Phase2 dry-run: pending A, probe, build-config client, gated task/add

Still default dry-run; ALLOW_LIVE_SUBMIT=0. Live gates closed.
2026-10-16 19:00:15 +08:00
parent 0d0f3b611a
commit bed3bc11ca
25 changed files with 952 additions and 110 deletions
--- a/18
+++ b/18
@@ -2,18 +2,21 @@
 # Allowed: test / build / smoke-health / dry-run
 # FORBIDDEN: register / submit / deploy / push targets

-.PHONY: test build smoke-health dry-run help
+.PHONY: build-config-dry probe-readonly test build smoke-health dry-run dry-run-pending help

 PYTHON ?= python3
 IMAGE  ?= modelhub-agent-contrib:local
 PORT   ?= 8080

 help:
+	@echo "build-config-dry generate YAML via API (requires token; never creates task)"
+	@echo "probe-readonly authenticated read-only task/page GET"
+	@echo "dry-run-pending read-only adapt pending A→B report"
 	@echo "test         stdlib unit tests + local SIGTERM integration (needs sockets)"
 	@echo "build        Docker build using platform base (needs Docker + TCR access)"
 	@echo "smoke-health local HTTP smoke (needs sockets; default test STRATEGY_ID)"
-	@echo "dry-run      local example report; pending-submit integration outstanding"
-	@echo "No register/submit/deploy/push — Phase1 hard boundary."
+	@echo "dry-run      read-only pending report (example fallback)"
+	@echo "No register/submit/deploy/push — default live off."

 test:
 	$(PYTHON) -m unittest discover -s tests -v
@@ -30,3 +33,12 @@ smoke-health:

 dry-run:
 	$(PYTHON) scripts/dry_run.py
+
+dry-run-pending:
+	PENDING_SUBMIT_MD=/workspace/modelhub-adapt/pending-submit.md PENDING_SUBMIT_TXT=/workspace/modelhub-adapt/pending-submit.txt $(PYTHON) scripts/dry_run.py
+
+probe-readonly:
+	$(PYTHON) scripts/probe_readonly.py
+
+build-config-dry:
+	$(PYTHON) scripts/build_config_dry.py
--- a/README.md
+++ b/README.md
@@ -72,3 +72,5 @@ tests/      # stdlib unittest
 ## 许可与边界

 本仓库用于 Victor 本地/Gitea 交付骨架；任何 ModelHub 写操作须 Victor 本人在控制台或显式批准后执行。
+
+Phase2：`make dry-run-pending` 只读生成 A→B 草稿，名称硬门与禁提过滤已接入；需 Victor 人工提交/确认。task/add 客户端默认不调用，必须显式注入并通过双闸门、队列及单次预算；测试仅 mock HTTP。
--- a/docs/api-probe-design.md
+++ b/docs/api-probe-design.md
@@ -1,7 +1,7 @@
 # API 探针设计（只读意图 · 禁止代调写接口）

 - 日期：2026-10-16 Asia/Shanghai
- Phase1：**仅设计文档**；本仓不实现真实 HTTP 写客户端
+- Phase2：客户端已实现但默认不调用；本轮禁止真实 task/add

 ## 管理 API `/agent_platform/strategies*`

@@ -22,15 +22,20 @@
 |------|------|----------|
 | GET | `/api/adapt/task/page` | 未来只读鉴权探针（对齐 huni-probe，`ALLOW_SUBMIT=0`） |
 | POST | `/api/adapt/task/build-config` | 设计可引用；Phase1 不实现 |
-| POST | `/api/adapt/task/add` | **禁止代调**；live 仅闸门 + 单测假客户端 |
+| POST | `/api/adapt/task/add` | 显式注入客户端 + 双闸门；本轮只做 mock 测试 |
 | GET | `/api/adapt/task/calculate-credit` | 只读索引；勿写死 KPI |

-容器内社区鉴权（C）：`Xc-Token` + body `strategyId` — **待 Victor 账号核实**后再实现探针。
+容器内社区鉴权（C）：`Xc-Token` + body `strategyId` — **待 Victor 账号核实**；60014 仍为 **D（unverified）**。

-## Phase1 落地边界
+## Phase2 落地边界

- `src/gate.py`：闸门逻辑 + `attempt_live_submit(..., client=Fake)` 仅供单测
- **无**默认 `urllib`/`requests` 客户端指向 `task/add` 或 `strategies` POST
+- `src/gate.py`：闸门逻辑 + `attempt_live_submit(..., client=Fake)` ；真实客户端只能显式注入，测试必须 mock HTTP
+- `src/task_add_client.py` 使用 urllib；import 和 dry-run 无网络调用，无自动重试
+- token 优先级：XC_TOKEN_FILE → XC_TOKEN → MODELHUB_XC_TOKEN → EXTERNAL_SERVICE_TOKEN；禁止输出 token
+- build_payload 中 GPU/framework/strategyId/configParams 占位须人工替换；真实调用要求 STRATEGY_ID 与 payload 一致
+- 队列 ≥95 阻断；无效队列提示也阻断；每进程一次调用预算在调用前保留，异常也不重试（防止响应丢失造成重复创建）
+- 名称/族硬门不能证明元数据 PASS；未跳过的 pending 行均标 needs_human
+- 仍禁止所有 agent_platform 写接口
 - Makefile **无** register / submit / deploy / push

