested reviewers: huronat

Priority: ⬇️ Low

Merge Risk: ⚪ Minimal · up to 86a2a

The new terminal error code is preserved for clients across buffered and streaming responses without changing the intended server-error behavior.

🚥 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 and concisely describes the main change: passing the invalid_guard_verdict worker code through the gateway.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 2 files.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

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