Document upstream review-only monthly dual-API audit

README.md

@@ -63,9 +63,25 @@ Use the platform repositories for broker credentials, dry-run/live switches, ord
 
 Use this README as a map of the project, not as live performance data. Before enabling or changing a live profile, rerun the relevant live-pool/release pipeline and review short, medium, and long windows: return, max drawdown, benchmark-relative return, turnover, data freshness, and artifact version. Monthly live-pool selection, ranking, and promotion evidence belong in CryptoLivePoolPipelines; strategy changes here should preserve that upstream authority. If evidence is stale, incomplete, or the profile is marked research-only, keep it out of live runtime settings.
 
-## Monthly report audit status
-
-The monthly report workflow in CryptoLivePoolPipelines generates review materials and a tracking issue only. It does not automatically dispatch AIAuditBridge, run a dual-AI audit, or apply AI fixes. A green report job or an open issue proves material preparation, not completed AI review. Any proposed fix must go through PR + CI.
+## Monthly report review
+
+The monthly report is published by
+[CryptoLivePoolPipelines](https://github.com/QuantStrategyLab/CryptoLivePoolPipelines).
+Its review-only audit reads that run's published report and upstream selector/shadow
+metrics, makes one single-turn OpenAI request and one single-turn Anthropic request,
+and posts results to the existing upstream `monthly-review` issue. CryptoStrategies
+does not run a second monthly audit or receive automated code changes. Upstream
+metrics do not establish downstream strategy performance or authorize live promotion.
+
+Defaults are 16,000 input tokens and 2,500 output tokens per request, a 60-second
+timeout, zero retries, and a $1 UTC-month budget reserved in issue comments before
+calls. Missing existing API secrets skips the API job; missing evidence, exhausted
+budget, or API failure produces `review unavailable`. See the
+[upstream runbook](https://github.com/QuantStrategyLab/CryptoLivePoolPipelines/blob/main/docs/operator_runbook.md#monthly-review-only-api-audit)
+for caps, budget accounting, and the human review path. This flow never runs
+Codex/Cursor CLI, edits repositories, or changes trading/live configuration.
+A green publisher or an open issue proves material preparation; completed review
+requires both provider results in the review comment. Proposed fixes require PR + CI.
 
 ## Repository layout
 

README.zh-CN.md

@@ -63,9 +63,13 @@ CryptoStrategies 是 QuantStrategyLab 的加密货币策略包。为 Binance 执
 
 README 只作为项目地图，不替代最新表现数据。启用或调整 live profile 前，需要重新运行相关 live-pool/release pipeline，并分别看短、中、长周期的收益、最大回撤、相对基准收益、换手、数据新鲜度和 artifact 版本。月度 live-pool 选择、ranking 和 promotion 证据属于 CryptoLivePoolPipelines；本仓库的策略改动应保留这个上游权威边界。证据过期、不完整，或者 profile 仍标记为 research-only，就不要放进 live runtime settings。
 
-## 月报审计状态
+## 月度报告审查
 
-CryptoLivePoolPipelines 的月报 workflow 目前只生成审查材料和跟踪 Issue，不会自动 dispatch AIAuditBridge、执行双 AI 审计或应用 AI 修复。月报任务通过或 Issue 已创建，只能证明材料已准备，不能证明 AI 审查完成。修复建议必须经过 PR 和 CI。
+月度报告由 [CryptoLivePoolPipelines](https://github.com/QuantStrategyLab/CryptoLivePoolPipelines) 发布。它的 review-only 作业读取同一次运行的已发布月报及上游 selector/shadow 指标，各发起一次 OpenAI 和 Anthropic API 请求，单轮结果写入上游既有 `monthly-review` issue 的评论。CryptoStrategies 不重复运行月审，也不接受自动代码修改；上游指标不能证明下游策略收益或授权 live 提升。
+
+默认每次输入上限 16,000 token、输出上限 2,500 token、超时 60 秒、零重试，UTC 自然月预算 1 美元，调用前通过 issue 评论预留费用。缺少既有 API Secret 时跳过 API 作业；材料缺失、预算不足或 API 失败时发布 `review unavailable`。配置、预算记录和人工审查入口见 [上游运行手册](https://github.com/QuantStrategyLab/CryptoLivePoolPipelines/blob/main/docs/operator_runbook.md#monthly-review-only-api-audit)。该流程不运行 Codex/Cursor CLI，不修改仓库或交易配置。
+
+月报发布任务通过或 issue 已创建，只能证明材料已准备；审查完成需要评论中两家 API 的结果。修复建议必须经过 PR 和 CI。
 
 ## 仓库结构