 ## 冲突裁决提醒
--- a/docs/autoconfig-probe-plan.md
+++ b/docs/autoconfig-probe-plan.md
@@ -0,0 +1,25 @@
+# configParams 配置来源与容器验证
+
+日期：2026-10-16 Asia/Shanghai。
+
+结论：**API**。配置来自服务端，前端将响应 `data` 写入 `configParams`。
+依据：[autoconfig-capture-2026-10-16.md](/workspace/modelhub-adapt/logs/autoconfig-capture-2026-10-16.md)。原「公开面无 build-config」缺口关闭：**未公开文档但已抓包证实**。
+
+## 已证实契约
+
+- `POST /api/adapt/task/build-config`
+- Query：`gpuType=Sunrise_pt-200-x1&framework=vllm&taskType=text-generation`
+- 空 body，无 JSON；模型地址不进入 query/body，也无需传 strategyId。
+- HTTP 2xx，JSON `code == 0`，非空字符串 `data` 为 YAML。
+- 当前 UI 无「自动配置」标签。新建适配弹层选择模型、GPU、文本生成后，选择框架触发 effect 并填入配置。
+- 抓包观察到 `Authorization` 头，值未记录。容器默认按社区实践发送 `Xc-Token`；`BUILD_CONFIG_USE_AUTHORIZATION=1` 改用同一 resolve_token 值作为 Authorization，不自动加 Bearer。
+
+## 客户端与验证
+
+`src/build_config_client.py` 的 `build_config()` 成功返回 YAML，失败抛出 `BuildConfigError`，包含安全归因、HTTP 状态和平台数字码；不返回服务端消息或传输异常细节。缺失/无效凭证不联网，40100、40101、其他平台码、传输失败和无效 data 均可区分。导入无 I/O。
+
+默认 dry-run 使用 `templates/config_params.sample.yaml`，无网络。`CONFIG_PARAMS_SOURCE=build-config` 强制 API；或 `ALLOW_BUILD_CONFIG=1` 且可解析 token 时使用 API。调用失败直接报错，不回退 sample。调用方须传真实 GPU enum。`make build-config-dry` 仅生成配置，缺 token 非零退出，不输出 YAML 或凭证。
+
+## 剩余 GAP 与边界
+
+尚无容器 token 鉴权实证；需运行时 token + API 调用并与只读探针对照，不能宣称鉴权就绪。sample 本身不是 production-ready。队列 ≥95 禁 live。本轮仅 mock 测试，不进行真实 API 调用，不调用 task/add，不保存秘密，不提交任务或推送 tag。
--- a/docs/gpu-perf-order.md
+++ b/docs/gpu-perf-order.md
@@ -0,0 +1,49 @@
+# GPU 缺卡序（契约）
+
+- 日期：2026-10-16 Asia/Shanghai
+- 对齐：模盒产品经理 + 模盒积分
+- 正式性能表：**无** → 报告字段 `perf_order=unset`（勿写成官方 TPS 排名）
+- 缺卡补齐所用序：**经验序 / 待平台核实**（下表）— 产品要求排序器按此序；非站内正式性能榜
+
+## A. 正式性能序
+
+| 优先级 | 展示名 | targetGpu | 依据 |
+|--------|--------|-----------|------|
+| — | — | — | 模盒积分定稿：无正式卡间性能表 → `perf_order=unset` |
+
+## B. 缺卡经验序（排序器用 · 待平台核实）
+
+跨卡总策略优先于本序：已验证/已下载落点 → 只打**未适配卡**；跳过正在验证 / 已验证成功 / 唯一性失败 / 同卡已知失败禁重试；队列 ≥95 停交（live）。
+
+| 优先级（高→低） | 平台展示名 | targetGpu（已观察，待再核） | 备注 |
+|-----------------|------------|------------------------------|------|
+| 1（一梯队并列） | 壁砺166M | `Biren_166m` | ≈ 曦望 S2；社区成功量壁砺更高 |
+| 1（一梯队并列） | 曦望 S2 | `Sunrise_pt-200-x1`（抓包证实） | 与壁砺并列，勿臆造先后「性能」 |
+| 2 | 天垓150 | Iluvatar 天垓族（字符串待核） | 条件回退；饱和+唯一性碰撞多 |
+| 3 | 天垓100 | （待核） | 一般不扇出 |
+| 4 | 智铠 mrv-100 | `Iluvatar_mrv-100` | 我方唯一成功样本族；不当最快卡宣传 |
+
+并列一梯队时：若 pending `try-next` 已给出相对顺序则保留；否则可按「壁砺 / 曦望」与池内缺口并列候选，报告标 `order_source=experience`。
+
+## C. 映射备忘（非排名）
+
+| 展示名 | 提交字段 | 备注 |
+|--------|----------|------|
+| 曦望 S2 | `Sunrise_pt-200-x1` | 2026-10-16 build-config query 抓包证实；非性能排名 |
+| （docs sample） | `MetaX_c-500` | 与中文悬赏卡无已核实映射 |
+
+## D. 依据文件（工作流/观察，非官方性能榜）
+
+- `/workspace/modelhub-adapt/pending-submit.md`
+- `/workspace/modelhub-adapt/submit-plan-2026-10-15-v2.md`
+- `/workspace/modelhub-adapt/uniqueness-collision-notes-2026-10-15-eve.md`
+- `/workspace/modelhub-adapt/gpu-skip-patterns-xiwang-vs-bili-2026-10-15-eve2.md`
+
+## E. configParams 缺口（产品补充）
+
+- 结论：**API**，未公开文档但已抓包证实，依据 [autoconfig-capture-2026-10-16.md](/workspace/modelhub-adapt/logs/autoconfig-capture-2026-10-16.md)。
+- `POST /api/adapt/task/build-config`，query 为 `gpuType` / `framework` / `taskType`，空 body；成功 JSON `code=0`，`data` 为 YAML。
+- UI 无「自动配置」标签；选择框架触发 effect 填入 configParams。模型地址不进入该请求。
+- 抓包观察 Authorization（值未记录）；容器默认 Xc-Token 社区实践，可显式切换 Authorization，尚无 token 鉴权实证。
+- 默认 dry-run 离线 sample；CONFIG_PARAMS_SOURCE=build-config 或 ALLOW_BUILD_CONFIG=1 且 token 可解析时调用 API，失败不可静默回退。
+- **live 前必须有 token + API 调用实证**；sample 单独不是 production-ready。队列 ≥95 继续禁 live，本轮无 live。
--- a/docs/issues.md
+++ b/docs/issues.md
@@ -8,9 +8,9 @@
 |-------------------|-----|--------------|
 | Phase 0 调研冻结 | M0 | 文档已落 `docs/` |
 | Phase 1 本地可部署样例 | M1 | **本轮交付** |
-| Phase 2 Dry-run 闭环 | M2 | 骨架：`scripts/dry_run.py`；**待对接 pending-submit**：A→B / PASS / 硬门尚未自动核验 |
+| Phase 2 Dry-run 闭环 | M2 | 已实现 pending A→B 解析、名称硬门/禁提过滤与 dry-run；元数据 PASS 仍需人工核实 |
 | Phase 3 人工注册 + 只读探针 | M3 | 仅 runbook / api-probe-design；**不代注册** |
-| Phase 4 一次 live 验证 | M4 | 闸门已备；**无真实客户端** |
+| Phase 4 一次 live 验证 | M4 | 闸门已备；客户端已实现；默认关闭，必须显式注入 |
 | Phase 5 规模化 | — | 未开 |

 ## P0 Issues
@@ -28,9 +28,9 @@

 Phase1 构建与运行验收须独立记录；环境阻塞不等于通过。

-## P1（未做）
+## P1

- [ ] 对接 `/workspace/modelhub-adapt/pending-submit.md` A→B 只读
+- [x] 对接 `/workspace/modelhub-adapt/pending-submit.md` A→B 只读
 - [ ] 结构化 JSON 日志（禁密钥）
 - [ ] evidence-agent 模板填写

@@ -43,5 +43,5 @@ Phase1 构建与运行验收须独立记录；环境阻塞不等于通过。

 - 无 register / submit / deploy / push Makefile 目标
 - 无 `/agent_platform/strategies*` 写调用
- 无 `/api/adapt/task/add` 真客户端
+- `/api/adapt/task/add` 客户端默认不调用；双闸门、队列≥95 阻断、每进程最多一次；本轮未授权真网
 - 适配 W1 优先；不抢队列
--- a/docs/tag-dryrun-notes-draft.md
+++ b/docs/tag-dryrun-notes-draft.md
@@ -0,0 +1,13 @@
+# Tag 文案草稿（仍默认 dry-run）— 未授权勿 push
+
+- 建议 Tag 名（待 Victor 确认）：`v0.2.0-phase2-dryrun` 或 `v0.2.0-submit-ready-dryrun`
+- 相对 Phase1 `v0.1.0-phase1@0d0f3b6`：pending A 池 + 经验序排序、只读 task/page 探针、task/add 客户端仍双重闸门默认关、configParams sample 仅 dry-run
+- **不含**：live 开闸、真实自动配置、开闸环境变量默认开启
+- 推 Tag 条件：仅当 Victor **明确**说「先上线 dry-run 客户端」；产品当前：**先别 push Tag**
+- 队列满 / configParams 缺口期间：**不开闸**
+
+推送检查清单（人工）：
+1. `make test` 绿
+2. ALLOW_LIVE_SUBMIT 默认非 1；LIVE_CONFIRM 未设
+3. 控制台重建后日志仍为 dry-run / Phase2 格式
+4. 不代注册、不代 live
--- a/scripts/build_config_dry.py
+++ b/scripts/build_config_dry.py
@@ -0,0 +1,26 @@
+#!/usr/bin/env python3
+"""Generate config only; never creates a task or prints returned YAML/secrets."""
+import argparse
+import sys
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+from src.build_config_client import build_config, BuildConfigError
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--gpu-type', default='Sunrise_pt-200-x1')
+    parser.add_argument('--framework', default='vllm')
+    parser.add_argument('--task-type', default='text-generation')
+    args = parser.parse_args()
+    try:
+        yaml = build_config(args.gpu_type, args.framework, args.task_type)
+    except BuildConfigError as error:
+        print(str(error), file=sys.stderr)
+        return 1
+    print(f'configParams_source=build-config; generated YAML characters={len(yaml)}; no task submitted')
+    return 0
+
+
+if __name__ == '__main__':
+    sys.exit(main())
--- a/scripts/dry_run.py
+++ b/scripts/dry_run.py
@@ -1,12 +1,5 @@
 #!/usr/bin/env python3
-"""Generate a dry-run adaptation report (no platform side effects).
-
-Reads config/candidates.example.json (or DRY_RUN_CANDIDATES path).
-Writes reports/dry-run-<Asia/Shanghai-timestamp>.md
-"""
-
-from __future__ import annotations
-
+"""Read-only Phase2 pending draft report. Offline by default; opt-in configuration generation only."""
 import json
 import os
 import sys
@@ -15,93 +8,78 @@ from pathlib import Path
 from zoneinfo import ZoneInfo

 ROOT = Path(__file__).resolve().parents[1]
-TZ = ZoneInfo("Asia/Shanghai")
+sys.path.insert(0, str(ROOT))
+from src.pending_pool import build_ordered_candidates, input_path, experience_order
+from src.config_params import build_config_params, config_params_source, SOURCE, GAP
+from src.queue_depth import queue_depth
+from src.task_add_client import build_payload
+
+TZ = ZoneInfo('Asia/Shanghai')


