diff --git a/SESSION-HANDOFF.md b/SESSION-HANDOFF.md
index 5c0c4e20b..f11775ce8 100644
--- 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 新增 40 篇论文与部署 Epoch Contract
+
+- status：`running`
+- objective：在用户明确授权“分十批新研究 40 篇论文，全流程部署”下，新增 10 批 × 4 篇公开 arXiv 论文研究卡，覆盖 foundation/scaling、开放模型、instruction tuning、reasoning prompt、agent/tool use、PEFT、长上下文/推理、多模态生成与评测安全。
+- scope：允许新增 `src/content/docs/papers/*.md`、`data/review-receipts/papers/*.json`，刷新 `data/note-index.json` 与 papers atlas 派生页，同步公开规模文案和本 handoff；不修改候选队列、policy/threshold、既有论文正文语义或远端配置。
+- activated_by：`explicit-user-request-2026-07-14-new-40-papers-full-deploy`
+- review_after：`2026-07-14`
+- acceptance_checks：
+ - arXiv API 元数据校验：40/40 条目可解析；
+ - `node scripts/quality-gate.mjs --changed-from main --json`：checked=40, pass=true；
+ - 40 份 `study-review-receipt-v1` 的 canonical note digest 与正文一致；
+ - `npm run atlas`：2024 notes, 69 chunks；
+ - `npm run audit:counts`；
+ - `npm run audit:content-contract`；
+ - `npm run audit:links`；
+ - `npm run audit:wikilinks`；
+ - `git ls-files -co --exclude-standard -z | node scripts/audit-public-redlines.mjs --stdin0`；
+ - `npm run build:strict -- --log /tmp/study-forty-build-clean.log`；
+ - `git diff --check`；
+ - 提交后使用 `STUDY_CHANGED_FROM=384787e09827c336baf5ac2b33e67e8c91b9df49 npm run verify:ci` 做 PR/Pages portable gate。
+- budget：10 个内容小批次、40 篇新增 paper、1 个可写切片、1 个本地 writer、1 次 branch/PR/merge/deploy 窗口。
+- external_outcome：40 篇新增论文笔记进入公开 study 站点；验证状态保持 `UNVERIFIED`，不声明实际运行论文 benchmark。
+- stop_conditions：规范 Node/npm 不可用；arXiv 来源不可核验；内容契约、红线审计、strict build 或 verify:ci 失败且无法在本 scope 内修复；需要修改 policy/threshold、候选队列或隐私敏感内容；远端 CI/Pages 失败且需要新权限；用户停止。
+- superseded_by：`none`
+
## 2026-07-14 新增 4 篇论文与部署 Epoch Contract
- status：`running`
diff --git a/data/note-index.json b/data/note-index.json
index 92d87edab..e543081ee 100644
--- a/data/note-index.json
+++ b/data/note-index.json
@@ -3,16 +3,16 @@
"taxonomy_version": "taxonomy-v1",
"stats": {
"summary": {
- "total": 1984,
- "classified": 1937,
+ "total": 2024,
+ "classified": 1977,
"unclassified": 47,
"unknown_difficulty": 1975,
"empty_description": 1970
},
"by_area": {
"papers": {
- "total": 1023,
- "classified": 1004,
+ "total": 1063,
+ "classified": 1044,
"unclassified": 19,
"unknown_difficulty": 1014,
"empty_description": 1013
@@ -27,7 +27,7 @@
},
"missing_curated_assignments": [],
"atlas": {
- "chunks": 68,
+ "chunks": 69,
"chunk_size": 100,
"max_chunk_entries": 100
}
@@ -2596,6 +2596,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-formal-methods-01/"
}
},
+ {
+ "id": "papers::big-bench-hard-2022",
+ "area": "papers",
+ "slug": "big-bench-hard-2022",
+ "title": "BIG-Bench Hard — 从大题库里挑出模型最头疼的 23 类题",
+ "description": "用 BBH 理解为什么 benchmark 需要难题子集和 CoT 对照。",
+ "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 / Evaluation"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/big-bench-hard-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::big-little-2011",
"area": "papers",
@@ -2660,6 +2692,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-hci-and-software-engineering-research-01/"
}
},
+ {
+ "id": "papers::bigbird-2020",
+ "area": "papers",
+ "slug": "bigbird-2020",
+ "title": "BigBird — 用稀疏 attention 拉长 Transformer 视野",
+ "description": "用 BigBird 理解局部、全局和随机 attention 怎样组成长序列模式。",
+ "difficulty": "advanced",
+ "canonical_topics": [
+ "papers-nlp-foundations-and-scaling"
+ ],
+ "classification": {
+ "state": "classified",
+ "source": "frontmatter-category",
+ "topic_id": "papers-nlp-foundations-and-scaling",
+ "matched_category": "nlp",
+ "raw_category": "NLP / Efficient Attention"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/bigbird-2020/",
+ "atlas": {
+ "chunk_id": "topic-papers-nlp-foundations-and-scaling-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
+ }
+ },
{
"id": "papers::biggan-2018",
"area": "papers",
@@ -3010,6 +3074,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-information-retrieval-and-recommendation-01/"
}
},
+ {
+ "id": "papers::bloom-2022",
+ "area": "papers",
+ "slug": "bloom-2022",
+ "title": "BLOOM — 把 176B 多语种模型做成开放科学工程",
+ "description": "用 BLOOM 理解大模型也可以用社区协作、数据治理和开放发布来推进。",
+ "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 / Open Science"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/bloom-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::bm25",
"area": "papers",
@@ -6226,6 +6322,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-cryptography-and-security-01/"
}
},
+ {
+ "id": "papers::controlnet-2023",
+ "area": "papers",
+ "slug": "controlnet-2023",
+ "title": "ControlNet — 给扩散模型加一条可控条件支路",
+ "description": "用 ControlNet 理解边缘、姿态和深度图如何稳定控制图像生成。",
+ "difficulty": "intermediate",
+ "canonical_topics": [
+ "papers-generative-models-and-diffusion"
+ ],
+ "classification": {
+ "state": "classified",
+ "source": "frontmatter-category",
+ "topic_id": "papers-generative-models-and-diffusion",
+ "matched_category": "diffusion",
+ "raw_category": "Diffusion / Control"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/controlnet-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-generative-models-and-diffusion-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-generative-models-and-diffusion-01/"
+ }
+ },
{
"id": "papers::cook-1984-distributed-ray-tracing",
"area": "papers",
@@ -8993,6 +9121,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-reinforcement-learning-01/"
}
},
+ {
+ "id": "papers::dreambooth-2022",
+ "area": "papers",
+ "slug": "dreambooth-2022",
+ "title": "DreamBooth — 用几张图把一个新主体塞进生成模型",
+ "description": "用 DreamBooth 理解 subject-driven generation 怎样让扩散模型记住特定对象。",
+ "difficulty": "intermediate",
+ "canonical_topics": [
+ "papers-generative-models-and-diffusion"
+ ],
+ "classification": {
+ "state": "classified",
+ "source": "frontmatter-category",
+ "topic_id": "papers-generative-models-and-diffusion",
+ "matched_category": "diffusion",
+ "raw_category": "Diffusion / Personalization"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/dreambooth-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-generative-models-and-diffusion-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-generative-models-and-diffusion-01/"
+ }
+ },
{
"id": "papers::dreamfusion-2022",
"area": "papers",
@@ -12660,6 +12820,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-computer-graphics-and-visualization-01/"
}
},
+ {
+ "id": "papers::gorilla-2023",
+ "area": "papers",
+ "slug": "gorilla-2023",
+ "title": "Gorilla — 让 LLM 学会查 API 文档再调用",
+ "description": "用 Gorilla 理解 API grounding 如何降低工具调用幻觉。",
