fix(chat): forward caller service_tier through the chat-to-responses conversion (#1886)

[olddonkey]

Summary

Independent bug-fix unit of the FastWire umbrella (#1886): the chat→responses converter (chatCompletionsToResponsesBody) dropped service_tier, so a tier declared by a /v1/chat/completions caller vanished before the responses pipeline could see it — fast-mode override, the capability gate, and serialization all behaved as if no tier was requested, while the identical request through /v1/responses worked. The two inbound entries now agree.

One production line: copy service_tier under the converter's existing optional-scalar convention (typeof === "string", matching the adjacent user / prompt_cache_key handling). No normalization, no injection — the downstream pipeline keeps owning tier semantics, so the fail-closed gate still strips it on supportsServiceTier: false routes.

Stacked on the A0 characterization PR #1892, deliberately: A0 locked this bug as a (known bug) characterization cell, and the design's flip protocol says the bug-fix unit updates that cell in the same change. Review this PR by its top commit. (#1892 is now merged as dec332c49; no rebase is needed — this branch already contains its commit 0cdd07d51 through common history.)

Tests

• Converter: caller-supplied tier preserved; no tier invented when absent.
• End-to-end through handleChatCompletions (forced chat→responses fallback): caller tier reaches the outbound body on a tier-capable route; stripped on an explicit supportsServiceTier: false route.
• A0 characterization cell flipped from drop-assertion to preserve-assertion, annotated with the unit reference.

Verification

• Full suite at this commit: 12,750 pass / 10 skip / 0 fail (12,760 tests across 823 files).
• bun x tsc --noEmit clean.

Part of #1886 (independent bug-fix unit; shrinks the later B1 migration surface).

🤖 Generated with Claude Code

Review readiness checklist

This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:

• All CI tests are green on my local testing.

• I pushed my PR to the latest dev commit.

• I resolved all correct Codex and CodeRabbit findings.

• My PR is ready for review.

Summary by CodeRabbit

• New Features

  • Service-tier options are now preserved when converting chat completion requests.
  • Service tiers are forwarded to compatible providers and omitted when unsupported.
  • Provider metadata and routing now accurately reflect service-tier availability.

• Tests

  • Added coverage for supported, unsupported, and unclassified service-tier scenarios.
  • Added validation for exact-model forwarding, compatibility behavior, logging, and request handling.

[github-actions]

✅ READY

• all PR quality gates passed; the review readiness checklist is complete.

Review readiness checklist

• ✅ All CI tests are green on my local testing.
• ✅ I pushed my PR to the latest dev commit.
• ✅ I resolved all correct Codex and CodeRabbit findings.
• ✅ My PR is ready for review.

✅ 4/4 boxes ticked.

This pull request is already Ready for Review.
The review-ready label marks this PR as ready; review automation runs independently.
Maintainers: @lidge-jun @Ingwannu @Wibias

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 6614d378-d4da-4364-a3f0-1ac1bd74656d

📥 Commits

Reviewing files that changed from the base of the PR and between c603dcd and 936aef8.

📒 Files selected for processing (1)

• tests/chat-completions-endpoint.test.ts

Included review availability: Your plan includes up to 10 reviews per rolling hour; 9 remain after this review.

---

📝 Walkthrough

Walkthrough

The Chat Completions translator now forwards service_tier to the Responses request body. Tests cover conversion, endpoint fallback behavior, routing compatibility, catalog metadata, adapter forwarding, logging, and raw-body handling.

Changes

Service tier support

Layer / File(s)	Summary
Chat conversion and endpoint forwarding src/chat/inbound.ts, tests/chat-completions-endpoint.test.ts	The converter preserves a supplied string service_tier and omits it when absent. Endpoint tests verify forwarding for capable providers and removal for unsupported providers.
Routing and metadata compatibility tests/fastwire-characterization-routing.test.ts	Routing tests cover service-tier capability checks, compatibility projections, and catalog metadata fields for supported and unsupported providers.
Adapter and wire characterization tests/fastwire-characterization-wire.test.ts	Wire tests cover service-tier forwarding and logging across response and chat adapters, plus raw-body observation and related adapter behavior.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Merge Risk: ⚪ Minimal · up to 936ae

This is a localized fix that preserves an optional service tier during chat-to-responses conversion, with coverage for supported and fail-closed routes. No actionable merge-blocking risk remains after normal checks and review.