-def _load_candidates(path: Path) -> list[dict]:
-    data = json.loads(path.read_text(encoding="utf-8"))
-    if isinstance(data, list):
-        return data
-    if isinstance(data, dict) and "candidates" in data:
-        return list(data["candidates"])
-    raise SystemExit(f"unexpected candidates schema in {path}")
+def _load_candidates(path):
+    data = json.loads(path.read_text())
+    return data if isinstance(data, list) else list(data['candidates'])


-def _render(candidates: list[dict], ts: str, src: Path) -> str:
-    lines = [
-        f"# Dry-run 适配建议报告 — {ts} (Asia/Shanghai)",
-        "",
-        "> **模式：dry-run** — 本报告不产生平台侧真实提交。",
-        "> **需 Victor 人工提交**：请登录模盒控制台按草稿确认后手工提交。",
-        "> **积分：待平台核实**（发布智能体本身不得分；成功适配按账本计）。",
-        "",
-        "## 闸门状态",
-        "",
-        f"- ALLOW_LIVE_SUBMIT is 1: {os.environ.get('ALLOW_LIVE_SUBMIT', '0').strip() == '1'}",
-        f"- LIVE_CONFIRM set: {'yes' if os.environ.get('LIVE_CONFIRM', '').strip() else 'no'}",
-        f"- STRATEGY_ID present: {'yes' if os.environ.get('STRATEGY_ID', '').strip() else 'no'}",
-        "",
-        "## 输入",
-        "",
-        f"- 候选文件：`{src}`",
-        f"- 候选条数：{len(candidates)}",
-        "",
-        "## 建议草稿（字段示意，非真实 POST）",
-        "",
-    ]
+def _render(candidates, ts, src):
+    candidates = [c for c in candidates if not os.environ.get('PENDING_SECTION') or
+                  c.get('section', 'B') == os.environ['PENDING_SECTION'].strip().upper()]
+    depth = queue_depth()
+    lines = [f'# 代提候选 / 需 Victor 确认 — Dry-run — {ts} (Asia/Shanghai)', '',
+             '> 模式：dry-run；零网络写；需 Victor 人工提交/确认',
+             '> 积分：待平台核实', '> perf_order=unset；order_source=experience（经验序 / 待平台核实）',
+             f'> configParams source: {SOURCE}', f'> configParams_source={config_params_source()}', f'> {GAP}', '',
+             '## A→B 顺序草稿', '', f'输入：{src}；候选条数：{len(candidates)}',
+             f'队列深度：{depth if depth is not None else "未知，需人工确认"}',
+             '队列≥95：禁 live；适配 W1 优先，不抢空位。', '']
+    if depth is not None and depth >= 95:
+        lines.extend(['> **禁 live / 队列≥95**', ''])
    for i, c in enumerate(candidates, 1):
-        model = c.get("modelAddress") or c.get("model_id") or "(missing model_id)"
-        lines.append(f"### {i}. `{model}`")
-        lines.append("")
-        lines.append("```json")
-        draft = {
-            "modelAddress": model,
-            "taskType": c.get("taskType", "text-generation"),
-            "targetGpu": c.get("targetGpu", "(unset)"),
-            "framework": c.get("framework", "(unset)"),
-            "strategyId": "<PLATFORM_INJECTED_STRATEGY_ID>",
-            "configParams": "<from build-config — not fetched in Phase1 dry-run>",
-            "notes": c.get("notes", ""),
-        }
-        lines.append(json.dumps(draft, ensure_ascii=False, indent=2))
-        lines.append("```")
-        lines.append("")
-        lines.append("- 风险：Phase1 未做唯一性/硬门预检；提交前请对照 do-not-resubmit 与队列空位。")
-        lines.append("- **需 Victor 人工提交**（或经双重闸门的受控 live；本骨架默认拒绝 live）。")
-        lines.append("- **积分：待平台核实**")
-        lines.append("")
-    lines.extend(
-        [
-            "## 后续对接（Phase2）",
-            "",
-            "- **待对接 pending-submit**：尚未读取 `/workspace/modelhub-adapt/pending-submit.md`，未验证 A→B、PASS、硬门与 do-not-resubmit；本报告仅保留输入顺序。",
-            "- 适配 W1 优先；同一 100 队列，≥95 全体停交；空位先 A 再 B，人工核验后决定，不凭示例报告提交。",
-            "- 禁止 Makefile / 脚本调用 `/api/adapt/task/add` 或 `/agent_platform/strategies*` 写接口。",
-            "",
-        ]
-    )
-    return "\n".join(lines)
+        model = c.get('model_address') or c.get('modelAddress') or c.get('model_id') or '<MODEL>'
+        gpus = experience_order(c.get('try_next_gpus', []))
+        target = gpus[0] if gpus else c.get('targetGpu', '<TARGET_GPU:需人工确认>')
+        status = c.get('status', 'needs_human')
+        lines.extend([f'### {i}. Section {c.get("section", "B")} — `{model}`', '',
+                      f'状态：{status}；needs_human={c.get("needs_human", True)}',
+                      f'过滤：{c.get("skip_reason") or "名称硬门未命中；元数据/唯一性待人工核实"}', ''])
+        lines.extend([f'未适配卡（仅 pending 缺口）：{", ".join(gpus) or "未提供，需确认"}',
+                      'perf_order=unset；order_source=experience；天垓100 一般不扇出', ''])
+        draft = build_payload(model, target, 'vllm', strategy_id='<STRATEGY_ID>',
+                              config_params=build_config_params(target))
+        # Display only; this truncated draft must never be submitted.
+        draft['configParams'] = draft['configParams'][:400] + '\n… [truncated; dry-run only]'
+        lines.extend(['build-config query: gpuType=<targetGpu>&framework=vllm&taskType=text-generation; POST empty body', 'POST 字段草稿（skipped 行禁止提交）：', '```json', json.dumps(draft, ensure_ascii=False, indent=2), '```',
+                      '需 Victor 人工提交/确认；积分：待平台核实', ''])
+    report = '\n'.join(lines)
+    # Defensive redaction even if an input happens to contain an injected credential.
+    secrets = [os.environ.get(k, '') for k in ('XC_TOKEN', 'MODELHUB_XC_TOKEN', 'EXTERNAL_SERVICE_TOKEN', 'Xc-Token')]
+    if os.environ.get('XC_TOKEN_FILE'):
+        try:
+            secrets.append(Path(os.environ['XC_TOKEN_FILE']).read_text().strip())
+        except OSError:
+            pass
+    for secret in secrets:
+        if secret:
+            report = report.replace(secret, '[REDACTED]')
+    return report


-def main() -> None:
-    src = Path(
-        os.environ.get(
-            "DRY_RUN_CANDIDATES",
-            str(ROOT / "config" / "candidates.example.json"),
-        )
-    )
-    if not src.is_file():
-        raise SystemExit(f"candidates file not found: {src}")
-
-    candidates = _load_candidates(src)
+def main():
+    md = input_path('PENDING_SUBMIT_MD', 'pending-submit.md')
+    txt = input_path('PENDING_SUBMIT_TXT', 'pending-submit.txt')
+    if md.is_file() or txt.is_file():
+        src = md if md.is_file() else txt
+        candidates = build_ordered_candidates(md, txt)
+    else:
+        src = Path(os.environ.get('DRY_RUN_CANDIDATES', ROOT / 'config/candidates.example.json'))
+        candidates = _load_candidates(src)
    now = datetime.now(TZ)
-    ts = now.strftime("%Y%m%d-%H%M%S")
-    out_dir = ROOT / "reports"
-    out_dir.mkdir(parents=True, exist_ok=True)
-    out_path = out_dir / f"dry-run-{ts}.md"
-    out_path.write_text(_render(candidates, now.strftime("%Y-%m-%d %H:%M:%S"), src), encoding="utf-8")
-    print(f"Wrote {out_path}")
+    out = ROOT / 'reports' / f'dry-run-{now:%Y%m%d-%H%M%S}.md'
+    out.parent.mkdir(exist_ok=True)
+    out.write_text(_render(candidates, now.strftime('%Y-%m-%d %H:%M:%S'), src), encoding='utf-8')
+    print(f'Wrote {out}')