+ "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 / Tool Use"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/gorilla-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::gortler-1996-lumigraph",
"area": "papers",
@@ -13326,6 +13518,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
}
},
+ {
+ "id": "papers::gsm8k-2021",
+ "area": "papers",
+ "slug": "gsm8k-2021",
+ "title": "GSM8K — 小学数学题把大模型算术短板照出来",
+ "description": "用 GSM8K 理解数学 word problem、verifier 和采样重排为什么重要。",
+ "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 / Math Evaluation"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/gsm8k-2021/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::hacl-star-2017",
"area": "papers",
@@ -14342,6 +14566,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-information-and-coding-theory-01/"
}
},
+ {
+ "id": "papers::hugginggpt-2023",
+ "area": "papers",
+ "slug": "hugginggpt-2023",
+ "title": "HuggingGPT — 让 ChatGPT 当任务调度员，模型库当工具箱",
+ "description": "用 HuggingGPT 理解 LLM 如何规划并调用专用模型完成多模态任务。",
+ "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 / Tool Orchestration"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/hugginggpt-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::hughes-fp-matters",
"area": "papers",
@@ -14754,6 +15010,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-databases-01/"
}
},
+ {
+ "id": "papers::inner-monologue-2022",
+ "area": "papers",
+ "slug": "inner-monologue-2022",
+ "title": "Inner Monologue — 让机器人把观察结果说回计划里",
+ "description": "用 Inner Monologue 理解闭环反馈如何让语言计划接上真实环境变化。",
+ "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": "LLM Agent / Robotics"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/inner-monologue-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::instant-ngp-2022",
"area": "papers",
@@ -15516,6 +15804,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-compilers-and-programming-language-theory-01/"
}
},
+ {
+ "id": "papers::kaplan-scaling-laws-2020",
+ "area": "papers",
+ "slug": "kaplan-scaling-laws-2020",
+ "title": "Scaling Laws — 大模型训练不是玄学，是幂律预算题",
+ "description": "用 Kaplan scaling laws 理解参数、数据和计算量怎样一起决定语言模型损失。",
+ "difficulty": "advanced",
+ "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 / Scaling Laws"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/kaplan-scaling-laws-2020/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::karger-1997-consistent-hashing",
"area": "papers",
@@ -16467,6 +16787,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-information-retrieval-and-recommendation-01/"
}
},
+ {
+ "id": "papers::lamda-2022",
+ "area": "papers",
+ "slug": "lamda-2022",
+ "title": "LaMDA — 聊天模型先学会有用、具体和不乱说",
+ "description": "用 LaMDA 理解开放域对话模型为什么需要质量、安全和 groundedness 三条线。",
+ "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 / Dialogue"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/lamda-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::lamport-1978",
"area": "papers",
@@ -16785,6 +17137,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-compilers-and-programming-language-theory-01/"
}
},
+ {
+ "id": "papers::least-to-most-prompting-2022",
+ "area": "papers",
+ "slug": "least-to-most-prompting-2022",
+ "title": "Least-to-Most — 先拆小题，再解大题",
+ "description": "用 Least-to-Most Prompting 理解复杂推理为什么要先分解再逐步求解。",
+ "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 / Reasoning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/least-to-most-prompting-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::lee-keystone-2020",
"area": "papers",
@@ -17200,6 +17584,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-distributed-systems-01/"
}
},
+ {
+ "id": "papers::linformer-2020",
+ "area": "papers",
+ "slug": "linformer-2020",
+ "title": "Linformer — 把 attention 矩阵投影成线性复杂度",
+ "description": "用 Linformer 理解低秩假设如何压缩 self-attention。",
+ "difficulty": "advanced",
+ "canonical_topics": [
+ "papers-nlp-foundations-and-scaling"
+ ],
+ "classification": {
+ "state": "classified",
+ "source": "frontmatter-category",
+ "topic_id": "papers-nlp-foundations-and-scaling",
+ "matched_category": "nlp",
+ "raw_category": "NLP / Efficient Attention"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/linformer-2020/",
+ "atlas": {
+ "chunk_id": "topic-papers-nlp-foundations-and-scaling-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
+ }
+ },
{
"id": "papers::linux-kernel",
"area": "papers",
@@ -17802,6 +18218,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
}
},
+ {
+ "id": "papers::longnet-2023",
+ "area": "papers",
+ "slug": "longnet-2023",
+ "title": "LongNet — 用 dilated attention 把上下文推到十亿 token 想象空间",
+ "description": "用 LongNet 理解扩张式 attention 如何在多尺度上连接超长序列。",
+ "difficulty": "advanced",
+ "canonical_topics": [
+ "papers-nlp-foundations-and-scaling"
+ ],
+ "classification": {
+ "state": "classified",
+ "source": "frontmatter-category",
+ "topic_id": "papers-nlp-foundations-and-scaling",
+ "matched_category": "nlp",
+ "raw_category": "NLP / Long Context"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/longnet-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-nlp-foundations-and-scaling-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
+ }
+ },
{
"id": "papers::loong-doc-mt",
"area": "papers",
@@ -19780,6 +20228,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-compilers-and-programming-language-theory-01/"
}
},
+ {
+ "id": "papers::minerva-2022",
+ "area": "papers",
+ "slug": "minerva-2022",
+ "title": "Minerva — 把语言模型拉进数学草稿纸",
+ "description": "用 Minerva 理解为什么数学推理需要专门的数据、逐步解题和采样验证。",
+ "difficulty": "advanced",
+ "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 / Math Reasoning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/minerva-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::minhash-broder-1997",
"area": "papers",
@@ -20004,6 +20484,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
}
},
+ {
+ "id": "papers::mistral-7b-2023",
+ "area": "papers",
+ "slug": "mistral-7b-2023",
+ "title": "Mistral 7B — 小模型靠架构细节打出性价比",
+ "description": "用 Mistral 7B 理解 grouped-query attention 和 sliding-window attention 如何服务高效开源模型。",
+ "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 / Efficient Model"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/mistral-7b-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::mitls-2014-triple-handshake",
"area": "papers",
@@ -20547,6 +21059,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-network-protocols-01/"
}
},
+ {
+ "id": "papers::mrkl-systems-2022",
+ "area": "papers",
+ "slug": "mrkl-systems-2022",
+ "title": "MRKL — 给大模型配一组专家工具和路由器",
