estelledc · estelledc · Jul 14, 2026 · Jul 14, 2026
diff --git a/SESSION-HANDOFF.md b/SESSION-HANDOFF.md
@@ -2,6 +2,31 @@
 
 > 状态：当前接班入口。旧的批量生产 session 快照已失效，不得用于恢复自动循环；持续运行使用只读 supervisor + 有界 writer epoch。
 
+## 2026-07-14 总结不足后再推进 4 篇补强论文
+
+- status：`local-ready`（已完成本地内容、receipt、atlas 和主要门禁；远端 PR / merge / Pages deploy 仍以实时外部状态为准）。
+- 起始 ref：`6dd71d8868a0142b88f2afefbdce353dba147678`（PR #32 merge 后的 `origin/main`）。
+- 分支：`study/papers-20260714-one-more-round`。
+- 本轮不足总结：上一轮 40 篇完成了规模和部署闭环，但多数卡片仍是 `STATIC_ANALYSIS` / `UNVERIFIED`；部分卡片 91 行、低于建议 100 行；L4 主要是 toy / manual simulation；主题上对 ML 工程 agent、终端 agent、长上下文有效窗口、视觉 Web GUI agent 的覆盖仍不够。
+- 本轮 objective：新增 4 篇更厚的 `study-v2` paper note，分别补强 `MLE-bench`、`Terminal-Bench`、`RULER`、`VisualWebArena`，保持 `UNVERIFIED` 边界，不声明运行真实 benchmark。
+- scope：允许新增 `src/content/docs/papers/*.md`、`data/review-receipts/papers/*.json`，刷新 `data/note-index.json`、papers atlas 派生页和公开计数文案；不修改候选队列、policy/threshold、既有论文正文语义。
+- activated_by：`explicit-user-request-2026-07-14-summarize-gaps-and-advance-one-more-round`
+- review_after：`2026-07-14`
+- acceptance_checks：
+  - `lr search arxiv` + arXiv API 元数据核验 4/4；
+  - `node scripts/quality-gate.mjs` 逐篇通过，行数分别为 135 / 137 / 147 / 138，无 advisory；
+  - `npm run audit:content-contract`：0 blocking，56 v2；
+  - `npm run atlas`：2028 notes，69 chunks；
+  - `npm run audit:counts`：projects=961、papers=1067、total=2028；
+  - `npm run audit:links` / `npm run audit:wikilinks`：无 blocking；
+  - `git ls-files -co --exclude-standard -z | node scripts/audit-public-redlines.mjs --stdin0`：0 blocking；
+  - `npm run build:strict -- --log /tmp/study-one-more-round-build-clean.log`：通过；首次失败由 stale `.astro` cache 触发 duplicate-id warning，删除 ignored cache 后恢复；
+  - `git diff --check`：通过。
+- budget：1 个内容小批次、4 篇新增 paper、1 个可写切片、1 个本地 writer。
+- external_outcome：当前为本地 review-ready branch；PR、merge 和 Pages deploy 需要以单独外部动作完成并复核。
+- stop_conditions：规范工具链不可用；arXiv 来源不可核验；content contract / redline / strict build 失败且无法在 scope 内修复；需要改 policy/threshold、候选队列或敏感内容；用户停止。
+- superseded_by：`none`
+
 ## 2026-07-14 新增 40 篇论文全流程完成记录
 
