diff --git a/README.md b/README.md
index 6c882ebe..08d60dcc 100644
--- a/README.md
+++ b/README.md
@@ -20,7 +20,7 @@
-Every factor proposed, every strategy mutated, every order routed — logged, versioned, reviewable. The LLM writes the code; the engineering harness signs every decision.
+Every factor proposed, every strategy mutated, every order routed — logged, versioned, reviewable. Every number the agents reason on — sourced, as_of-stamped, freshness-checked. The LLM writes the code; the engineering harness signs every decision.
@@ -30,6 +30,8 @@
Inalpha is a **professional quant agent framework, governed by engineering discipline**. It treats LLM agents not as black-box signal generators, but as code-writing collaborators bounded by hooks, permissions, plan-then-execute approval, and a one-shot signature on every order path.
+**Source-attributed by default.** Below the decision harness sits a data discipline: every bar, every quote, every macro print the agents reason on carries its source, its `as_of` timestamp, and a freshness check. Financial reasoning that quietly ages into stale data is the most common way an agent fails *without anyone noticing* — Inalpha refuses to compile that failure mode.
+
Four capability lines sit on top of that harness:
- **Factor lab** — agents formalize, compute, IC-test, multiple-testing-check, and register factors; every hypothesis is logged with author, timestamp, and the economic-story gate decision.
@@ -124,6 +126,35 @@ Inalpha splits *scheduling* from *compute*. The agent runtime fans out the grid
---
+## Roadmap
+
+Where each capability stands today. Live module inventory and the end-to-end decision sequence diagram live in [`docs/04-current-state.md`](docs/04-current-state.md).
+
+| Status | Capability | Phase | Highlight |
+|---|---|---|---|
+| ✅ Shipped | Plan/Exec audit trail + Hooks + Permission Engine | D-8a | three-step orders · one-shot signing token · 5 lifecycle hook events · allow / ask / deny tri-state |
+| ✅ Shipped | Research → strategy → backtest lineage | D-8c | `deep_dive → compose_strategy → run_backtest` with `research_id` / `backtest_id` threaded through |
+| ✅ Shipped | LLM-authored strategies — E1 MVP | D-9 | three sandbox gates (AST · subprocess · `Strategy` contract) + multi-objective fitness + baseline auto-run |
+| ✅ Shipped | Risk engine at the HTTP boundary | D-9 | declarative `risk_rules.toml` · pre-trade `enforce` · `risk_locks` table with independent commit |
+| ✅ Shipped | Bull / bear researcher debate | D-9 | opposing-stance researchers under `services/research` |
+| ✅ Shipped | Scheduler / cron agent mode | D-9 | `scheduler_jobs` + advisory lock + `/api/scheduler/*` management plane |
+| ✅ Shipped | RiskGuard per-account isolation | D-9.1a | `RiskGuardFactory` removes cross-account state bleed |
+| ⏭️ In Flight | Risk engine — remaining rules | D-9 closing | `cooldown` / `stoploss_guard` / `market_hours` need real `trade_repo` + `market_calendar` |
+| ⏭️ In Flight | `askUserChoice` front-end | D-10 (issue #2) | brings the `ask` permission state back from workaround |
+| ⏭️ In Flight | `permissions.yaml` configuration | D-8b (issue #4) | replaces the hard-coded `defaults.ts` |
+| 🗓️ Planned | Live runner | D-10 (issue #1) | tick-driven `on_bar` writing `paper_positions` / `paper_trades` |
+| 🗓️ Planned | Strategy evolution — E2 | D-11 | multi-generation loop + MAP-Elites + Island Model + `unified-diff` mutations |
+| 🗓️ Planned | Research-hub nested supervisor | D-10+ | 4 analysts + bull/bear/risk debate as a single closed loop |
+| 🗓️ Planned | Factor discovery — L0 → L1 | D-11+ | walk-forward IC + multiple-testing correction + `factor_candidates` table |
+| 🔬 Exploring | Skills as procedural memory | TBD | reusable markdown skills with auto-discovery |
+| 🔬 Exploring | Alpha Zoo cold start | E1+ | seed factor library with public alphas (Qlib / Kakushadze / GTJA) |
+| 🔬 Exploring | E4 `evolve_strategy` MCP tool | E4 | evolution loop exposed to the orchestrator as one MCP tool |
+| 🔬 Exploring | Analog backtesting | TBD | similarity-window-driven backtest range selection (STUMPY) |
+
+> **Legend** — ✅ Shipped: behavior already lives in `main` · ⏭️ In Flight: actively in this phase · 🗓️ Planned: scoped for an upcoming phase, not started · 🔬 Exploring: research recorded, no commit date.