+ "description": "用 MRKL Systems 理解 neuro-symbolic agent 为什么要把 LLM、检索和计算模块拆开。",
+ "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 / Tool Architecture"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/mrkl-systems-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::ms-marco-2016",
"area": "papers",
@@ -20898,6 +21442,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-distributed-systems-01/"
}
},
+ {
+ "id": "papers::natural-instructions-v2-2022",
+ "area": "papers",
+ "slug": "natural-instructions-v2-2022",
+ "title": "Super-NaturalInstructions — 1600+ 任务教模型读懂说明书",
+ "description": "用 Super-NaturalInstructions 理解 declarative instructions 如何评测任务泛化。",
+ "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 / Instruction Benchmark"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/natural-instructions-v2-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::nbeats-2020",
"area": "papers",
@@ -21978,6 +22554,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-operating-systems-and-cluster-management-01/"
}
},
+ {
+ "id": "papers::opt-2022",
+ "area": "papers",
+ "slug": "opt-2022",
+ "title": "OPT — 把 GPT-3 级训练日志打开给研究社区",
+ "description": "用 OPT 理解开放权重、训练日志和复现实验对 LLM 研究的重要性。",
+ "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 / Open Model"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/opt-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::optuna",
"area": "papers",
@@ -22074,6 +22682,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
}
},
+ {
+ "id": "papers::orca-explanation-tuning-2023",
+ "area": "papers",
+ "slug": "orca-explanation-tuning-2023",
+ "title": "Orca — 小模型不只抄答案，还学解释轨迹",
+ "description": "用 Orca 理解 explanation tuning 为什么比只蒸馏最终答案更像教学生。",
+ "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 / Distillation"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/orca-explanation-tuning-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::oscar-int2-kv",
"area": "papers",
@@ -22202,6 +22842,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-distributed-training-and-gpu-01/"
}
},
+ {
+ "id": "papers::p-tuning-v2-2021",
+ "area": "papers",
+ "slug": "p-tuning-v2-2021",
+ "title": "P-Tuning v2 — 把 prompt tuning 深插到每一层",
+ "description": "用 P-Tuning v2 理解深层连续提示为什么能跨规模和任务接近 fine-tuning。",
+ "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 / Efficient Finetuning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/p-tuning-v2-2021/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::p4-2014",
"area": "papers",
@@ -22425,6 +23097,70 @@
"chunk_route": "/study/atlas/papers/topic-papers-hci-and-software-engineering-research-01/"
}
},
+ {
+ "id": "papers::pal-code-reasoning-2022",
+ "area": "papers",
+ "slug": "pal-code-reasoning-2022",
+ "title": "PAL — 让 Python 成为语言模型的草稿纸",
+ "description": "用 PAL 理解 Program-aided Language Models 如何把推理转成可运行代码。",
+ "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 / Tool Reasoning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/pal-code-reasoning-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
+ {
+ "id": "papers::palm-2022",
+ "area": "papers",
+ "slug": "palm-2022",
+ "title": "PaLM — Pathways 把 540B LLM 扩成统一底座",
+ "description": "用 PaLM 理解大规模 dense decoder 如何在多任务、推理和代码能力上同时冒头。",
+ "difficulty": "advanced",
+ "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 / Foundation Model"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/palm-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::panel",
"area": "papers",
@@ -23192,6 +23928,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-distributed-systems-01/"
}
},
+ {
+ "id": "papers::plan-and-solve-prompting-2023",
+ "area": "papers",
+ "slug": "plan-and-solve-prompting-2023",
+ "title": "Plan-and-Solve — 零样本推理先写计划再执行",
+ "description": "用 Plan-and-Solve 理解为什么 prompt 可以显式拆成 plan 和 solve 两段。",
+ "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 / Reasoning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/plan-and-solve-prompting-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::plan9-1995",
"area": "papers",
@@ -23539,6 +24307,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
}
},
+ {
+ "id": "papers::prefix-tuning-2021",
+ "area": "papers",
+ "slug": "prefix-tuning-2021",
+ "title": "Prefix-Tuning — 不改模型，只给每层塞一段可训练前缀",
+ "description": "用 Prefix-Tuning 理解连续 prompt 如何成为参数高效微调方法。",
+ "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 / Efficient Finetuning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/prefix-tuning-2021/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::presumed-abort-1986",
"area": "papers",
@@ -23634,6 +24434,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-hci-and-software-engineering-research-01/"
}
},
+ {
+ "id": "papers::program-of-thoughts-2022",
+ "area": "papers",
+ "slug": "program-of-thoughts-2022",
+ "title": "Program of Thoughts — 让模型写程序，把计算交给解释器",
+ "description": "用 Program of Thoughts 理解自然语言推理和精确计算为什么要分工。",
+ "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 / Tool Reasoning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/program-of-thoughts-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::program-shepherding-2002",
"area": "papers",
@@ -23761,6 +24593,70 @@
"chunk_route": "/study/atlas/papers/topic-papers-compilers-and-programming-language-theory-01/"
}
},
+ {
+ "id": "papers::prompt-to-prompt-2022",
+ "area": "papers",
+ "slug": "prompt-to-prompt-2022",
+ "title": "Prompt-to-Prompt — 改词不改构图的 cross-attention 编辑",
+ "description": "用 Prompt-to-Prompt 理解扩散模型里文本 token 和图像布局如何对齐。",
+ "difficulty": "intermediate",
+ "canonical_topics": [
+ "papers-generative-models-and-diffusion"
+ ],
+ "classification": {
+ "state": "classified",
+ "source": "frontmatter-category",
+ "topic_id": "papers-generative-models-and-diffusion",
+ "matched_category": "diffusion",
+ "raw_category": "Diffusion / Editing"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/prompt-to-prompt-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-generative-models-and-diffusion-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-generative-models-and-diffusion-01/"
+ }
+ },
+ {
+ "id": "papers::prompt-tuning-2021",
+ "area": "papers",
+ "slug": "prompt-tuning-2021",
+ "title": "Prompt Tuning — 规模变大后，软提示也能接近微调",
+ "description": "用 Prompt Tuning 理解为什么 soft prompt 在大模型上突然变得有效。",
+ "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 / Efficient Finetuning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/prompt-tuning-2021/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::prototypical-networks-2017",
"area": "papers",