-if __name__ == "__main__":
+if __name__ == '__main__':
    main()
--- a/scripts/probe_readonly.py
+++ b/scripts/probe_readonly.py
@@ -0,0 +1,17 @@
+#!/usr/bin/env python3
+"""Run only the task/page GET; print credential-free summary."""
+import json
+import sys
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+from src.readonly_probe import probe_readonly
+
+
+def main():
+    result = probe_readonly()
+    print(json.dumps(result, ensure_ascii=False))
+    return 0 if result['auth_ok'] else 1
+
+
+if __name__ == '__main__':
+    sys.exit(main())
--- a/src/build_config_client.py
+++ b/src/build_config_client.py
@@ -0,0 +1,60 @@
+"""Explicit configuration generation; import performs no I/O.
+
+Returns YAML on success, otherwise raises BuildConfigError with safe attribution
+and HTTP/platform status. Server messages and transport details are never exposed.
+"""
+import json
+import os
+import urllib.error
+import urllib.parse
+import urllib.request
+from src.task_add_client import resolve_token
+
+
+class BuildConfigError(RuntimeError):
+    def __init__(self, reason, http_status=None, platform_code=None):
+        self.error_attribution = reason
+        self.http_status = http_status
+        self.platform_code = platform_code
+        super().__init__('build-config: ' + reason)
+
+
+def build_config(gpu_type, framework='vllm', task_type='text-generation'):
+    try:
+        token = resolve_token()
+    except (ValueError, OSError):
+        raise BuildConfigError('missing/invalid credential') from None
+    if not isinstance(gpu_type, str) or not gpu_type.strip() or '<' in gpu_type:
+        raise BuildConfigError('gpu_type required (platform enum)')
+    query = urllib.parse.urlencode(dict(gpuType=gpu_type, framework=framework, taskType=task_type))
+    header = 'Authorization' if os.environ.get('BUILD_CONFIG_USE_AUTHORIZATION') == '1' else 'Xc-Token'
+    try:
+        request = urllib.request.Request(
+            os.environ.get('MODELHUB_API_BASE', 'https://www.modelhub.org.cn').rstrip('/')
+            + '/api/adapt/task/build-config?' + query,
+            headers={header: token, 'Accept': 'application/json'}, method='POST')
+        with urllib.request.urlopen(request, timeout=30) as response:
+            status, raw = response.status, response.read()
+    except urllib.error.HTTPError as error:
+        status, raw = error.code, error.read()
+    except (OSError, urllib.error.URLError, ValueError):
+        raise BuildConfigError('transport/timeout') from None
+    try:
+        body = json.loads(raw)
+    except (ValueError, UnicodeError):
+        raise BuildConfigError('invalid JSON response', status) from None
+    if not isinstance(body, dict):
+        raise BuildConfigError('invalid response object', status)
+    code = body.get('code')
+    # Only numeric codes are retained; arbitrary server strings may echo secrets.
+    code = code if type(code) is int else None
+    if code in (40100, 40101):
+        raise BuildConfigError({40100: '40100 NOT_LOGIN_ERROR', 40101: '40101 NO_AUTH_ERROR'}[code], status, code)
+    if not 200 <= status < 300 or code != 0:
+        raise BuildConfigError('other HTTP/platform code', status, code)
+    data = body.get('data')
+    if not isinstance(data, str) or not data.strip():
+        raise BuildConfigError('empty/non-str data', status, code)
+    if token in data:
+        raise BuildConfigError('credential echoed in data', status, code)
+    return data
--- a/src/config_params.py
+++ b/src/config_params.py
@@ -0,0 +1,31 @@
+"""Offline sample by default; explicit opt-in enables captured build-config API."""
+import os
+from pathlib import Path
+from src.build_config_client import build_config, BuildConfigError
+from src.task_add_client import resolve_token
+
+SOURCE = 'POST /api/adapt/task/build-config (discovered via capture); default templates/config_params.sample.yaml'
+GAP = 'GAP: build-config API discovered via capture (autoconfig-capture-2026-10-16.md); live/实锤 needs token + API call and auth proof. Sample is dry-run draft only, NOT production-ready.'
+
+
+def config_params_source():
+    source = os.environ.get('CONFIG_PARAMS_SOURCE', '').strip()
+    if source not in ('', 'sample', 'build-config'):
+        raise BuildConfigError('unsupported CONFIG_PARAMS_SOURCE')
+    if source == 'build-config':
+        return 'build-config'
+    if os.environ.get('ALLOW_BUILD_CONFIG') == '1':
+        try:
+            resolve_token()
+        except (ValueError, OSError):
+            pass
+        else:
+            return 'build-config'
+    return 'sample'
+
+
+def build_config_params(gpu_type=None, framework='vllm', task_type='text-generation'):
+    """Return YAML; opted-in API failures propagate, never fall back to sample."""
+    if config_params_source() == 'build-config':
+        return build_config(gpu_type, framework, task_type)
+    return (Path(__file__).resolve().parents[1] / 'templates/config_params.sample.yaml').read_text()
--- a/src/gate.py
+++ b/src/gate.py
@@ -5,18 +5,23 @@ Live is allowed only when ALL of:
  - LIVE_CONFIRM=I_CONFIRM_LIVE_SUBMIT
  - STRATEGY_ID present

-Phase1: no real HTTP client for /api/adapt/task/add. Live path accepts an
-injected fake client for unit tests only.
+Phase2 client exists but requires explicit injection. No automatic wiring.
+Queue ≥95 blocks; at most one injected submit per process.
 """

 from __future__ import annotations

 import os
+import threading
+from .queue_depth import queue_depth
 from dataclasses import dataclass
 from typing import Any, Callable, Dict, Optional, Protocol

 from .strategy import strategy_id_present

+_submit_count = 0
+_submit_lock = threading.Lock()
+
 LIVE_CONFIRM_VALUE = "I_CONFIRM_LIVE_SUBMIT"


@@ -62,6 +67,11 @@ def evaluate_gate() -> GateDecision:
            allowed=False,
            reason=f"LIVE_CONFIRM must be exactly {LIVE_CONFIRM_VALUE!r}",
        )
+    depth = queue_depth()
+    if depth is None and ("ASYNC_QUEUE_DEPTH" in os.environ or "QUEUE_SNAPSHOT_JSON" in os.environ):
+        return GateDecision("blocked", False, "queue depth invalid/unknown; human confirmation required")
+    if depth is not None and depth >= 95:
+        return GateDecision("blocked", False, "queue ≥95: 禁 live")
    return GateDecision(
        mode="live",
        allowed=True,
@@ -75,10 +85,10 @@ def attempt_live_submit(
 ) -> Dict[str, Any]:
    """Attempt a live submit only if the gate opens and a client is injected.