Possibly related PRs

• lidge-jun/opencodex#1512: Both changes modify Chat Completions service-tier forwarding and its capability-gated tests.
• lidge-jun/opencodex#1892: This change addresses the characterized Chat-to-Responses service_tier behavior.
• lidge-jun/opencodex#1893: This change extends the related FastWire service-tier forwarding tests.

Suggested labels: review-ready

Suggested reviewers: lidge-jun

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 25.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
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.
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly and concisely describes forwarding caller-provided service_tier through the chat-to-responses conversion.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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

[olddonkey]

Gate transparency for the merged head (936aef8be, dev merged in at 90fdc3a62):

• Full suite: 12,783 pass / 10 skip / 1 fail (12,794 tests, 825 files).
• The single failure — CLI key-login live-update overlay preservation > notify after key login pushes the merged row and keeps modelCosts on live and disk (tests/key-login-live-update.test.ts:100, live row comes back with modelCosts: undefined) — reproduces identically on pristine upstream/dev at 90fdc3a62 with none of this PR's changes present. It is a pre-existing dev-side regression, unrelated to this unit (which touches the chat→responses converter and service-tier tests only).
• At this PR's pre-merge base the suite was fully green: 12,750 pass / 0 fail.

Flagging so the baseline failure gets triaged on dev rather than attributed to this branch.

[lidge-jun]

Its baseline is in: #1892 merged as dec332c49, so the A0 characterization is now on dev.

This PR is next in the Wave 5B order and is being held only by its own draft state — all four
readiness boxes are unticked, and enforce-target and label are CANCELLED at head, so there
is no passing gate to merge against. That is the checklist doing its job, not an objection to
the change.

Two things worth knowing before you tick the boxes:

• No rebase is required. This branch already contains test(fastwire): A0 characterization — lock current service-tier behavior (#1886) #1892's commit 0cdd07d51, so the
  shared fastwire-characterization-* files resolve through common history rather than as an
  add/add conflict. I verified a sequential merge of the 5B train onto origin/dev in a scratch
  worktree: five merges, zero conflicts. The body's "will rebase once test(fastwire): A0 characterization — lock current service-tier behavior (#1886) #1892 merges" is now
  optional tidiness.
• The one file whose content differs from test(fastwire): A0 characterization — lock current service-tier behavior (#1886) #1892 is the intentional flip —
  characterization (known bug): ...drops service_tier becoming
  characterization: ...preserves service_tier — which is exactly the A0 protocol working as
  designed. A reviewer seeing that diff should read it as the point, not a conflict.

Ready for review-readiness whenever you are.

[olddonkey]

Thanks — and the draft state you saw is now cleared: head 4b74f6359 has all four readiness boxes ticked, hygiene / enforce-target / label / resolve-pr all green, and the gate comment flipped to ✅ READY at that SHA.

On your two points:

• No rebase — agreed, and the branch took the merge route rather than a rebase, so the shared history you verified is intact. I've dropped the "will rebase" line from the description as obsolete.
• The flipped cell is the point — exactly right. characterization (known bug): …drops service_tier → characterization: …preserves service_tier is the A0 protocol: A0 locks the bug as current behavior, and the bug-fix unit flips that same cell in the same change, so the two PRs read as one intentional sequence.

Local full-suite evidence at 4b74f6359 (which now contains dev tip 2a9f08324): 12,788 pass / 10 skip / 1 fail.

The single failure is the pre-existing dev-side one I flagged above — tests/key-login-live-update.test.ts:100, live row returns modelCosts: undefined — and it reproduces on pristine upstream/dev with none of this branch's changes present. Unrelated to this unit; worth its own triage on dev. Happy to open a separate issue for it if that's useful.

One caveat for the 5B ordering: four suites (oauth-refresh-lock-multiprocess, port-reclaim, codex-shim, autostart-health) are load-sensitive here — they time out under a fully parallel run and pass in isolation — so a red in those on a busy runner is likely noise rather than a real regression.

[lidge-jun]

Validation before merge: scratch-worktree merge onto dev (post-#1965) — fastwire-characterization-wire + chat-completions-endpoint suites 112/0, tsc --noEmit clean. Review threads all resolved; hunk overlap with landed B1 confirmed disjoint by the plan-audit reviewer.