 - status：`complete`

diff --git a/data/note-index.json b/data/note-index.json
@@ -3,16 +3,16 @@
   "taxonomy_version": "taxonomy-v1",
   "stats": {
     "summary": {
-      "total": 2024,
-      "classified": 1977,
+      "total": 2028,
+      "classified": 1981,
       "unclassified": 47,
       "unknown_difficulty": 1975,
       "empty_description": 1970
     },
     "by_area": {
       "papers": {
-        "total": 1063,
-        "classified": 1044,
+        "total": 1067,
+        "classified": 1048,
         "unclassified": 19,
         "unknown_difficulty": 1014,
         "empty_description": 1013
@@ -20611,6 +20611,38 @@
         "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
       }
     },
+    {
+      "id": "papers::mle-bench",
+      "area": "papers",
+      "slug": "mle-bench",
+      "title": "MLE-bench — 用 Kaggle 任务衡量机器学习工程 agent",
+      "description": "用 MLE-bench 理解 ML 工程 agent 为什么不能只靠单元测试和代码 benchmark 来评估。",
+      "difficulty": "intermediate",
+      "canonical_topics": [
+        "papers-agents-and-llm-systems"
+      ],
+      "classification": {
+        "state": "classified",
+        "source": "frontmatter-category",
+        "topic_id": "papers-agents-and-llm-systems",
+        "matched_category": "agent",
+        "raw_category": "AI Agent / MLE Benchmark"
+      },
+      "trust": {
+        "contract_state": "v2",
+        "verification_status": "UNVERIFIED"
+      },
+      "freshness": {
+        "state": "NOT_EVALUATED",
+        "reviewed_at": "2026-07-14",
+        "review_after": null
+      },
+      "route": "/study/papers/mle-bench/",
+      "atlas": {
+        "chunk_id": "topic-papers-agents-and-llm-systems-01",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+      }
+    },
     {
       "id": "papers::mlflow",
       "area": "papers",
@@ -26401,6 +26433,38 @@
         "chunk_route": "/study/atlas/papers/topic-papers-network-protocols-01/"
       }
     },
+    {
+      "id": "papers::ruler-long-context",
+      "area": "papers",
+      "slug": "ruler-long-context",
+      "title": "RULER — 真实长上下文能力不能只看 NIAH",
+      "description": "用 RULER 理解长上下文模型的有效窗口、检索幻觉和聚合推理为什么要分开评测。",
+      "difficulty": "intermediate",
+      "canonical_topics": [
+        "papers-agents-and-llm-systems"
+      ],
+      "classification": {
+        "state": "classified",
+        "source": "frontmatter-category",
+        "topic_id": "papers-agents-and-llm-systems",
+        "matched_category": "llm",
+        "raw_category": "LLM / Long Context Evaluation"
+      },
+      "trust": {
+        "contract_state": "v2",
+        "verification_status": "UNVERIFIED"
+      },
+      "freshness": {
+        "state": "NOT_EVALUATED",
+        "reviewed_at": "2026-07-14",
+        "review_after": null
+      },
+      "route": "/study/papers/ruler-long-context/",
+      "atlas": {
+        "chunk_id": "topic-papers-agents-and-llm-systems-01",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+      }
+    },
     {
       "id": "papers::rwkv-2023",
       "area": "papers",
@@ -30633,6 +30697,38 @@
         "chunk_route": "/study/atlas/papers/topic-papers-distributed-training-and-gpu-01/"
       }
     },
+    {
+      "id": "papers::terminal-bench",
+      "area": "papers",
+      "slug": "terminal-bench",
+      "title": "Terminal-Bench — 在真实命令行任务里测试 agent",
+      "description": "用 Terminal-Bench 理解终端环境为什么能暴露 agent 的长程执行、环境理解和验证能力。",
+      "difficulty": "intermediate",
+      "canonical_topics": [
+        "papers-agents-and-llm-systems"
+      ],
+      "classification": {
+        "state": "classified",
+        "source": "frontmatter-category",
+        "topic_id": "papers-agents-and-llm-systems",
+        "matched_category": "agent",
+        "raw_category": "AI Agent / Terminal Benchmark"
+      },
+      "trust": {
+        "contract_state": "v2",
+        "verification_status": "UNVERIFIED"
+      },
+      "freshness": {
+        "state": "NOT_EVALUATED",
+        "reviewed_at": "2026-07-14",
+        "review_after": null
+      },
+      "route": "/study/papers/terminal-bench/",
+      "atlas": {
+        "chunk_id": "topic-papers-agents-and-llm-systems-01",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+      }
+    },
     {
       "id": "papers::tesla-architecture-2008",
       "area": "papers",
@@ -32513,6 +32609,38 @@
         "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-02/"
       }
     },
+    {
+      "id": "papers::visualwebarena",
+      "area": "papers",
+      "slug": "visualwebarena",
+      "title": "VisualWebArena — 让网页 agent 真正看见界面",
+      "description": "用 VisualWebArena 理解多模态 web agent 为什么不能只读 DOM 文本，还要处理视觉线索。",
+      "difficulty": "intermediate",
+      "canonical_topics": [
+        "papers-agents-and-llm-systems"
+      ],
+      "classification": {
+        "state": "classified",
+        "source": "frontmatter-category",
+        "topic_id": "papers-agents-and-llm-systems",
+        "matched_category": "agent",
+        "raw_category": "AI Agent / Multimodal Web"
+      },
+      "trust": {
+        "contract_state": "v2",
+        "verification_status": "UNVERIFIED"
+      },
+      "freshness": {
+        "state": "NOT_EVALUATED",
+        "reviewed_at": "2026-07-14",
+        "review_after": null
+      },
+      "route": "/study/papers/visualwebarena/",
+      "atlas": {
+        "chunk_id": "topic-papers-agents-and-llm-systems-01",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+      }
+    },
     {
       "id": "papers::vit",
       "area": "papers",
@@ -32761,8 +32889,8 @@
       },
       "route": "/study/papers/voyager/",
       "atlas": {
-        "chunk_id": "topic-papers-agents-and-llm-systems-01",
-        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+        "chunk_id": "topic-papers-agents-and-llm-systems-02",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/"
       }
     },
     {
@@ -33112,8 +33240,8 @@
       },
       "route": "/study/papers/webgpt-2021/",
       "atlas": {
-        "chunk_id": "topic-papers-agents-and-llm-systems-01",
-        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+        "chunk_id": "topic-papers-agents-and-llm-systems-02",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/"
       }
     },
     {
@@ -33176,8 +33304,8 @@
       },
       "route": "/study/papers/webxskill/",
       "atlas": {
-        "chunk_id": "topic-papers-agents-and-llm-systems-01",
-        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+        "chunk_id": "topic-papers-agents-and-llm-systems-02",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/"
       }
     },
     {
@@ -33400,8 +33528,8 @@
       },
       "route": "/study/papers/wizardlm-2023/",
       "atlas": {
-        "chunk_id": "topic-papers-agents-and-llm-systems-01",
-        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+        "chunk_id": "topic-papers-agents-and-llm-systems-02",
+        "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/"
       }
     },
     {
@@ -64331,7 +64459,7 @@
       "page": 2,
       "pages": 2,
       "route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/",
-      "entries": 1
+      "entries": 5
     },
     {
       "id": "topic-papers-ai-safety-and-interpretability-01",

