fix(catalog): account for auxiliary model charges

[luvs01]

Summary

• Discovered pricing classification in discoveredPricingStatus only inspected the prompt/input and completion/output pair, so a model whose token rates are zero but that publishes a positive auxiliary charge (request, image, web_search, ...) could classify as free and appear under a free-only filter.
• The classifier now scans every published component of the provider pricing object through discoveredPricingRate: any positive numeric rate is paid, any unusable component is unknown, and free still requires the complete zero prompt/completion pair.
• Regression tests cover a positive per-request auxiliary charge, fully-zero auxiliary dimensions, and unsupported/malformed auxiliary pricing shapes.

Verification

• Head: 8694d6cc6e6b791d720a0d0d20ffbe927138ccf3 (tree 0e72c21e033f4fc14742c73dc33608170378d395), based on dev ad9eab393b1125ca7bdfa3e83ed4fa0e0b0475e9.
• bun test tests/codex-integration/catalog-free-pricing-status.test.ts - 13 pass / 0 fail, including the two new auxiliary-charge cases.
• bun x tsc --noEmit - clean.
• bun run structure:check - passed.
• bun run privacy:scan - passed.
• bun scripts/file-size-ratchet.ts - passed.

Remaining gates

• Fork CI dispatched via gh workflow run ci.yml on this branch: https://github.com/luvs01/opencodex/actions/runs/35289987101. The macOS control job is expected to cancel at the 30-minute mark; that is the known upstream limitation tracked in [Bug]: macOS control is cancelled near its 30-minute limit in full dispatch CI #4905, not a signal from this change. Occasional shard flakes are likewise known upstream noise.
• No gui/ files are touched, so the UI-screenshot gate does not apply.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed.
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults.

• All CI tests are green on my local testing. Local gates passed (see Verification); fork run 35289987101 dispatched on the exact head.
• 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

• Bug Fixes

  • Pricing status now accounts for all published pricing components, including auxiliary charges.
  • Models with any positive charge are correctly classified as paid.
  • Models with unsupported or invalid pricing values are classified as unknown instead of free.
  • Models are marked free only when all supported pricing components are zero and required pricing details are present.

• Tests

  • Added coverage for auxiliary charges, zero-cost models, and unsupported pricing formats.

[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: Advanced

Run ID: 0a7075d1-26a0-491e-8752-7651f4dbfee7

📥 Commits

Reviewing files that changed from the base of the PR and between 6467235 and 8694d6c.

📒 Files selected for processing (2)

• src/codex/catalog/model-hints.ts
• tests/codex-integration/catalog-free-pricing-status.test.ts

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

---

📝 Walkthrough

Walkthrough

Changes

Pricing status classification

Layer / File(s)	Summary
Fail-closed pricing classification and regression coverage src/codex/catalog/model-hints.ts, tests/codex-integration/catalog-free-pricing-status.test.ts	discoveredPricingStatus now scans all pricing components. Positive numeric rates classify a model as paid; unsupported values classify it as unknown; a complete all-zero pricing set classifies it as free. Tests cover auxiliary charges and unsupported pricing shapes.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Merge Risk: ⚪ Minimal · up to 8694d

No actionable merge-blocking risk remains after the targeted pricing checks and regression coverage.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 50.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 2 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 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: updating catalog pricing classification to account for auxiliary model charges.
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.

• Fix all pre-merge checks with AI

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

---

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.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[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

[lidge-jun]

리뷰 · 우선순위 73 / 80

이 PR은 카탈로그 Free 필터용 discoveredPricingStatus가 prompt/completion(또는 input/output)만 보고 보조 요금을 무시하던 구멍을 막습니다. 지금 tip(e80e571f63, package 2.59.0)의 src/codex/catalog/model-hints.ts는 쌍이 모두 0이면 free입니다. 토큰 단가는 0인데 request/image/web_search 같은 양수 보조 요금이 있으면 Free 목록에 올라가 사용자 돈을 쓰게 됩니다. #3666 계약의 "유료를 Free에 넣지 말 것(fail-closed)"과 정면으로 어긋납니다.

고침은 Object.values(pricing)를 discoveredPricingRate로 훑어, 양수가 하나라도 있으면 paid, 파싱 불가 성분이 있으면 unknown, 전부 비음수 0이고 prompt/completion 쌍이 완전할 때만 free입니다. 기존 문자열/숫자 0·OpenRouter 소수 문자열·input/output 별칭·metadata.pricing 경로는 유지됩니다. tests/codex-integration/catalog-free-pricing-status.test.ts에 보조 요금 paid, 전 차원 0 free, 객체/음수 보조 → unknown 케이스가 추가됩니다. review-ready이고 checklist도 채워져 있습니다. types/config 스플릿과 무관하고, 중복 PR은 없습니다. Preview deploy는 계획에 없습니다.

점수는 높게 잡습니다. tip 버그가 코드로 확인되고, fail-closed 방향·회귀 테스트가 한 축입니다. 다만 모든 pricing 값을 rate로 해석하므로, 숫자/소수 문자열이 아닌 형제 키(예: 통화 표기 문자열)가 섞이면 예전엔 free/paid였던 행이 unknown으로 떨어질 수 있습니다. 의도된 보수성이지만, 실제공자 payload에 비요금 키가 있는지는 랜딩 후 Free 필터 표본으로 한 번 보면 좋습니다.

src/codex/catalog/model-hints.ts (discoveredPricingStatus) - tip은 prompt/completion만 판단. PR은 양수 성분 우선 paid → 미지원 성분 unknown → 완전 0쌍만 free. 주석도 계약에 맞게 고쳐졌습니다.
tests/codex-integration/catalog-free-pricing-status.test.ts - request 0.05 → paid, 전 차원 0 → free, request 객체/image -1 → unknown. 기존 OpenRouter free 문자열 케이스와 충돌 없습니다.
경로/심볼 discoveredPricingRate - 음수·비숫자·빈 문자열은 계속 undefined. 보조 키에도 같은 파서를 쓰는 선택이 일관됩니다.

메인테이너의 판단이 필요한 지점

• pricing 객체에 rate가 아닌 형제 키가 실측으로 있으면 allowlist(알려진 요금 키만 스캔)로 좁힐지, 지금처럼 전체 fail-closed를 유지할지.
• tip rebase·fork CI(macos [Bug]: macOS control is cancelled near its 30-minute limit in full dispatch CI #4905)만 보고 바로 랜딩할지.

너의 추천

머지하세요. Free 필터 금액 사고 방지라 가치가 분명하고 변경 표면이 작습니다. tip에 올린 뒤 카탈로그 Free 표본만 한번 훑으면 됩니다. 라벨은 바꾸지 마세요.

이 댓글은 grok-bot이 작성했습니다

[lidge-jun]

Merging with one macOS red that is a known dev regression rather than anything in this change.

The only failure at this head is Codex history injection preflight on a WAL store with no live writer > reaches a verdict on a cleanly-closed WAL store rather than the catch-all refusal, an 8ms assertion on macos 1/2. It comes from #4957, which landed with full Linux and all-nine-Windows-shard evidence and a macOS gap in its one platform-dependent case. A dedicated lane owns the fix. Everything else at this head is green.

That failure is deterministic and fast, so it is distinguishable from the timeout and cancellation pattern tracked in #4956; I am treating fast assertions as real results and timeouts as the known defect, rather than waving both through.