[HaoooLee]

Which issue or RFC does this PR close?

Refs #2088.

Rationale for this change

_has_resolved_failure_evidence loads each Artifact citation individually.
Each get() delegates to a single-reference get_many(), repeating revision
and lineage queries for every citation.

Batching these reads reduces the SQL statements issued by a complete
experience.propose operation:

Artifact citations	Before	After
4	13	7
32	69	7

Counts were measured at the SQLAlchemy engine level on in-memory SQLite.
Each measurement was repeated three times with identical results.

What changes are included in this PR?

Use the existing get_many() operation to read Artifact citations for each
matching failure item. If batch validation raises RepositoryError, read
the references in their original order to preserve the original error.

Add a complete-proposal SQL budget test, a focused scenario covering missing
references, historical revisions and early return, and three error-order
regressions. Reuse shared Task Outcome fixtures and correct the cue-matching
test with valid positive and negative controls.

Are there any user-facing changes?

Failure-evidence validation uses fewer database statements while preserving
proposal acceptance, error handling and stored candidate results.

How was this change tested?

• uv run --locked prek run -a — all hooks passed, including Ruff and ty.
• uv run --locked pytest tests/builtin/review/ -q — 33 passed.
• Existing Source eligibility and failure-evidence rule tests — 10 passed.
• git diff --check — passed.

The SQL budget test fails on the original implementation and passes with
this change, requiring equal counts for 4 and 32 citations within an
8-statement ceiling.

The three error-order tests reject a batch-only implementation and pass
with ordered fallback. The corrected cue-matching test also catches removal
of the matching check.

AI usage statement

OpenAI Codex assisted with implementation, test design and review.

[HaoooLee]

I’m interested in PowerContext and would like to contribute more. Could you please review this PR when you have a chance? I’d appreciate any feedback or suggestions.

[hireview001]

• CI failed: Upgrade release gate (failure).
• CI failed: tests (3.13, unit) (failure).
• CI failed: tests (3.14, unit) (failure).
• CI failed: tests (3.13, e2e) (failure).
• CI failed: tests (3.12, e2e) (failure).
• CI failed: tests (3.12, unit) (failure).
• CI failed: tests (3.11, unit) (failure).
• CI failed: tests (3.14, e2e) (failure).
• CI failed: tests (3.11, e2e) (failure).
• CI failed: upgrade-acceptance / Prepare upgrade artifacts and matrix (failure).
• Additional CI issues: PR checks.

[AlexStocks]

结论：无新增 Finding；11 条红与本 PR 变更面无关

head 57fc4d25（2 文件：src/powercontext/builtin/review/service.py + tests/builtin/review/test_service.py）。

行为验证

• pytest tests/builtin/review/test_service.py = 23 passed。
• 变异反证：把 get_many + 逐条回退整体替换成原逐条 artifacts.get 后，同一文件 = 1 failed / 22 passed，失败为 test_failure_proposal_statement_count_does_not_grow_with_artifact_evidence（assert [14, 14, 14] == [70, 70, 70]，即 4 条证据 14 条语句 vs 32 条证据 70 条语句）。该用例真实锁定"语句数不随证据数增长"，_FAILURE_EVIDENCE_STATEMENT_CEILING = 8 在 head 上有效（small == large 且 max(large) <= 8）。
• 回退分支不引入第二次全量读取：ArtifactRepository.get() 的非 for_update 路径本身就是 return (await self.get_many(connection, scope_id, (ref,)))[0]，所以 except RepositoryError 里的逐条调用只用于还原原首个失败的错误优先级，不改变正常路径语句数。

11 条红 job 的归因

与本 PR 变更面无关：quality + tests (3.11|3.12|3.13|3.14, unit) 均为 upgrade_release_gate.GateError: Persistence changed (file set or content; CRLF/LF normalized)（test_repository_baseline_covers_all_three_databases），tests (*, e2e) 均为 test_owner_recovery_http.py::test_http_creation_and_restart_recover_the_original_owner → MigrationError: incompatible_schema。本 PR 不含 MANAGED_INPUTS 下任何文件；base 927bb6ee 的 managed_digest 实测 f0a1d364… 与 release/upgrade-support.json 声明的 9b67d639… 已不一致，修复来自 master 的 de435aa3/59572c6b（#2212），属基线滞后，rebase 到当前 master 即可。

[hidb4ai]

LGTM