-    Phase1: without an injected client, even an open gate returns blocked with
-    a message that real HTTP submit is not implemented in this skeleton.
-    Never calls platform write APIs from this module.
+    Phase2 TaskAddClient must be explicitly injected and dual-gated.
+    Tests must use fake clients or mock HTTP only.
    """
+    global _submit_count
    decision = evaluate_gate()
    if not decision.allowed:
        return {
@@ -92,11 +102,17 @@ def attempt_live_submit(
            "ok": False,
            "mode": "blocked",
            "reason": (
-                "Phase1 skeleton: no real task/add client; "
-                "inject a fake client in tests only"
+                "Phase2 task/add client exists; explicit client injection required; "
+                "no real request without injection and dual gate"
            ),
            "submitted": False,
        }
+    with _submit_lock:
+        if _submit_count >= 1:
+            return dict(ok=False, mode="blocked", submitted=False,
+                        reason="process limit: at most 1 real add")
+        # Reserve before invocation: a transport failure may still have created a task.
+        _submit_count += 1
    result = client.submit(payload)
    return {
        "ok": True,
--- a/src/pending_pool.py
+++ b/src/pending_pool.py
@@ -0,0 +1,106 @@
+"""Read-only pending pool. Name checks cannot establish metadata/uniqueness PASS."""
+import os
+import re
+from pathlib import Path
+
+BASE = Path('/workspace/modelhub-adapt')
+GPUS = r'曦望(?: S2)?|壁砺(?:166M)?|天垓150|天垓100|智铠(?: mrv-100)?|Iluvatar_mrv-100|Biren_166m|Sunrise_pt-200-x1'
+FAMILIES = ('lfm2', 'olmo-2', 'olmo2', 'spark-x2.5', 'spark2_5', 'nemotron-h')
+
+
+def input_path(env, filename):
+    return Path(os.environ.get(env) or BASE / filename)
+
+
+def load_do_not_resubmit(path):
+    # Exact ids only; includes collision bullets and overnight tables.
+    return {s for s in re.findall(r'`([^`]+)`', Path(path).read_text())
+            if re.fullmatch(r'[\w.-]+/[\w.\-/]+', s)}
+
+
+def load_banned_name_tokens(path):
+    text = Path(path).read_text().lower()
+    return {s for s in ('abliterated', 'heretic', 'uncensored', 'unsensored') if s in text}
+
+
+def experience_order(gpus):
+    """Stable tier sort of missing chips only; tier-one pending order wins.
+
+    docs/gpu-perf-order.md Section B is experience, never official performance.
+    Unknown names retain their relative order after known tiers; no chips added.
+    """
+    def tier(gpu):
+        if any(name in gpu for name in ('壁砺', '曦望', 'Biren_166m', 'Sunrise_pt-200-x1')):
+            return 1
+        if '天垓150' in gpu:
+            return 2
+        if '天垓100' in gpu:
+            return 3
+        if '智铠' in gpu or 'mrv-100' in gpu:
+            return 4
+        return 5
+    return sorted(dict.fromkeys(gpus), key=tier)
+
+
+def _candidate(model, section, gpu_text, notes):
+    return dict(model_address=model, section=section,
+                try_next_gpus=experience_order(re.findall(GPUS, gpu_text)),
+                perf_order="unset", order_source="experience",
+                notes=notes, comment=notes, skip_reason=None,
+                needs_human=False, status='candidate')
+
+
+def parse_pending_md(path):
+    rows, section = [], None
+    for line in Path(path).read_text().splitlines():
+        heading = re.match(r'##\s+Section\s+([AB])\b', line)
+        if heading:
+            section = heading[1]
+        elif line.startswith('## '):
+            section = None
+        if section and re.match(r'\|\s*\d+\s*\|', line):
+            cols = [s.strip() for s in line.strip('|').split('|')]
+            model = re.search(r'`([^`]+)`', cols[1])
+            if model:
+                rows.append(_candidate(model[1], section, cols[3] if section == 'A' else cols[-1], cols[-1]))
+    return rows
+
+
+def parse_pending_txt(path):
+    rows = []
+    for line in Path(path).read_text().splitlines():
+        if not line.strip() or line.startswith('#'):
+            continue
+        parts = re.split(r'\t|\\t', line, maxsplit=1)
+        comment = parts[1] if len(parts) == 2 else ''
+        match = re.match(r'#([AB])', comment)
+        section = match[1] if match else 'B'
+        gpu_text = re.split(r'(?:still\s+)?try\s*:', comment, maxsplit=1)[-1] if 'try:' in comment else (comment if section == 'B' else '')
+        row = _candidate(parts[0].strip(), section, gpu_text, comment)
+        if not match:
+            row.update(needs_human=True, status='needs_human')
+        rows.append(row)
+    return rows
+
+
+def build_ordered_candidates(md_path=None, txt_path=None, dnr_path=None, banned_path=None):
+    md = Path(md_path) if md_path is not None else input_path('PENDING_SUBMIT_MD', 'pending-submit.md')
+    txt = Path(txt_path) if txt_path is not None else input_path('PENDING_SUBMIT_TXT', 'pending-submit.txt')
+    rows = parse_pending_md(md) if md.is_file() else parse_pending_txt(txt) if txt.is_file() else []
+    dnr = Path(dnr_path) if dnr_path is not None else input_path('DO_NOT_RESUBMIT_MD', 'do-not-resubmit.md')
+    banned = Path(banned_path) if banned_path is not None else input_path('BANNED_PATTERNS_MD', 'banned-patterns.md')
+    excluded = load_do_not_resubmit(dnr) if dnr.is_file() else set()
+    tokens = load_banned_name_tokens(banned) if banned.is_file() else set()
+    section = os.environ.get('PENDING_SECTION', '').strip().upper()
+    if section:
+        rows = [row for row in rows if row['section'] == section]
+    for row in rows:
+        name = row['model_address']
+        reason = ('do-not-resubmit' if name in excluded else
+                  'hard-gate: banned name token' if any(t in name.lower() for t in tokens) else
+                  'hard-gate: banned family' if any(t in name.lower() for t in FAMILIES) else None)
+        row['skip_reason'] = reason
+        # Metadata gates (weights, context, riskTag, uniqueness) remain unverified.
+        row['needs_human'] = not bool(reason)
+        row['status'] = 'skipped' if reason else 'needs_human'
+    return sorted(rows, key=lambda r: r['section'] != 'A')
--- a/src/queue_depth.py
+++ b/src/queue_depth.py
@@ -0,0 +1,22 @@
+"""Read queue hints without contacting the platform; unknown is conservative."""
+import json
+import os
+from pathlib import Path
+
+
+def queue_depth():
+    value = os.environ.get('ASYNC_QUEUE_DEPTH')
+    if value is None and os.environ.get('QUEUE_SNAPSHOT_JSON'):
+        try:
+            data = json.loads(Path(os.environ['QUEUE_SNAPSHOT_JSON']).read_text())
+            if isinstance(data, dict):
+                value = next((data[k] for k in ('depth', 'queue_depth', 'async_queue_depth', 'ASYNC_QUEUE_DEPTH') if k in data), None)
+        except (OSError, ValueError):
+            return None
+    try:
+        if isinstance(value, bool):
+            return None
+        depth = int(str(value))
+        return depth if depth >= 0 else None
+    except (TypeError, ValueError):
+        return None
--- a/src/readonly_probe.py
+++ b/src/readonly_probe.py
@@ -0,0 +1,44 @@
+"""Explicit read-only auth probe. Import performs no I/O; never calls task/add."""
+import json
+import os
+import urllib.error
+import urllib.request
+from src.task_add_client import resolve_token
+
+
+def probe_readonly():
+    try:
+        token = resolve_token()
+    except (ValueError, OSError):
+        return dict(http_status=None, platform_code=None, auth_ok=False,
+                    record_summary={'count': 0}, error_attribution='missing/invalid credential')
+    request = urllib.request.Request(
+        os.environ.get('MODELHUB_API_BASE', 'https://modelhub.org.cn').rstrip('/')
+        + '/api/adapt/task/page?onlyMine=true&pageSize=5&current=1',
+        headers={'Xc-Token': token, 'Accept': 'application/json'}, method='GET')
+    try:
+        with urllib.request.urlopen(request, timeout=30) as response:
+            status, raw = response.status, response.read()
+    except urllib.error.HTTPError as error:
+        status, raw = error.code, error.read()
+    except (OSError, urllib.error.URLError, ValueError):
+        return dict(http_status=None, platform_code=None, auth_ok=False,
+                    record_summary={'count': 0}, error_attribution='transport/timeout')
+    # Never expose response strings: servers can echo credentials in any field.
+    try:
+        body = json.loads(raw)
+    except (ValueError, UnicodeError):
+        body = {}
+    body = body if isinstance(body, dict) else {}
+    code = body.get('code')
+    code = code if type(code) is int else int(code) if isinstance(code, str) and code in ('0', '40100', '40101') else None
+    ok = 200 <= status < 300 and code == 0
+    data = body.get('data')
+    records = data.get('records') if isinstance(data, dict) else None
+    summary = {'count': len(records) if isinstance(records, list) else 0}
+    if isinstance(data, dict) and type(data.get('total')) is int:
+        summary['total'] = data['total']
+    attribution = {40100: '40100 NOT_LOGIN_ERROR', 40101: '40101 NO_AUTH_ERROR'}.get(
+        code, 'success' if ok else 'other HTTP/platform/invalid response')
+    return dict(http_status=status, platform_code=code, auth_ok=ok,
+                record_summary=summary, error_attribution=attribution)
--- a/src/task_add_client.py
+++ b/src/task_add_client.py
@@ -0,0 +1,58 @@
+"""Explicit task/add client; import never sends requests. No retries.
+
+Community Xc-Token authentication is C; platform code 60014 is D (unverified).
+GPU enums/framework/configParams require human verification. Draft placeholders
+must be replaced before a real call. Use through the dual gate with injection.
+"""
+import json
+import os
+import re
+import urllib.request
+import urllib.error
+from pathlib import Path
+
+
+def build_payload(model_address, target_gpu='<TARGET_GPU>', framework='<FRAMEWORK>',
+                  strategy_id=None, config_params='<CONFIG_PARAMS>'):
+    return dict(modelAddress=model_address, taskType='text-generation', targetGpu=target_gpu,
+                framework=framework, strategyId=strategy_id or os.environ.get('STRATEGY_ID') or '<STRATEGY_ID>',
+                configParams=config_params)
+
+
+def resolve_token():
+    file = os.environ.get('XC_TOKEN_FILE')
+    token = Path(file).read_text().strip() if file else next(
+        (os.environ[k].strip() for k in ('XC_TOKEN', 'MODELHUB_XC_TOKEN', 'EXTERNAL_SERVICE_TOKEN') if os.environ.get(k)), '')
+    if not token or re.search(r'(?i)(?:^|[^a-z0-9])(placeholder|changeme|tmp)(?:$|[^a-z0-9])', token):
+        raise ValueError('Missing or invalid service credential')
+    return token
+
+
+class TaskAddClient:
+    def submit(self, payload):
+        if not os.environ.get('STRATEGY_ID', '').strip():
+            raise ValueError('STRATEGY_ID required')
+        required = ('modelAddress', 'taskType', 'targetGpu', 'framework', 'strategyId', 'configParams')
+        if any(k not in payload or not payload[k] or '<' in str(payload[k]) for k in required):
+            raise ValueError('Replace all required payload placeholders')
+        if payload['strategyId'] != os.environ['STRATEGY_ID'].strip():
+            raise ValueError('strategyId must match STRATEGY_ID')
+        token = resolve_token()
+        request = urllib.request.Request(os.environ.get('MODELHUB_API_BASE', 'https://www.modelhub.org.cn').rstrip('/') + '/api/adapt/task/add',
+            data=json.dumps(payload).encode(), headers={'Xc-Token': token, 'Content-Type': 'application/json'}, method='POST')
+        try:
+            with urllib.request.urlopen(request, timeout=30) as response:
+                status, raw = response.status, response.read()
+        except urllib.error.HTTPError as error:
+            status, raw = error.code, error.read()
+        except (OSError, urllib.error.URLError):
+            return dict(http_status=None, body=None, platform_code=None, error_attribution='transport/timeout')
+        # A server may echo credentials; redact before parsing or returning anything.
+        text = raw.decode('utf-8', errors='replace').replace(token, '[REDACTED]')
+        try:
+            body = json.loads(text)
+        except ValueError:
+            body = text
+        code = body.get('code') if isinstance(body, dict) else None
+        attribution = '60014: D (unverified)' if str(code) == '60014' else 'HTTP error' if status >= 400 else 'platform response (unverified)'
+        return dict(http_status=status, body=body, platform_code=code, error_attribution=attribution)
--- a/templates/config_params.sample.yaml
+++ b/templates/config_params.sample.yaml
@@ -0,0 +1,44 @@
+framework: vllm
+api: completion
+lang: en
+max_model_len: 4096
+max_tokens: 1024
+temperature: 0.7
+repetition_penalty: 1.1
+top_p: 0.9
+sut_config:
+  gpu_num: 1
+  values:
+    command:
+    - /opt/conda/bin/vllm
+    - serve
+    - /model
+    - --port
+    - '20644'
+    - --served-model-name
+    - llm
+    - --max-model-len
+    - '4096'
+    - --gpu-memory-utilization
+    - '0.9'
+    - -tp
+    - '1'
+    - --enforce-eager
+    - --trust-remote-code
+ref_config:
+  gpu_num: 1
+  values:
+    command:
+    - vllm
+    - serve
+    - /model
+    - --port
+    - '80'
+    - --served-model-name
+    - llm
+    - --max-model-len
+    - '4096'
+    - -tp
+    - '1'
+    - --enforce-eager
+    - --trust-remote-code
--- a/tests/test_build_config_client.py
+++ b/tests/test_build_config_client.py
@@ -0,0 +1,73 @@
+import runpy
+import io
+import json
+import os
+import unittest
+import urllib.error
+from unittest.mock import MagicMock, patch
+from urllib.parse import urlparse, parse_qs
+from src.build_config_client import build_config, BuildConfigError
+from scripts.build_config_dry import main
+
+
+class TestBuildConfigClient(unittest.TestCase):
+    def response(self, body, status=200):
+        response = MagicMock()
+        response.__enter__.return_value = response
+        response.status = status
+        response.read.return_value = json.dumps(body).encode()
+        return response
+
+    def test_success_request_and_auth(self):
+        for authorization in ('0', '1'):
+            with patch.dict(os.environ, {'XC_TOKEN': 'test-secret', 'BUILD_CONFIG_USE_AUTHORIZATION': authorization}, clear=True), patch('urllib.request.urlopen', return_value=self.response({'code': 0, 'data': 'framework: vllm\n'})) as call:
+                self.assertEqual(build_config('Sunrise_pt-200-x1'), 'framework: vllm\n')
+            request = call.call_args.args[0]
+            self.assertEqual(request.get_method(), 'POST')
+            self.assertIsNone(request.data)
+            parsed = urlparse(request.full_url)
+            self.assertEqual(parsed.path, '/api/adapt/task/build-config')
+            self.assertNotIn('/task/add', request.full_url)
+            self.assertEqual(parse_qs(parsed.query), {'gpuType': ['Sunrise_pt-200-x1'], 'framework': ['vllm'], 'taskType': ['text-generation']})
+            self.assertEqual(request.get_header('Authorization' if authorization == '1' else 'Xc-token'), 'test-secret')
+            self.assertIsNone(request.get_header('Xc-token' if authorization == '1' else 'Authorization'))
+
+    def test_missing_invalid_no_network(self):
+        for env in ({}, {'XC_TOKEN': 'placeholder'}, {'XC_TOKEN_FILE': '/nonexistent/token'}):
+            with patch.dict(os.environ, env, clear=True), patch('urllib.request.urlopen') as call:
+                with self.assertRaisesRegex(BuildConfigError, 'missing/invalid credential'):
+                    build_config('gpu')
+                with patch('sys.argv', ['build_config_dry']), patch('sys.stderr', io.StringIO()):
+                    self.assertEqual(main(), 1)
+                call.assert_not_called()
+
+    def test_platform_and_data_errors(self):
+        for body, reason in [({'code': 40100}, '40100'), ({'code': 40101}, '40101'), ({'code': 9}, 'other HTTP/platform'), ({'code': 0, 'data': ''}, 'empty/non-str'), ({'code': 0, 'data': []}, 'empty/non-str'), ({'code': 0, 'data': 'test-secret'}, 'credential echoed')]:
+            with patch.dict(os.environ, {'XC_TOKEN': 'test-secret'}, clear=True), patch('urllib.request.urlopen', return_value=self.response(body)):
+                with self.assertRaisesRegex(BuildConfigError, reason):
+                    build_config('gpu')
+
+    def test_http_transport_invalid_response(self):
+        for code in (40100, 40101):
+            error = urllib.error.HTTPError('https://example.test', 403, '', {}, io.BytesIO(json.dumps({'code': code}).encode()))
+            with patch.dict(os.environ, {'XC_TOKEN': 'test-secret'}, clear=True), patch('urllib.request.urlopen', side_effect=error):
+                with self.assertRaisesRegex(BuildConfigError, str(code)):
+                    build_config('gpu')
+        with patch.dict(os.environ, {'XC_TOKEN': 'test-secret'}, clear=True):
+            for error in (TimeoutError('secret'), urllib.error.URLError('secret')):
+                with patch('urllib.request.urlopen', side_effect=error), self.assertRaisesRegex(BuildConfigError, 'transport/timeout'):
+                    build_config('gpu')
+            for response in (self.response([], 200), self.response({'code': 0, 'data': 'yaml'}, 500)):
+                with patch('urllib.request.urlopen', return_value=response), self.assertRaises(BuildConfigError):
+                    build_config('gpu')
+            response = self.response({})
+            response.read.return_value = b'not-json'
+            with patch('urllib.request.urlopen', return_value=response), self.assertRaisesRegex(BuildConfigError, 'invalid JSON'):
+                build_config('gpu')
+
+    def test_import_no_io(self):
+        import src.build_config_client
+        with patch('urllib.request.urlopen') as call, patch('pathlib.Path.read_text') as read:
+            runpy.run_path('src/build_config_client.py')
+            call.assert_not_called()
+            read.assert_not_called()
--- a/tests/test_config_params.py
+++ b/tests/test_config_params.py
@@ -0,0 +1,47 @@
+import json
+import re
+import unittest
+import os
+from unittest.mock import patch
+from pathlib import Path
+from src.config_params import build_config_params, config_params_source, GAP
+from src.build_config_client import BuildConfigError
+from scripts.dry_run import _render
+
+
+class TestConfigParams(unittest.TestCase):
+    def setUp(self):
+        env = patch.dict(os.environ, {}, clear=True)
+        env.start()
+        self.addCleanup(env.stop)
+
+    def test_source_selection_and_failure(self):
+        with patch("urllib.request.urlopen") as network:
+            self.assertEqual(config_params_source(), "sample")
+            build_config_params()
+            network.assert_not_called()
+        self.assertIn("discovered via capture", GAP)
+        with patch.dict(os.environ, {"CONFIG_PARAMS_SOURCE": "build-config"}):
+            with self.assertRaisesRegex(BuildConfigError, "missing/invalid credential"):
+                build_config_params("gpu")
+        with patch.dict(os.environ, {"ALLOW_BUILD_CONFIG": "1", "XC_TOKEN": "test-secret"}), patch("src.config_params.build_config", return_value="api: yaml") as client:
+            self.assertEqual(config_params_source(), "build-config")
+            self.assertEqual(build_config_params("gpu"), "api: yaml")
+            client.assert_called_once_with("gpu", "vllm", "text-generation")
+        with patch.dict(os.environ, {"CONFIG_PARAMS_SOURCE": "build-config"}), patch("src.config_params.build_config", side_effect=BuildConfigError("transport/timeout")):
+            with self.assertRaisesRegex(BuildConfigError, "transport/timeout"):
+                build_config_params("gpu")
+
+    def test_official_sample_exact(self):
+        scrape = Path('/workspace/modelhub-adapt/logs/api-docs-scrape-2026-10-16.md')
+        if scrape.is_file():
+            body = json.loads(re.search(r'```json\n(.*?)\n```', scrape.read_text(), re.S)[1])
+            self.assertEqual(build_config_params(), body['configParams'])
+        self.assertIn('max_model_len: 4096', build_config_params())
+        self.assertIn('framework: vllm', build_config_params())
+        self.assertIn('NOT production-ready', GAP)
+
+    def test_report_draft_and_gap(self):
+        report = _render([{'model_address': 'org/a', 'try_next_gpus': ['壁砺']}], 'test', 'fixture')
+        for value in ('代提候选 / 需 Victor 确认', 'configParams source', 'configParams_source=sample', 'GAP:', 'truncated', 'max_model_len: 4096', 'perf_order=unset', 'order_source=experience'):
+            self.assertIn(value, report)
--- a/tests/test_gate.py
+++ b/tests/test_gate.py
@@ -19,6 +19,27 @@ class FakeClient:


