docs(jev): add Q1-Q4 study-selection brief

[mikamikasuki]

Goal And Delivered Outcome

• Outcome basis / optional anchor: Q1–Q4 decision brief requested by [Task][RFC]: Evaluate optional independent assessment under the RFC decision gates #5213.
• Goal/source and gap: The RFC leaves the caller, finite question, comparator, rubric, and study gates unresolved. This brief recommends deferring M1 and retaining D7 as the sole next owner decision.
• Observable before → after, with the validation row that proves it: Q1–Q4 previously had no bounded contributor recommendation in the package decision records; the paired English and Chinese records now state a proposed D7 question, fair comparators, independent rubric, evidence limits, zero-spend data boundary, and stop rule. See static and manual validation below.
• Issue/task and intended base: Related to [Task][RFC]: Evaluate optional independent assessment under the RFC decision gates #5213; base main at 42e55a8.

Author Declaration

• Written by: model_agent; GPT-6 Luna; OpenAI

Implemented against

• Specification and revision: [Task][RFC]: Evaluate optional independent assessment under the RFC decision gates #5213 acceptance and the integrated RFC docs/architecture/rfcs/optional-semantic-assistance-jev-v0.md at 42e55a8.

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Resolve Q1–Q4 for one evidenced caller and finite question	implemented as a contributor recommendation; owner decision remains pending	packages/loopx-jev/DESIGN_DECISIONS.md; Chinese mirror	Source audit against RFC §§2–3.3, 12 and current contracts
Freeze rubric/error reporting and retain D7/D8 eligibility and authority boundaries	implemented in the brief; no runtime behavior changed	Both decision records	git diff --check; loopx check public-boundary scan
Keep study execution and adoption behind separate approval gates	implemented as defer recommendation and explicit stop rule	Both decision records	Manual source review; no study or provider calls made

• Self-check before submission: Reviewed the current RFC, the start and action-selection contracts, issue acceptance, and the paired decision records. Confirmed both language versions carry the same recommendation and gates. No new study, provider call, private data access, cost claim, or runtime change is included.

Scope And Continuation

• Completed scope and remaining work: Provides the requested source-grounded Q1–Q4 decision proposal. It recommends defer because there is no independently labeled D7 outcome set or authorized data/model/budget plan.
• Slice boundary / successor: The responsible RFC owners decide whether to accept the question and gates. Any M1 study requires their separate data, destination, model, cap, and retention decisions.

Validation

• Tested revision: 26a786b (based on 42e55a8)
• Run state: finished
• Input classes: public source, synthetic evidence only

Check kind	Result	Public-safe evidence / limitation
static	passed	git diff --check; no whitespace errors.
static	passed	mise exec node@22.22.3 -- uv run --extra test loopx check --scan-path packages/loopx-jev/DESIGN_DECISIONS.md --scan-path packages/loopx-jev/DESIGN_DECISIONS.zh-CN.md; public-boundary scan clean for both files. Two warnings report the absent local registry, not the changed files.
manual	passed	Reviewed source claims and bilingual parity against the cited RFC and current contracts; no new experiment or result asserted.
unit	not_run	Documentation-only change; runtime behavior is unchanged.

• Coverage and gaps: Static boundary scanning covers both changed public documents; source review covers the referenced contracts. No runtime or empirical study validation is claimed.

Frontend / Visual Evidence

• UI impact: none
• Before: N/A
• After: N/A
• States and viewports shown: N/A
• Source data: none
• Attention review: N/A; no interface changed.

Type of Change

• Documentation update

LoopX Area

• Public docs or presentation surface (README, protocols, dashboard)

Technical Direction

• Direction / acceptance reference, when applicable: Architecture and research incubator; [Task][RFC]: Evaluate optional independent assessment under the RFC decision gates #5213 / S11 decision required.

Shared-authority RFC fixture impact

• Production-scale fixture schema: N/A; this PR does not claim shared-authority or migration progress.
• Semantic dimensions changed, or reviewed no-impact rationale: N/A.
• Provider conformance arms run: N/A.
• Read-only legacy/file/PostgreSQL three-arm rehearsal: N/A.

Boundary Checklist

• Neither the diff nor this PR body/comments/attachments disclose private state, credentials, raw traces or verifier output, internal links, or local machine paths.
• I did not duplicate maintainer-owned benchmark work unless a maintainer split out a public issue for it.
• I kept the change scoped to the linked issue/task.
• I completed the visual evidence section for UI changes, or marked UI impact none.
• Every commit includes a DCO Signed-off-by trailer (git commit -s).