@@ -23920,6 +24816,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-compilers-and-programming-language-theory-01/"
}
},
+ {
+ "id": "papers::qlora-2023",
+ "area": "papers",
+ "slug": "qlora-2023",
+ "title": "QLoRA — 4-bit 量化底座上贴 LoRA 也能微调",
+ "description": "用 QLoRA 理解 NF4、double quantization 和 paged optimizers 如何降低微调门槛。",
+ "difficulty": "advanced",
+ "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 / Efficient Finetuning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/qlora-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::quantum-supremacy-2019",
"area": "papers",
@@ -25919,6 +26847,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-information-retrieval-and-recommendation-01/"
}
},
+ {
+ "id": "papers::saycan-2022",
+ "area": "papers",
+ "slug": "saycan-2022",
+ "title": "SayCan — 机器人不只问“想做什么”，还问“我能做什么”",
+ "description": "用 SayCan 理解语言模型和机器人 affordance 如何合成可执行动作。",
+ "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": "LLM Agent / Robotics"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/saycan-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::scads-database-2008",
"area": "papers",
@@ -26461,6 +27421,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
}
},
+ {
+ "id": "papers::self-instruct-2022",
+ "area": "papers",
+ "slug": "self-instruct-2022",
+ "title": "Self-Instruct — 让模型自己造指令数据再学习",
+ "description": "用 Self-Instruct 理解指令微调数据如何从少量种子任务扩展出来。",
+ "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 / Instruction Tuning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/self-instruct-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::self-pic",
"area": "papers",
@@ -28208,6 +29200,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
}
},
+ {
+ "id": "papers::speculative-decoding-2022",
+ "area": "papers",
+ "slug": "speculative-decoding-2022",
+ "title": "Speculative Decoding — 小模型先猜，大模型只验收",
+ "description": "用 Speculative Decoding 理解如何不改变分布地加速自回归生成。",
+ "difficulty": "advanced",
+ "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 / Inference"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/speculative-decoding-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::splade-2021",
"area": "papers",
@@ -28461,6 +29485,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-compilers-and-programming-language-theory-02/"
}
},
+ {
+ "id": "papers::star-self-taught-reasoner-2022",
+ "area": "papers",
+ "slug": "star-self-taught-reasoner-2022",
+ "title": "STaR — 模型先试着讲理由，再用对的理由训练自己",
+ "description": "用 STaR 理解 rationale bootstrapping 怎样减少人工推理标注。",
+ "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 / Reasoning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/star-self-taught-reasoner-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::starcoder-2023",
"area": "papers",
@@ -28967,8 +30023,8 @@
},
"route": "/study/papers/sycophancy-2023/",
"atlas": {
- "chunk_id": "topic-papers-nlp-foundations-and-scaling-01",
- "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
+ "chunk_id": "topic-papers-nlp-foundations-and-scaling-02",
+ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-02/"
}
},
{
@@ -29127,8 +30183,8 @@
},
"route": "/study/papers/t0-2021/",
"atlas": {
- "chunk_id": "topic-papers-nlp-foundations-and-scaling-01",
- "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
+ "chunk_id": "topic-papers-nlp-foundations-and-scaling-02",
+ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-02/"
}
},
{
@@ -29158,8 +30214,8 @@
},
"route": "/study/papers/t5/",
"atlas": {
- "chunk_id": "topic-papers-nlp-foundations-and-scaling-01",
- "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-01/"
+ "chunk_id": "topic-papers-nlp-foundations-and-scaling-02",
+ "chunk_route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-02/"
}
},
{
@@ -29609,6 +30665,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-operating-systems-and-cluster-management-01/"
}
},
+ {
+ "id": "papers::textual-inversion-2022",
+ "area": "papers",
+ "slug": "textual-inversion-2022",
+ "title": "Textual Inversion — 给新概念学一个专属 token",
+ "description": "用 Textual Inversion 理解冻结扩散模型时如何只学习概念 embedding。",
+ "difficulty": "intermediate",
+ "canonical_topics": [
+ "papers-generative-models-and-diffusion"
+ ],
+ "classification": {
+ "state": "classified",
+ "source": "frontmatter-category",
+ "topic_id": "papers-generative-models-and-diffusion",
+ "matched_category": "diffusion",
+ "raw_category": "Diffusion / Personalization"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/textual-inversion-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-generative-models-and-diffusion-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-generative-models-and-diffusion-01/"
+ }
+ },
{
"id": "papers::tfidf-classic",
"area": "papers",
@@ -30084,6 +31172,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
}
},
+ {
+ "id": "papers::toolllm-2023",
+ "area": "papers",
+ "slug": "toolllm-2023",
+ "title": "ToolLLM — 用 16000+ API 训练模型进入真实工具世界",
+ "description": "用 ToolLLM 理解大规模 API 数据集、工具检索和工具评测如何支撑 agent。",
+ "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 / Tool Use"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/toolllm-2023/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::tor-2004",
"area": "papers",
@@ -30116,6 +31236,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-network-protocols-01/"
}
},
+ {
+ "id": "papers::toxigen-2022",
+ "area": "papers",
+ "slug": "toxigen-2022",
+ "title": "ToxiGen — 用生成模型造隐性仇恨测试集",
+ "description": "用 ToxiGen 理解安全评测为什么要覆盖隐性、对抗性和群体相关文本。",
+ "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 / Safety Evaluation"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/toxigen-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::toy-models-superposition",
"area": "papers",
@@ -30530,6 +31682,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-information-retrieval-and-recommendation-01/"
}
},
+ {
+ "id": "papers::truthfulqa-2021",
+ "area": "papers",
+ "slug": "truthfulqa-2021",
+ "title": "TruthfulQA — 专门问模型容易学人类谬误的问题",
+ "description": "用 TruthfulQA 理解语言模型为什么会模仿常见假话而不是坚持事实。",
+ "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 / Evaluation"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/truthfulqa-2021/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::turchin-supercompilation",
"area": "papers",
@@ -30753,6 +31937,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-databases-01/"
}
},
+ {
+ "id": "papers::ul2-2022",
+ "area": "papers",
+ "slug": "ul2-2022",
+ "title": "UL2 — 一个模型同时练完补空、续写和长文本",
+ "description": "用 UL2 理解 mixture-of-denoisers 如何统一不同语言模型训练范式。",
+ "difficulty": "advanced",