diff --git a/data/review-receipts/papers/mle-bench.json b/data/review-receipts/papers/mle-bench.json
@@ -0,0 +1,55 @@
+{
+  "schema_version": "study-review-receipt-v1",
+  "generation": 1,
+  "predecessor_digest_sha256": null,
+  "note": {
+    "area": "papers",
+    "slug": "mle-bench",
+    "digest_sha256": "b093cd115076731f58de53743a0cf5a47825c14c74d0f59731cf1f8f01b3415c"
+  },
+  "source_revision": "arXiv:2410.07095v6",
+  "research_input_sha256": "8693c68493b19f35534fc4630738ec7c7e77bdc70ddbe247e870e49aae2d3ae6",
+  "reviewers": [
+    {
+      "role": "ZERO_BASE",
+      "reviewer_version": "study-static-review-20260714-one-more-round",
+      "decision": "PASS_WITH_NOTES",
+      "score": 86,
+      "warnings": [
+        "Explains Kaggle-style ML engineering with a toy workflow; no Kaggle task was executed."
+      ],
+      "execution": {
+        "review_mode": "STATIC_REVIEW",
+        "code_mode": "MANUAL_SIMULATION"
+      }
+    },
+    {
+      "role": "ENGINEER",
+      "reviewer_version": "study-static-review-20260714-one-more-round",
+      "decision": "PASS_WITH_NOTES",
+      "score": 84,
+      "warnings": [
+        "Engineering claims are limited to static paper reading; AIDE scaffold and benchmark harness were not run."
+      ],
+      "execution": {
+        "review_mode": "STATIC_REVIEW",
+        "code_mode": "NOT_APPLICABLE"
+      }
+    },
+    {
+      "role": "ACADEMIC",
+      "reviewer_version": "study-static-review-20260714-one-more-round",
+      "decision": "PASS_WITH_NOTES",
+      "score": 84,
+      "warnings": [
+        "Citation identity was checked through arXiv metadata; reported medal rates are not independently reproduced."
+      ],
+      "execution": {
+        "review_mode": "STATIC_REVIEW",
+        "code_mode": "NOT_APPLICABLE"
+      }
+    }
+  ],
+  "waivers": [],
+  "created_at": "2026-07-14T14:30:00.000Z"
+}
diff --git a/data/review-receipts/papers/ruler-long-context.json b/data/review-receipts/papers/ruler-long-context.json
@@ -0,0 +1,55 @@
+{
+  "schema_version": "study-review-receipt-v1",
+  "generation": 1,
+  "predecessor_digest_sha256": null,
+  "note": {
+    "area": "papers",
+    "slug": "ruler-long-context",
+    "digest_sha256": "fdea3ef9179c032b4040312f8720d8e43b0a09628bbf8aef951386d00a1cea81"
+  },
+  "source_revision": "arXiv:2404.06654v3",
+  "research_input_sha256": "585001909e88932e9aa31c90a0fb62d22fe20ca6e3e5588dcbcc0cc841768ac8",
+  "reviewers": [
+    {
+      "role": "ZERO_BASE",
+      "reviewer_version": "study-static-review-20260714-one-more-round",
+      "decision": "PASS_WITH_NOTES",
+      "score": 87,
+      "warnings": [
+        "Explains RULER through a toy long-context tracing example; benchmark tasks were not executed."
+      ],
+      "execution": {
+        "review_mode": "STATIC_REVIEW",
+        "code_mode": "MANUAL_SIMULATION"
+      }
+    },
+    {
+      "role": "ENGINEER",
+      "reviewer_version": "study-static-review-20260714-one-more-round",
+      "decision": "PASS_WITH_NOTES",
+      "score": 84,
+      "warnings": [
+        "Engineering claims are limited to static evaluation design reading; no model inference was run."
+      ],
+      "execution": {
+        "review_mode": "STATIC_REVIEW",
+        "code_mode": "NOT_APPLICABLE"
+      }
+    },
+    {
+      "role": "ACADEMIC",
+      "reviewer_version": "study-static-review-20260714-one-more-round",
+      "decision": "PASS_WITH_NOTES",
+      "score": 85,
+      "warnings": [
+        "Citation identity was checked through arXiv metadata; reported long-context results are not independently reproduced."
+      ],
+      "execution": {
+        "review_mode": "STATIC_REVIEW",
+        "code_mode": "NOT_APPLICABLE"
+      }
+    }
+  ],
+  "waivers": [],
+  "created_at": "2026-07-14T14:30:00.000Z"
+}