+
+---
+
## Built on the shoulders of
Inalpha is not invented from scratch. It selectively inherits proven designs from prior work, with explicit boundaries around **what we take and what we leave**:
diff --git a/README.zh-CN.md b/README.zh-CN.md
index 74effc9e..f1305571 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -20,7 +20,7 @@
-每个因子提案、每次策略变异、每笔订单路由——都有日志、有版本、可复核。LLM 只负责写代码，工程纪律为每个决策背书。
+每个因子提案、每次策略变异、每笔订单路由——都有日志、有版本、可复核。Agent 推理依据的每一个数字——都有来源、有 as_of 时点、有 freshness 校验。LLM 只负责写代码，工程纪律为每个决策背书。
@@ -30,6 +30,8 @@
Inalpha 是一个**用工程纪律驱动的专业量化 agent 框架**。它不把 LLM 当作黑箱信号源，而把它视作受 hooks / permissions / plan-exec / 一次性签名约束的代码协作者——每一步关键动作都留痕、可版本化、可复核。
+**数据默认溯源。** 决策护栏之下还有一道数据纪律：agent 推理依据的每一根 K 线、每一笔报价、每一个宏观数据点，都带来源、带 `as_of` 时间戳、带 freshness 校验。金融推理悄无声息地"用着过期数据继续讲故事"，是 agent 最常见、也最不易被察觉的失败模式——Inalpha 拒绝把这种失败模式编译进系统。
+
在这套护栏之上，铺开四条能力线：
- **因子实验室。** Agent 负责 formalize、compute、IC 检验、多重检验校正、register；每个假设都带作者、时间戳与经济故事门的判定记录。
@@ -124,6 +126,35 @@ Inalpha 把*调度*和*算力*分开：agent runtime 负责扇出网格、聚合
---
+## Roadmap
+
+每条能力的当前进度。已落地模块清单与端到端决策时序图见 [`docs/04-current-state.md`](docs/04-current-state.md)。
+
+| 状态 | 能力 | Phase | 关键点 |
+|---|---|---|---|
+| ✅ 已上线 | Plan/Exec 审计链 + Hooks + Permission Engine | D-8a | 三步下单 · 一次性签名 token · 5 类生命周期 hook · allow / ask / deny 三态 |
+| ✅ 已上线 | 研究 → 策略 → 回测 lineage | D-8c | `deep_dive → compose_strategy → run_backtest` 全链路串 `research_id` / `backtest_id` |
+| ✅ 已上线 | LLM 自创策略 — E1 MVP | D-9 | 三道沙盒（AST 审计 / 子进程 / `Strategy` 协议契约） + 多目标 fitness + baseline 自动并跑 |
+| ✅ 已上线 | 风控引擎落到 HTTP 边界 | D-9 | 声明式 `risk_rules.toml` · 撮合前 `enforce` · `risk_locks` 表（独立 commit） |
+| ✅ 已上线 | Bull / Bear 研究员辩论 | D-9 | `services/research` 立场对抗研究员 |
+| ✅ 已上线 | Scheduler / cron agent 模式 | D-9 | `scheduler_jobs` + advisory lock + `/api/scheduler/*` 管理面 |
+| ✅ 已上线 | RiskGuard 账户级隔离 | D-9.1a | `RiskGuardFactory` 去除跨账户状态串联 |
+| ⏭️ 进行中 | 风控引擎 — 剩余规则激活 | D-9 收尾 | `cooldown` / `stoploss_guard` / `market_hours` 需接 `trade_repo` + `market_calendar` |
+| ⏭️ 进行中 | `askUserChoice` 前端 | D-10（issue #2） | 把 permission 的 `ask` 路径从 workaround 救回 |
+| ⏭️ 进行中 | `permissions.yaml` 配置化 | D-8b（issue #4） | 替代 `defaults.ts` 硬编码 |
+| 🗓️ 已规划 | Live runner | D-10（issue #1） | 行情 tick 驱动 `on_bar`、写 `paper_positions` / `paper_trades` |
+| 🗓️ 已规划 | 策略进化 — E2 | D-11 | 多代演化 + MAP-Elites + Island Model + `unified-diff` 变异 |
+| 🗓️ 已规划 | Research-hub 嵌套 supervisor | D-10+ | 4 analyst + bull/bear/risk debate 闭环 |
+| 🗓️ 已规划 | 因子发现 — L0 → L1 | D-11+ | walk-forward IC + 多重检验校正 + `factor_candidates` 表 |
+| 🔬 探索中 | Skills as procedural memory | 待定 | 可复用 markdown skill + auto-discovery |
+| 🔬 探索中 | Alpha Zoo 冷启动 | E1+ | 公开 alpha 库播种（Qlib / Kakushadze / GTJA） |
+| 🔬 探索中 | E4 `evolve_strategy` MCP tool | E4 | 进化循环以单个 MCP tool 暴露给 orchestrator |
+| 🔬 探索中 | Analog backtesting | 待定 | 历史相似窗口驱动回测区间（STUMPY） |
+
+> **图例** — ✅ 已上线：行为已在 `main` 中 · ⏭️ 进行中：当前 phase 在做 · 🗓️ 已规划：未来 phase 已 scope、未开工 · 🔬 探索中：调研在档、无 commit 日期。
+
+---
+
## Built on the shoulders of
Inalpha 不是从零发明——它有选择地继承前人的最优解，并明确**借鉴边界**：