+ "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 / Pretraining Objective"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "route": "/study/papers/ul2-2022/",
+ "atlas": {
+ "chunk_id": "topic-papers-agents-and-llm-systems-01",
+ "chunk_route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/"
+ }
+ },
{
"id": "papers::unified-memory-2014",
"area": "papers",
@@ -31868,6 +33084,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-databases-01/"
}
},
+ {
+ "id": "papers::webgpt-2021",
+ "area": "papers",
+ "slug": "webgpt-2021",
+ "title": "WebGPT — 让模型带着浏览器回答问题",
+ "description": "用 WebGPT 理解检索、引用和人类偏好如何组合成可追溯问答。",
+ "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": "LLM / Browser Agent"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "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/"
+ }
+ },
{
"id": "papers::websocket-rfc-6455",
"area": "papers",
@@ -32124,6 +33372,38 @@
"chunk_route": "/study/atlas/papers/topic-papers-network-protocols-01/"
}
},
+ {
+ "id": "papers::wizardlm-2023",
+ "area": "papers",
+ "slug": "wizardlm-2023",
+ "title": "WizardLM — 用 Evol-Instruct 自动变难训练题",
+ "description": "用 WizardLM 理解 instruction 数据不只要多，还要逐步变复杂。",
+ "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 / Instruction Tuning"
+ },
+ "trust": {
+ "contract_state": "v2",
+ "verification_status": "UNVERIFIED"
+ },
+ "freshness": {
+ "state": "NOT_EVALUATED",
+ "reviewed_at": "2026-07-14",
+ "review_after": null
+ },
+ "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/"
+ }
+ },
{
"id": "papers::word2vec",
"area": "papers",
@@ -32565,8 +33845,8 @@
},
"route": "/study/papers/zombie-agents-2602/",
"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/"
}
},
{
@@ -63035,9 +64315,23 @@
"en": "Agents and LLM Systems"
},
"page": 1,
- "pages": 1,
+ "pages": 2,
"route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-01/",
- "entries": 68
+ "entries": 100
+ },
+ {
+ "id": "topic-papers-agents-and-llm-systems-02",
+ "area": "papers",
+ "kind": "topic",
+ "topic_id": "papers-agents-and-llm-systems",
+ "labels": {
+ "zh": "智能体与 LLM 系统",
+ "en": "Agents and LLM Systems"
+ },
+ "page": 2,
+ "pages": 2,
+ "route": "/study/atlas/papers/topic-papers-agents-and-llm-systems-02/",
+ "entries": 1
},
{
"id": "topic-papers-ai-safety-and-interpretability-01",
@@ -63247,7 +64541,7 @@
"page": 1,
"pages": 1,
"route": "/study/atlas/papers/topic-papers-generative-models-and-diffusion-01/",
- "entries": 10
+ "entries": 14
},
{
"id": "topic-papers-hci-and-software-engineering-research-01",
@@ -63345,7 +64639,7 @@
"page": 2,
"pages": 2,
"route": "/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-02/",
- "entries": 11
+ "entries": 14
},
{
"id": "topic-papers-operating-systems-and-cluster-management-01",
diff --git a/data/review-receipts/papers/big-bench-hard-2022.json b/data/review-receipts/papers/big-bench-hard-2022.json
new file mode 100644
index 000000000..78ff52a56
--- /dev/null
+++ b/data/review-receipts/papers/big-bench-hard-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "big-bench-hard-2022",
+ "digest_sha256": "6d7d582826d9fa135f2d1c8e8603b6be560a6aa0f5f53558f6203f902d4ab2c6"
+ },
+ "source_revision": "arXiv:2210.09261v1",
+ "research_input_sha256": "66ebd2a100d7b0beda16f53dc8dd7401edc2f2874ea1cb2011c136a517e555aa",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/bigbird-2020.json b/data/review-receipts/papers/bigbird-2020.json
new file mode 100644
index 000000000..6d923eae8
--- /dev/null
+++ b/data/review-receipts/papers/bigbird-2020.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "bigbird-2020",
+ "digest_sha256": "995a4d69e73eb9f098d8f9ebf5982bccec613f573acf9ebf597b41f6409699a8"
+ },
+ "source_revision": "arXiv:2007.14062v2",
+ "research_input_sha256": "d744b073ba50ddce66b6e067b911ce4433c301048d15f1e47263de305b6ec012",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/bloom-2022.json b/data/review-receipts/papers/bloom-2022.json
new file mode 100644
index 000000000..32538f1c8
--- /dev/null
+++ b/data/review-receipts/papers/bloom-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "bloom-2022",
+ "digest_sha256": "74aa5fefd63016af4b0cc709028a3118e00cb4e7daebf1a1064435392e1d18b7"
+ },
+ "source_revision": "arXiv:2211.05100v4",
+ "research_input_sha256": "2d5b18618bc3f4dae1cc7814442b3da7f8c445270a53f3d16cc9b5e53f7c83fe",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/controlnet-2023.json b/data/review-receipts/papers/controlnet-2023.json
new file mode 100644
index 000000000..fcaa35287
--- /dev/null
+++ b/data/review-receipts/papers/controlnet-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "controlnet-2023",
+ "digest_sha256": "2c4dfbaf90df781a252aad8d6e31fbe4d20918a97a3aa204722ab1675d883113"
+ },
+ "source_revision": "arXiv:2302.05543v3",
+ "research_input_sha256": "0dfae02d64e50c17d5d50a7fa6c9cd8b05954c101e33169776b6381761781a72",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/dreambooth-2022.json b/data/review-receipts/papers/dreambooth-2022.json
new file mode 100644
index 000000000..2e6b096ed
--- /dev/null
+++ b/data/review-receipts/papers/dreambooth-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "dreambooth-2022",
+ "digest_sha256": "8224e32165149dc618fec8b5efefd918fcd01f46a69e2aeb581588aa85818c0e"
+ },
+ "source_revision": "arXiv:2208.12242v2",
+ "research_input_sha256": "26e088769607243ef0f4b5dad39c327c7c5ac4ec1c91e05a388b05d9580f8b85",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/gorilla-2023.json b/data/review-receipts/papers/gorilla-2023.json
new file mode 100644
index 000000000..90d862d8f
--- /dev/null
+++ b/data/review-receipts/papers/gorilla-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "gorilla-2023",
+ "digest_sha256": "6150edbbabffdca5905d6b32a0bd4eef0dac774a1d1e23cf716355596ff05784"
+ },
+ "source_revision": "arXiv:2305.15334v1",
+ "research_input_sha256": "b7b92dacdb5b5524c4fecf784fa9560ee61a0d18378393631841c0f57179bbbe",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/gsm8k-2021.json b/data/review-receipts/papers/gsm8k-2021.json
new file mode 100644
index 000000000..6757fb09c
--- /dev/null
+++ b/data/review-receipts/papers/gsm8k-2021.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "gsm8k-2021",
+ "digest_sha256": "ce0b4e02509476e0d2bfbdeaaf2b7f7684c44405ef374cc13bd84c2d7147e0e7"
+ },
+ "source_revision": "arXiv:2110.14168v2",
+ "research_input_sha256": "efb9a56f0ff87949a517bdb03118dece407326f31c60e04cd3a98b020ae40785",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/hugginggpt-2023.json b/data/review-receipts/papers/hugginggpt-2023.json
new file mode 100644