[loopx-agent]

Reviewer: model_agent; gpt-6.1-sol; OpenAI; runtime_reported; reasoning_effort=xhigh

Reviewed exact head: 57a9d2f0586a19111095f7db976dd15dc4cf25fb. 当前变更为 2 个文档、+59/-0；已经重新核验本轮 rebase 后的完整 PR 差异。此前 head 的检查不自动转成新 head 的结论。

动机

使用 Jev（提供额外只读判断的模型）的研究负责人，需要先判断它是否值得加入 Agent 选任务流程。此前 Q1–Q4 仍缺少一份可供 owner 决策的具体建议，容易把小型示例或已通过的实现测试误当成投资依据。本 PR 提议只讨论 D7：在同优先级、已有资格的 Todo 中选下一项，并因缺少独立结果集和数据/预算批准而暂缓 M1 研究。可观察改善是读完两份决策记录即可知道比较什么、依据什么、哪些条件尚未满足以及何时停止。这里交付的是决策简报；模型效果、真实效率提升和研究批准仍未成立。

改动思路

以 #5213 当前允许的 source-grounded decision brief 为范围，复用现有 RFC 和选择器；不为了讨论研究先增加生产路径。spec_ref: docs/architecture/rfcs/optional-semantic-assistance-jev-v0.md；spec_revision: 42e55a809eb94f13443d303d76118735d4112182。该规范及相关 start/action-selection 契约在当前 PR base 3596467d6e5b3bbb7fc32603a1841e9170b7c4f6 与上述版本之间未变化。

逐项判断：Q1 已给出 D7-only、defer 的贡献者建议，最终方向由 product/domain owner 决定；Q2 给出有限问题和派发前检查点，有限样本与窗口须由 caller/evaluation owner 在结果前选定；Q3 保留完整 Agent/工具/评审基线，另加现有模型和 Jev 两个只读对照，独立盲评，误选、漏选和总成本分开报告，数值门槛/误差成本仍待 owner 冻结；Q4 明确本简报零新增调用/外部支出，数据、目的地、固定模型、调用/金额上限和留存尚待 data/operations owner；D7 §3.3 保留优先级、资格、依赖、hold、claim/lease 和 quota，不修改规范顺序；D8 §3.4 只保留固定研究组合/预算边界，未来另案讨论。上述未决项是研究入场条件，不能将建议写成批准。

具体改动

英文 packages/loopx-jev/DESIGN_DECISIONS.md 与中文镜像各加入同一 Q1–Q4 表、停止规则与证据边界。核对 build_goal_start_contract 的 planner→写入顺序 tie-break，以及 apply_action_selection_agent_gate 的非绑定建议/显式选择要求，文档描述与现有 owner 一致。正常路径是 owner 读取简报并决定是否值得补齐研究条件；负向路径是没有有限样本、独立标签或数据/预算批准时停在 defer，不调用提供方，也不宣称得到实验结果。既有 96 项检查与构造评测保留原修订归属，未被转作 D7 效果证据。

对主干的风险

完整 diff 仅增加讨论文档，没有运行时、状态、默认 skill、调度、quota、网络或 UI 变化。git diff --check 通过；原生 loopx check 对两份文件的公共边界检查通过，0 errors/0 warnings；中英建议、停止规则、authority 与证据限制逐项一致。没有新增实验、模型调用或成本测量，故没有把静态验证包装成真实结果。最大的剩余风险是后来把这份贡献者建议当成 owner 决定；当前文字明确防止这一点。

我的整体评价

无阻断发现，APPROVE。长程价值偏正向：它把下一步缩成一个明确的 owner 判断，减少为未证明的问题提前建设或重复投票；效率收益目前是少走准备弯路的合理预期，尚非测得的执行提升。做更小的纯链接整理不能提供有限问题、比较对象与停止条件；提前实现 ranker 则增加成本且越过当前任务范围。相关整理已检查：既有双语决策记录是合适归属，无需新协议/模块；下一步仍由原 RFC 的 caller/evaluation/data/operations owners 决定 M1，不创建形式化续项，也不关闭上层产品验收。

English verdict: APPROVE — exact head 57a9d2f; a useful bounded Q1–Q4 decision brief with explicit defer, full-workflow comparators and preserved authority. Static/bilingual checks passed; empirical D7 benefit and owner study admission remain unproven.