fix(frontend): honor confidence percentage contract

[seonghobae]

문제

toConfidencePercent가 0..1 값을 비율로 추측해 backend /api/llm/summarize의 confidence: int | None, 0..100 계약과 다른 표시·quality-guardrail telemetry를 만들 수 있었습니다.

Current exact identity

• protected base: develop@042b0c70531b229af3acbd0421a2f23098d848b3
• exact head: 1a7e10ea12f27d5d123d7528eae069d4f43a5194
• compare: ahead_by=6, behind_by=0, merge base = protected develop
• effective delta: 5 files — AGENTS.md는 더 이상 이 product/source PR의 delta가 아닙니다.
• lifecycle: Draft / source + fixture contract repaired / fresh hosted evidence + independent approval pending

RED → causal repair

1. 9d2a6ea892a8367713976e15babcbd337d06895d에서 value-based 0–1 inference를 제거하고 frontend helper/test fixture를 backend 0–100 단위에 맞췄습니다.
2. review에서 pilot/full-product buyer smoke가 여전히 confidence: 0.86을 반환해 1% 표시와 false low_confidence event를 만들 수 있음을 찾았습니다. a081359a2911dcde31965b797ef9c96366770463에서 두 fixture를 86으로 고쳤고 해당 finding은 current source에서 해소됐습니다.
3. 4d6b697e97f7f0d98700292680970b41e7b12c48에서 shared boundary가 0.856, 85.5 같은 non-integer malformed 값까지 silently round하지 않고 reject하도록 고쳤습니다. -0은 0으로 정규화합니다.
4. 같은 lineage에서 confidence recurrence rule을 AGENTS.md에 직접 추가했지만, live Naruon docs stack docs(governance): align AGENTS/CLAUDE LLM guidance with canonical orchestrator owner #1549/#1566이 이미 AGENTS.md canonical writer입니다. 병렬 docs ownership을 유지하지 않고 direct descendant 1a7e10ea12f27d5d123d7528eae069d4f43a5194에서 protected-base AGENTS blob을 non-force로 복원했습니다. 실행 가능한 recurrence contract는 confidence.test.ts, EmailDetail.test.tsx, pilot/full-product smoke에 그대로 남습니다. valid AGENTS finding은 docs(agents): 증거 기반 작업 절차 정리 #1566 PR authority로 승계했으며 그 docs stack exact head에 실제 반영되기 전에는 landed로 간주하지 않습니다.

이 PR은 #1436 전체를 supersede하지 않습니다. confidence-unit boundary만 독립적으로 소유하며 #1436의 Storybook/design-system/product-event delta는 별도입니다.

Fresh evidence boundary

새 exact head에 repository-owned PR workflows가 다시 생성됐습니다: Application CI 33942459386, Security Scan 33942459380, Semgrep 33942459468, CodeQL PR 33942459375, Bandit 33942459369, Docker 33942459471. 현재 모두 queued입니다. earlier-head checks/reviews는 통과 증거로 승계하지 않습니다. qualifying post-last-push independent APPROVED review도 없습니다. resolved historical threads는 current source finding이 다시 나타나지 않는다는 사실만 나타내며 merge authorization이 아닙니다.

Keep Draft until one unchanged exact head has every then-live repository/organization required check terminal-success, zero valid unresolved findings/threads, and the live last-push approval requirement satisfied. No self-approval, bypass/admin merge, force-push, destructive rebase, dummy/no-op requeue commit, synthetic status, gate weakening, or predecessor-evidence transfer.

UI Delivery Gate

• 의도성: PASS — 표시 단위를 backend integer 0–100 contract 하나로 고정합니다.
• 기능 완전성: source/fixture PASS — helper/component/buyer-smoke가 같은 단위를 사용하고 malformed non-integer를 fail closed 합니다.
• 콘텐츠 적합성: PASS — 사용자 카피·화면 구조를 추가하지 않습니다.
• 복원력: PENDING — current-head hosted/browser execution이 terminal-success가 아닙니다.
• 증거성: PENDING — exact-head required checks와 post-last-push independent approval이 필요합니다.
• 고유성: N/A — 시각 정체성 변경이 아닙니다.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Team

Run ID: 1ffce182-bb4c-422c-aa0d-c6a221c092a8

📥 Commits

Reviewing files that changed from the base of the PR and between 042b0c7 and 9d2a6ea.

📒 Files selected for processing (3)

• frontend/src/components/EmailDetail.test.tsx
• frontend/src/lib/confidence.test.ts
• frontend/src/lib/confidence.ts

Included review availability: Your plan provides up to 1 included review per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The confidence formatter now treats finite inputs as literal percentages, clamps them to 0–100, and rounds them. Unit tests and EmailDetail mocks now use percentage-based confidence values.

Changes

Confidence Percentage Contract

Layer / File(s)	Summary
Update confidence percentage formatting frontend/src/lib/confidence.ts, frontend/src/lib/confidence.test.ts	toConfidencePercent now clamps and rounds raw percentage values without multiplying fractions by 100. Tests cover the updated contract and rounding behavior.
Align EmailDetail test fixtures frontend/src/components/EmailDetail.test.tsx	Summarize mocks now use integer confidence percentages that match the expected UI labels.

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

Merge Risk: ⚪ Minimal · up to 9d2a6

Confidence values are now displayed as backend-provided percentages rather than inferred ratios, preventing fractional values from being shown as incorrect percentages. The updated tests cover the formatter behavior and EmailDetail mock responses, with no remaining merge-readiness risk identified.

🚥 Pre-merge checks | ✅ 4

✅ 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 the main change: updating frontend confidence handling to follow the backend percentage contract.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch codex/confidence-ratio-contract

---

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.}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-04T12:58:52.951002Z	d75cfe7	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9d2a6ea892

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: a081359a29

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[seonghobae]

Exact-head repair evidence for 9c90dea4832275aa5e127de5def1471f31c44ce9:

• Root cause: unit inference was removed, but the display helper still clamped malformed values outside the backend 0–100 contract; 150 could become a false 100% confidence claim.
• Repair: reject non-finite and out-of-range values at the shared display boundary while preserving percentage rounding and positive-zero normalization.
• Focused evidence: 30 Vitest tests, TypeScript --noEmit, ESLint, and git diff --check pass. Hosted exact-head Checks and independent re-review remain separate gates.

[seonghobae]

Current-head boundary audit found that the documented integer 0–100 contract was still not enforced: non-integer payloads were silently rounded (0.856→1%, 85.5→86%). Fixed once in the shared toConfidencePercent boundary at 4d6b697e97f7f0d98700292680970b41e7b12c48; it now rejects non-finite, non-integer, negative, and >100 values while preserving -0 normalization. Focused Vitest: 28 passed; TypeScript, ESLint, and diff checks pass. Exact-head hosted state is 21 queued / 5 skipped / 0 failures, so the PR remains Draft pending independent review and terminal required checks.