index 000000000..b4b995c93
--- /dev/null
+++ b/data/review-receipts/papers/hugginggpt-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "hugginggpt-2023",
+ "digest_sha256": "d276c3384e7eb0128aeddc0387bc466f2857bfe2bb24ff52569266282fc78a44"
+ },
+ "source_revision": "arXiv:2303.17580v4",
+ "research_input_sha256": "5fc26a9fb2d037e638ee0ab76db7488d659293518cd60adeb9da9b36af9d1c82",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/inner-monologue-2022.json b/data/review-receipts/papers/inner-monologue-2022.json
new file mode 100644
index 000000000..05380230c
--- /dev/null
+++ b/data/review-receipts/papers/inner-monologue-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "inner-monologue-2022",
+ "digest_sha256": "97e57acf5affaab65285d5d80ae469b5601743a3adab2f423c85dae2110ec3cb"
+ },
+ "source_revision": "arXiv:2207.05608v1",
+ "research_input_sha256": "44063aa526ff4eed240cfb43292146864657d674e00b2e923cdc6e2a2d021fdf",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/kaplan-scaling-laws-2020.json b/data/review-receipts/papers/kaplan-scaling-laws-2020.json
new file mode 100644
index 000000000..ea2866b9d
--- /dev/null
+++ b/data/review-receipts/papers/kaplan-scaling-laws-2020.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "kaplan-scaling-laws-2020",
+ "digest_sha256": "290943ad677bacccf3b829e9f3e8771eff0a5342278c65276d93fe6c89c5729b"
+ },
+ "source_revision": "arXiv:2001.08361v1",
+ "research_input_sha256": "65167231bcdfc0382012ea45cd0fe4f3c9bff8862d9f0812f9d9c5c56dcc9357",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/lamda-2022.json b/data/review-receipts/papers/lamda-2022.json
new file mode 100644
index 000000000..30355b489
--- /dev/null
+++ b/data/review-receipts/papers/lamda-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "lamda-2022",
+ "digest_sha256": "f21fb0b5917772720acaa133954545fe5677a4721c4b2dc4a8b5e81f874e9e0b"
+ },
+ "source_revision": "arXiv:2201.08239v3",
+ "research_input_sha256": "c0c685404880afee4e9e893d24915fa5ce1397e3f6a4c2ac29a4b909417bb240",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/least-to-most-prompting-2022.json b/data/review-receipts/papers/least-to-most-prompting-2022.json
new file mode 100644
index 000000000..1805337db
--- /dev/null
+++ b/data/review-receipts/papers/least-to-most-prompting-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "least-to-most-prompting-2022",
+ "digest_sha256": "5f6e7e61e72ac38a137e3a2dff081be8cf7276ef5e04bc81c6600ca12cd6be3c"
+ },
+ "source_revision": "arXiv:2205.10625v3",
+ "research_input_sha256": "5b22c3a273bcb0b7b503443adc6e4eb6d5d8df3e692e88e20edbadbe9691174c",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/linformer-2020.json b/data/review-receipts/papers/linformer-2020.json
new file mode 100644
index 000000000..1aa78f1d1
--- /dev/null
+++ b/data/review-receipts/papers/linformer-2020.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "linformer-2020",
+ "digest_sha256": "de526de470d111145d859cc69bcb2110e9f0268f22cfbdef15233cec44d2f308"
+ },
+ "source_revision": "arXiv:2006.04768v3",
+ "research_input_sha256": "3cdf8c4489a0e3edff6eef56803b19b2fca69d9111b77172fd0274bd225f8676",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/longnet-2023.json b/data/review-receipts/papers/longnet-2023.json
new file mode 100644
index 000000000..07910abd6
--- /dev/null
+++ b/data/review-receipts/papers/longnet-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "longnet-2023",
+ "digest_sha256": "bb44d774d93dd128af704f0d2152294291d694822f19bc484f7456883a7db5ac"
+ },
+ "source_revision": "arXiv:2307.02486v2",
+ "research_input_sha256": "6f18d710038430180361e1fd013e29864abe7f6a624452ff2bc9e7cf657770bd",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/minerva-2022.json b/data/review-receipts/papers/minerva-2022.json
new file mode 100644
index 000000000..978d29d8c
--- /dev/null
+++ b/data/review-receipts/papers/minerva-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "minerva-2022",
+ "digest_sha256": "dca9c2d973a768c36acff3bf33195b500832857e569d639e6884da74df34d906"
+ },
+ "source_revision": "arXiv:2206.14858v2",
+ "research_input_sha256": "981905e4676af347d3e699e5df252f824ab8b3c1a015e6680f69279eb9285f00",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/mistral-7b-2023.json b/data/review-receipts/papers/mistral-7b-2023.json
new file mode 100644
index 000000000..40b5e849a
--- /dev/null
+++ b/data/review-receipts/papers/mistral-7b-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "mistral-7b-2023",
+ "digest_sha256": "7ca28dbf4f5fda56c45882bf5d68369a13ad0785fa5b2c21a6e8c4a0f714a13d"
+ },
+ "source_revision": "arXiv:2310.06825v1",
+ "research_input_sha256": "8152a932b02ec3408b2e752a4869a4b3562167bb60576c41cdb5b0c5f6af1b0e",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/mrkl-systems-2022.json b/data/review-receipts/papers/mrkl-systems-2022.json
new file mode 100644
index 000000000..1d18d8854
--- /dev/null
+++ b/data/review-receipts/papers/mrkl-systems-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "mrkl-systems-2022",
+ "digest_sha256": "0f275a7b87ac803d9219d9d4118c9e37dbe19b27c66cf43647f2d946a646f4c4"
+ },
+ "source_revision": "arXiv:2205.00445v1",
+ "research_input_sha256": "ba86ab684e69bcf3d4795a93b4fd2048ef89bf1f6c30dea443977dcaa5ba2b16",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/natural-instructions-v2-2022.json b/data/review-receipts/papers/natural-instructions-v2-2022.json
new file mode 100644
index 000000000..2407b3440
--- /dev/null
+++ b/data/review-receipts/papers/natural-instructions-v2-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "natural-instructions-v2-2022",
+ "digest_sha256": "b2c0b67d86aef722c0d89cf12804ba7e222b8274ad8a75ff202614d90148619b"
+ },
+ "source_revision": "arXiv:2204.07705v3",
+ "research_input_sha256": "130079a0d10b62ae7cb8c4bc3089b1fee767aca2ce534913e42f705840c4682a",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/opt-2022.json b/data/review-receipts/papers/opt-2022.json
new file mode 100644
index 000000000..f8ac4dfbb
--- /dev/null
+++ b/data/review-receipts/papers/opt-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "opt-2022",
+ "digest_sha256": "869006ce0fbf9be2de2bf590cc2a737000569259ce246a332317352541ad1587"
+ },
+ "source_revision": "arXiv:2205.01068v4",
+ "research_input_sha256": "473ea8798985546dd37d15e6246f29b91edab40a619aa45972664e415e80fa7c",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/orca-explanation-tuning-2023.json b/data/review-receipts/papers/orca-explanation-tuning-2023.json
new file mode 100644
index 000000000..7469c0e22
--- /dev/null