 class TestGate(unittest.TestCase):
+    def setUp(self):
+        self.counter = mock.patch('src.gate._submit_count', 0)
+        self.counter.start()
+        self.addCleanup(self.counter.stop)
+        self.env = mock.patch.dict(os.environ, {}, clear=True)
+        self.env.start()
+        self.addCleanup(self.env.stop)
+
+    def test_queue_and_process_limit(self):
+        client = FakeClient()
+        with mock.patch.dict(os.environ, {
+            'STRATEGY_ID': 'sid', 'ALLOW_LIVE_SUBMIT': '1',
+            'LIVE_CONFIRM': LIVE_CONFIRM_VALUE, 'ASYNC_QUEUE_DEPTH': '95'
+        }):
+            self.assertIn('≥95', attempt_live_submit({}, client)['reason'])
+            self.assertEqual(client.calls, [])
+            os.environ['ASYNC_QUEUE_DEPTH'] = '94'
+            self.assertTrue(attempt_live_submit({}, client)['submitted'])
+            self.assertFalse(attempt_live_submit({}, client)['submitted'])
+            self.assertEqual(len(client.calls), 1)
+
    def _clear_gate_env(self):
        return mock.patch.dict(
            os.environ,
@@ -78,7 +99,7 @@ class TestGate(unittest.TestCase):
        self.assertEqual(d.mode, "live")