+++ b/data/review-receipts/papers/orca-explanation-tuning-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "orca-explanation-tuning-2023",
+ "digest_sha256": "05b36217e949245ed2afe74ce10fc9a06141ff9f2e748e22c6f1ce319c33288b"
+ },
+ "source_revision": "arXiv:2306.02707v1",
+ "research_input_sha256": "d10bcd6d1110c5034dea0d701d7355e90e5297641492c0bc064d9675b9fb2d6b",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/p-tuning-v2-2021.json b/data/review-receipts/papers/p-tuning-v2-2021.json
new file mode 100644
index 000000000..735eccaad
--- /dev/null
+++ b/data/review-receipts/papers/p-tuning-v2-2021.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "p-tuning-v2-2021",
+ "digest_sha256": "7e29970dd1c54b08232dcacf0424d15144db65fddf56a9513e349bcc2d66fc61"
+ },
+ "source_revision": "arXiv:2110.07602v3",
+ "research_input_sha256": "1cded8dc23b7fc4e5e2ca05fba142265ad7d5af2230f76394fea4e7598ab84a5",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/pal-code-reasoning-2022.json b/data/review-receipts/papers/pal-code-reasoning-2022.json
new file mode 100644
index 000000000..30184aafd
--- /dev/null
+++ b/data/review-receipts/papers/pal-code-reasoning-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "pal-code-reasoning-2022",
+ "digest_sha256": "857cec22591d8950302786bf99210f37e3ce7a6b0fa364c8d4d2d4161dbae2fd"
+ },
+ "source_revision": "arXiv:2211.10435v2",
+ "research_input_sha256": "3ca6a6ca60f5c6f71eb6b9d4938a302f25bb41767256c00cfa537682a58999ab",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/palm-2022.json b/data/review-receipts/papers/palm-2022.json
new file mode 100644
index 000000000..704a5a99c
--- /dev/null
+++ b/data/review-receipts/papers/palm-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "palm-2022",
+ "digest_sha256": "4a972542ca2f8b22240eeea1dbdb0c7228e32209323f41745c614a60d7ef76da"
+ },
+ "source_revision": "arXiv:2204.02311v5",
+ "research_input_sha256": "f805c977a505eb3be69f5cf62d093a5b8612c7acb100dcf5c7c77f02c4262f71",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/plan-and-solve-prompting-2023.json b/data/review-receipts/papers/plan-and-solve-prompting-2023.json
new file mode 100644
index 000000000..3b5091478
--- /dev/null
+++ b/data/review-receipts/papers/plan-and-solve-prompting-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "plan-and-solve-prompting-2023",
+ "digest_sha256": "873a84eb76b832aab87074cce0e2cfc58e56bb070e96ac8aa68547156cfceb85"
+ },
+ "source_revision": "arXiv:2305.04091v3",
+ "research_input_sha256": "d2d4c19f25152b8a3d715afd4cce80015db445b28b4478b4f7e3be0c268393d9",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/prefix-tuning-2021.json b/data/review-receipts/papers/prefix-tuning-2021.json
new file mode 100644
index 000000000..9993c4f9d
--- /dev/null
+++ b/data/review-receipts/papers/prefix-tuning-2021.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "prefix-tuning-2021",
+ "digest_sha256": "7a79e09d4d75a26bbd5ba80ee281206d64f9c05d13049a5097022a2fac458fe8"
+ },
+ "source_revision": "arXiv:2101.00190v1",
+ "research_input_sha256": "5e1f594b5843377dc37b38b9203e2d1b735e1b43029c2176aff65db9047501ff",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/program-of-thoughts-2022.json b/data/review-receipts/papers/program-of-thoughts-2022.json
new file mode 100644
index 000000000..b63109286
--- /dev/null
+++ b/data/review-receipts/papers/program-of-thoughts-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "program-of-thoughts-2022",
+ "digest_sha256": "3a1a76391b1207890db702882aa0d616d99582f37847bfd9ef646103a59e2023"
+ },
+ "source_revision": "arXiv:2211.12588v4",
+ "research_input_sha256": "47ce46b950048baada6a41cc2ae0d641b81b7ff36e197babeb265893941ea0f2",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/prompt-to-prompt-2022.json b/data/review-receipts/papers/prompt-to-prompt-2022.json
new file mode 100644
index 000000000..855911463
--- /dev/null
+++ b/data/review-receipts/papers/prompt-to-prompt-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "prompt-to-prompt-2022",
+ "digest_sha256": "90315253f92ad99c47c2a6fa10333e472969eba5ec09481ab3f6b12361d1927b"
+ },
+ "source_revision": "arXiv:2208.01626v1",
+ "research_input_sha256": "fb544725021fffe4b3be6284890ee73dc150b5113c8b391cc1ff0660d1d50f99",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/prompt-tuning-2021.json b/data/review-receipts/papers/prompt-tuning-2021.json
new file mode 100644
index 000000000..011b12948
--- /dev/null
+++ b/data/review-receipts/papers/prompt-tuning-2021.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "prompt-tuning-2021",
+ "digest_sha256": "40305a0a2c27182badba740098a7116d259e98be14bcb9c8647496e99fc0f090"
+ },
+ "source_revision": "arXiv:2104.08691v2",
+ "research_input_sha256": "d4f45f4b93bce04cff857847dc315f698383353fb23c00b95f3ea247cfb67bf1",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/qlora-2023.json b/data/review-receipts/papers/qlora-2023.json
new file mode 100644
index 000000000..dfc614e66
--- /dev/null
+++ b/data/review-receipts/papers/qlora-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "qlora-2023",
+ "digest_sha256": "0e2421e11efb7222e959bc39379427ece730742793e14ced9c6848b41e79a6e5"
+ },
+ "source_revision": "arXiv:2305.14314v1",
+ "research_input_sha256": "02545f5e4b90d560eda2a434b1a1e2f694bce57662d17a64440fe98f35f929d8",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/saycan-2022.json b/data/review-receipts/papers/saycan-2022.json
new file mode 100644
index 000000000..62058ca7c
--- /dev/null
+++ b/data/review-receipts/papers/saycan-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "saycan-2022",
+ "digest_sha256": "f9f36b3e3771766975e1dd4871bcf99a8b188701bbcdf7cecc104dd1c554e646"
+ },
+ "source_revision": "arXiv:2204.01691v2",
+ "research_input_sha256": "6a4fb549e218f3e257706bea543d383d755b36b0bd6695cb350c91569cb34eaa",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/self-instruct-2022.json b/data/review-receipts/papers/self-instruct-2022.json
new file mode 100644
index 000000000..a0776639b
--- /dev/null
+++ b/data/review-receipts/papers/self-instruct-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "self-instruct-2022",
+ "digest_sha256": "968fb2595e4c994a2bbf9b3c56385b1b9018340ca33cb0a56e9f2114ac9fb6fe"
+ },
+ "source_revision": "arXiv:2212.10560v2",
+ "research_input_sha256": "083aa5e217b02607a74d75cbbf9871d6c6f30c3b34cef4620ac938dc665be469",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/speculative-decoding-2022.json b/data/review-receipts/papers/speculative-decoding-2022.json
new file mode 100644
index 000000000..7897de27f
--- /dev/null