    def test_open_gate_without_client_still_blocked(self) -> None:
-        """Phase1: no real HTTP client even when gate is open."""
+        """Phase2: an open gate never constructs a network client."""
        with mock.patch.dict(
            os.environ,
            {
--- a/tests/test_pending_pool.py
+++ b/tests/test_pending_pool.py
@@ -0,0 +1,49 @@
+import tempfile
+import unittest
+from pathlib import Path
+from src.pending_pool import build_ordered_candidates, parse_pending_txt
+
+
+class TestPendingPool(unittest.TestCase):
+    def test_order_filters_and_first_gpu(self):
+        with tempfile.TemporaryDirectory() as directory:
+            root = Path(directory)
+            md, txt, dnr, banned = [root / n for n in ('pending.md', 'pending.txt', 'dnr.md', 'banned.md')]
+            md.write_text('''## Section B
+| 1 | `org/cold` | 50 | | | 
+## Section A
+| 1 | `org/good` | 曦望 | 壁砺, 天垓150 | — |
+| 2 | `org/exact` | 曦望 | 壁砺 | — |
+| 3 | `org/Heretic-model` | 曦望 | 壁砺 | — |
+| 4 | `org/OLMo-2-model` | 曦望 | 壁砺 | — |
+''')
+            dnr.write_text('| 1 | `org/exact` | fail |\n## Uniqueness collisions\n- `org/collision`\n## Overnight\n- `org/overnight`')
+            banned.write_text('`abliterated` `heretic` `uncensored` `unsensored`')
+            rows = build_ordered_candidates(md, txt, dnr, banned)
+            self.assertEqual([r['section'] for r in rows], ['A'] * 4 + ['B'])
+            self.assertEqual(rows[0]['try_next_gpus'][0], '壁砺')
+            self.assertEqual(rows[1]['skip_reason'], 'do-not-resubmit')
+            self.assertEqual(rows[2]['status'], 'skipped')
+            self.assertIn('family', rows[3]['skip_reason'])
+            self.assertTrue(rows[-1]['needs_human'])
+            txt.write_text('org/b\t#B cold\norg/a\t#A done:曦望; still try:壁砺+天垓150\norg/c\\t#A try:天垓150\n')
+            self.assertEqual(parse_pending_txt(txt)[1]['try_next_gpus'], ['壁砺', '天垓150'])
+            fallback = build_ordered_candidates(root/'absent', txt, dnr, banned)
+            self.assertEqual([r['model_address'] for r in fallback], ['org/a', 'org/c', 'org/b'])
+            self.assertEqual(len(build_ordered_candidates(md, txt, dnr, banned)), 5)
+
+    def test_experience_order_missing_only_and_ties(self):
+        from src.pending_pool import experience_order, parse_pending_md
+        self.assertEqual(experience_order(['智铠 mrv-100', '天垓100', '曦望', '天垓150', '壁砺']),
+                         ['曦望', '壁砺', '天垓150', '天垓100', '智铠 mrv-100'])
+        self.assertEqual(experience_order(['天垓150']), ['天垓150'])
+        with tempfile.TemporaryDirectory() as directory:
+            path = Path(directory) / 'pending.md'
+            path.write_text('## Section A\n| 1 | `org/a` | 壁砺:验证失败; 曦望:正在验证 | 天垓150 | no 壁砺 retry |\n## Section B\n| 1 | `org/b` | 0 | 壁砺 |\n')
+            from unittest.mock import patch
+            import os
+            with patch.dict(os.environ, {'PENDING_SECTION': 'A'}):
+                rows = build_ordered_candidates(path, dnr_path=Path(directory)/'none', banned_path=Path(directory)/'none')
+            self.assertEqual(len(rows), 1)
+            self.assertEqual(rows[0]['try_next_gpus'], ['天垓150'])
+            self.assertEqual(rows[0]['perf_order'], 'unset')
--- a/tests/test_readonly_probe.py
+++ b/tests/test_readonly_probe.py
@@ -0,0 +1,61 @@
+import io
+import json
+import os
+import unittest
+import urllib.error
+from unittest.mock import patch, MagicMock
+from src.readonly_probe import probe_readonly
+from scripts.probe_readonly import main
+
+
+class TestReadonlyProbe(unittest.TestCase):
+    def test_missing_token_no_network(self):
+        with patch.dict(os.environ, {}, clear=True), patch('urllib.request.urlopen') as call:
+            self.assertFalse(probe_readonly()['auth_ok'])
+            with patch('sys.stdout', io.StringIO()):
+                self.assertEqual(main(), 1)
+            call.assert_not_called()
+
+    def test_success_and_exact_request(self):
+        response = MagicMock()
+        response.__enter__.return_value = response
+        response.status = 200
+        response.read.return_value = json.dumps({'code': 0, 'data': {
+            'records': [{'modelId': 'secret-value'}], 'total': 10}}).encode()
+        with patch.dict(os.environ, {'EXTERNAL_SERVICE_TOKEN': 'secret-value'}, clear=True), patch('urllib.request.urlopen', return_value=response) as call:
+            result = probe_readonly()
+        request = call.call_args.args[0]
+        self.assertEqual(request.get_method(), 'GET')
+        self.assertEqual(request.full_url, 'https://modelhub.org.cn/api/adapt/task/page?onlyMine=true&pageSize=5&current=1')
+        self.assertEqual(request.get_header('Xc-token'), 'secret-value')
+        self.assertTrue(result['auth_ok'])
+        self.assertEqual(result['record_summary'], {'count': 1, 'total': 10})
+        self.assertNotIn('secret-value', json.dumps(result))
+
+    def test_errors_no_fallback_or_echo(self):
+        for code in (40100, 40101, 50000):
+            error = urllib.error.HTTPError('https://example.test', 403, '', {}, io.BytesIO(json.dumps({'code': code, 'message': 'secret-value'}).encode()))
+            with patch.dict(os.environ, {'XC_TOKEN': 'secret-value'}, clear=True), patch('urllib.request.urlopen', side_effect=error) as call:
+                result = probe_readonly()
+            self.assertEqual(call.call_count, 1)
+            self.assertFalse(result['auth_ok'])
+            self.assertEqual(result['platform_code'], code)
+            self.assertIn(str(code) if code != 50000 else 'other', result['error_attribution'])
+            self.assertNotIn('secret-value', json.dumps(result))
+
+    def test_transport_and_invalid_json(self):
+        for error in (OSError('secret-value'), urllib.error.URLError('secret-value')):
+            with patch.dict(os.environ, {'XC_TOKEN': 'secret-value'}, clear=True), patch('urllib.request.urlopen', side_effect=error):
+                self.assertEqual(probe_readonly()['error_attribution'], 'transport/timeout')
+        response = MagicMock()
+        response.__enter__.return_value = response
+        response.status, response.read.return_value = 200, b'invalid secret-value'
+        with patch.dict(os.environ, {'XC_TOKEN': 'secret-value'}, clear=True), patch('urllib.request.urlopen', return_value=response):
+            self.assertFalse(probe_readonly()['auth_ok'])
+
+    def test_import_no_network(self):
+        import importlib
+        import src.readonly_probe
+        with patch('urllib.request.urlopen') as call:
+            importlib.reload(src.readonly_probe)
+            call.assert_not_called()
--- a/tests/test_report_safety.py
+++ b/tests/test_report_safety.py
@@ -21,9 +21,23 @@ class TestReportSafety(unittest.TestCase):
        }, clear=True):
            report = _render([], "test-time", Path("example.json"))
        self.assertNotIn(sentinel, report)
-        for required in ("待对接 pending-submit", "A→B", "需 Victor 人工提交", "积分：待平台核实", "≥95"):
+        for required in ("A→B", "需 Victor 人工提交", "积分：待平台核实", "≥95"):
            self.assertIn(required, report)

+    def test_pending_draft_queue_banner_and_skips(self):
+        rows = [dict(model_address='org/a', section='A', try_next_gpus=['壁砺', '天垓150'],
+                     status='skipped', skip_reason='do-not-resubmit', needs_human=False),
+                dict(model_address='org/b', section='B', try_next_gpus=[],
+                     status='needs_human', needs_human=True)]
+        with mock.patch.dict(os.environ, {'ASYNC_QUEUE_DEPTH': '95'}, clear=True):
+            report = _render(rows, 'test', Path('pending.md'))
+        self.assertIn('禁 live / 队列≥95', report)
+        self.assertIn('do-not-resubmit', report)
+        self.assertIn('needs_human', report)
+        self.assertIn('"targetGpu": "壁砺"', report)
+        self.assertNotIn('"targetGpu": "天垓150"', report)
+        self.assertLess(report.index('Section A'), report.index('Section B'))
+
    def test_http_log_omits_request_details(self):
        output = io.StringIO()
        with contextlib.redirect_stderr(output):
--- a/tests/test_task_add_client.py
+++ b/tests/test_task_add_client.py
@@ -0,0 +1,69 @@
+import importlib
+import io
+import os
+import tempfile
+import unittest
+import urllib.error
+from pathlib import Path
+from unittest import mock
+from src import task_add_client as module
+
+
+class TestTaskAddClient(unittest.TestCase):
+    def setUp(self):
+        patch = mock.patch.dict(os.environ, {'STRATEGY_ID': 'sid', 'XC_TOKEN': 'test-secret-credential'}, clear=True)
+        patch.start()
+        self.addCleanup(patch.stop)
+        self.payload = module.build_payload('org/model', '壁砺', 'vllm', config_params={ 'max_model_len': 4096 })
+
+    def test_import_no_network(self):
+        with mock.patch('urllib.request.urlopen') as urlopen:
+            importlib.reload(module)
+        urlopen.assert_not_called()
+
+    def test_success(self):
+        response = mock.MagicMock()
+        response.__enter__.return_value = response
+        response.status = 200
+        response.read.return_value = b'{"code":0,"message":"test-secret-credential"}'
+        with mock.patch('urllib.request.urlopen', return_value=response) as urlopen:
+            result = module.TaskAddClient().submit(self.payload)
+        self.assertEqual(result['http_status'], 200)
+        self.assertNotIn(os.environ['XC_TOKEN'], str(result))
+        request = urlopen.call_args.args[0]
+        self.assertEqual(request.get_header('Xc-token'), os.environ['XC_TOKEN'])
+        self.assertEqual(request.method, 'POST')
+        urlopen.assert_called_once()
+
+    def test_http_4xx_60014(self):
+        error = urllib.error.HTTPError('https://example.invalid', 403, 'forbidden', {}, io.BytesIO(b'{"code":60014}'))
+        with mock.patch('urllib.request.urlopen', side_effect=error) as urlopen:
+            result = module.TaskAddClient().submit(self.payload)
+        self.assertEqual(result['http_status'], 403)
+        self.assertEqual(result['platform_code'], 60014)
+        self.assertIn('D (unverified)', result['error_attribution'])
+        urlopen.assert_called_once()
+
+    def test_transport_no_retry(self):
+        with mock.patch('urllib.request.urlopen', side_effect=TimeoutError('test-secret-credential')) as urlopen:
+            result = module.TaskAddClient().submit(self.payload)
+        self.assertNotIn(os.environ['XC_TOKEN'], str(result))
+        self.assertEqual(result['error_attribution'], 'transport/timeout')
+        urlopen.assert_called_once()
+
+    def test_token_precedence_and_rejections(self):
+        with tempfile.TemporaryDirectory() as directory:
+            path = Path(directory)/'credential'
+            path.write_text(' file-credential \n')
+            with mock.patch.dict(os.environ, {'XC_TOKEN_FILE': str(path)}):
+                self.assertEqual(module.resolve_token(), 'file-credential')
+        for token in ('', 'tmp', 'PLACEHOLDER', 'changeme', 'tmp-token'):
+            with self.subTest(token=token), mock.patch.dict(os.environ, {'XC_TOKEN': token}):
+                with self.assertRaises(ValueError):
+                    module.resolve_token()
+
+    def test_placeholder_never_requests(self):
+        with mock.patch('urllib.request.urlopen') as urlopen:
+            with self.assertRaises(ValueError):
+                module.TaskAddClient().submit(module.build_payload('org/model'))
+        urlopen.assert_not_called()