+++ b/data/review-receipts/papers/speculative-decoding-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "speculative-decoding-2022",
+ "digest_sha256": "1ba16b1b21153d74fd4acccb2e9acc65e3b82ad95d0584fdd5b296e22da85741"
+ },
+ "source_revision": "arXiv:2211.17192v2",
+ "research_input_sha256": "d7e61964ae99d9b8912de0e1e166bda00dfc05c013daf4715f24c82bdcfb7c25",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/star-self-taught-reasoner-2022.json b/data/review-receipts/papers/star-self-taught-reasoner-2022.json
new file mode 100644
index 000000000..1829cc7b9
--- /dev/null
+++ b/data/review-receipts/papers/star-self-taught-reasoner-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "star-self-taught-reasoner-2022",
+ "digest_sha256": "6905615a871b43a3b1d2872b16033f56ca41e5b8ff13d5b43380975173f73b78"
+ },
+ "source_revision": "arXiv:2203.14465v2",
+ "research_input_sha256": "b96386383efe76f87d11da6ec72b86f7bf8aac6b0f699d721ff68f3f0ba97479",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/textual-inversion-2022.json b/data/review-receipts/papers/textual-inversion-2022.json
new file mode 100644
index 000000000..29e93dc15
--- /dev/null
+++ b/data/review-receipts/papers/textual-inversion-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "textual-inversion-2022",
+ "digest_sha256": "7fe4f396ad1124ecae5664b8d7942a0a6d2acf82f80fd104b5733b70622359df"
+ },
+ "source_revision": "arXiv:2208.01618v1",
+ "research_input_sha256": "15e6724d58d9ff5a0ab3591ea4727a3d79508a0a8dd7a49f906ad0fa8100e4af",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/toolllm-2023.json b/data/review-receipts/papers/toolllm-2023.json
new file mode 100644
index 000000000..9b028c386
--- /dev/null
+++ b/data/review-receipts/papers/toolllm-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "toolllm-2023",
+ "digest_sha256": "a9d7ead607a2d552ece1b642c9b261aa14583395f923c67e2c57f4d244385215"
+ },
+ "source_revision": "arXiv:2307.16789v2",
+ "research_input_sha256": "8c75cb0174d4816a8a8c321f8d0c80b769fa564438b5f2efca1bd4eaa7f4937b",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/toxigen-2022.json b/data/review-receipts/papers/toxigen-2022.json
new file mode 100644
index 000000000..e9f05d56c
--- /dev/null
+++ b/data/review-receipts/papers/toxigen-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "toxigen-2022",
+ "digest_sha256": "6864b9bb37134b3d24847401c2f5ee4d1a6491fd19703ba9c4a25ad94d100eb0"
+ },
+ "source_revision": "arXiv:2203.09509v4",
+ "research_input_sha256": "6001f226a5327b067710c905af594b2e719efe1d9843d49ce4c2e09ca30e9be5",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/truthfulqa-2021.json b/data/review-receipts/papers/truthfulqa-2021.json
new file mode 100644
index 000000000..ee04ef8b4
--- /dev/null
+++ b/data/review-receipts/papers/truthfulqa-2021.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "truthfulqa-2021",
+ "digest_sha256": "1e9137fe2aaf875b54815039fe9d53aca9f7e1d332f25d2a00a25969f239f992"
+ },
+ "source_revision": "arXiv:2109.07958v2",
+ "research_input_sha256": "31a0a9c012ace4fb2ab4cbb1ab6d210a79bd2281c1202dc9443f2a572ce11287",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/ul2-2022.json b/data/review-receipts/papers/ul2-2022.json
new file mode 100644
index 000000000..24b1d6ce9
--- /dev/null
+++ b/data/review-receipts/papers/ul2-2022.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "ul2-2022",
+ "digest_sha256": "bfdc46c05d78a872106fa71977a43b220f5e04681d9ecd60bff6e327025c6b4f"
+ },
+ "source_revision": "arXiv:2205.05131v3",
+ "research_input_sha256": "f1fcd8b4c348fef6a4739e1ff75f9d99f036512b9173021876bc93cdeef873f3",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/webgpt-2021.json b/data/review-receipts/papers/webgpt-2021.json
new file mode 100644
index 000000000..d1f558c0f
--- /dev/null
+++ b/data/review-receipts/papers/webgpt-2021.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "webgpt-2021",
+ "digest_sha256": "17eb3c519cde92d0915004103d77dbf5cda7a6cad3e9b8c6b05b85804de4dd78"
+ },
+ "source_revision": "arXiv:2112.09332v3",
+ "research_input_sha256": "0d50a33d16b8002e8ee87f736887de6299bad7cb3ab984b0d46c046becd62ed9",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/data/review-receipts/papers/wizardlm-2023.json b/data/review-receipts/papers/wizardlm-2023.json
new file mode 100644
index 000000000..4d22e12fa
--- /dev/null
+++ b/data/review-receipts/papers/wizardlm-2023.json
@@ -0,0 +1,55 @@
+{
+ "schema_version": "study-review-receipt-v1",
+ "generation": 1,
+ "predecessor_digest_sha256": null,
+ "note": {
+ "area": "papers",
+ "slug": "wizardlm-2023",
+ "digest_sha256": "cd4746f0aaeabc9658be82632cbefdc7765774d6c1a2df8f82a97aaa57f859c7"
+ },
+ "source_revision": "arXiv:2304.12244v3",
+ "research_input_sha256": "c34bd2b904d86510d666641430bc3c9ef615dd9fe1a60da2af8fc927cfcbeb3d",
+ "reviewers": [
+ {
+ "role": "ZERO_BASE",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 84,
+ "warnings": [
+ "Zero-base explanation includes an analogy and toy mechanism check; paper benchmark was not executed."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "MANUAL_SIMULATION"
+ }
+ },
+ {
+ "role": "ENGINEER",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 83,
+ "warnings": [
+ "Claims are limited to arXiv metadata and static paper reading; no repository or service integration was run."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ },
+ {
+ "role": "ACADEMIC",
+ "reviewer_version": "study-static-review-20260714-40-paper-batch",
+ "decision": "PASS_WITH_NOTES",
+ "score": 82,
+ "warnings": [
+ "Citation identity was checked through arXiv metadata; reported experimental numbers are not independently reproduced."
+ ],
+ "execution": {
+ "review_mode": "STATIC_REVIEW",
+ "code_mode": "NOT_APPLICABLE"
+ }
+ }
+ ],
+ "waivers": [],
+ "created_at": "2026-07-14T12:00:00.000Z"
+}
diff --git a/src/content/docs/about.md b/src/content/docs/about.md
index 7b83fd0a2..d0c762085 100644
--- a/src/content/docs/about.md
+++ b/src/content/docs/about.md
@@ -1,6 +1,6 @@
---
title: 关于这个站点
-description: 1900+ 篇精读笔记，按"未来工程师该懂什么"的标准筛选——给想成为 AI 时代工程师的同路人
+description: 2000+ 篇精读笔记，按"未来工程师该懂什么"的标准筛选——给想成为 AI 时代工程师的同路人
sidebar:
order: 0
label: 立场宣言
@@ -14,7 +14,7 @@ sidebar:
写到今天的硬数字：
-- **1023 篇论文笔记** + **961 篇项目笔记**，合计 **1900+ 篇**
+- **1063 篇论文笔记** + **961 篇项目笔记**，合计 **2000+ 篇**
- 横跨 19 个主题：分布式系统 76 / 编程语言 76 / 数据库 47 / 操作系统 46 / 机器学习 44 / 区块链 44 / 后端 API 40 / 基础设施 38 / 网络协议 37 / 图形学 36 / 形式化方法 27 / 通信 27 / 信息检索 25 / Agent 24 / CLI 23 / NLP 11 / 编译器 11 / 等
- 近 30 天集中产出：基础设施（444 commits）、编译器与 PL（72）、自演化 Agent（10+ 新建）、分布式（47）、区块链（44）
@@ -166,7 +166,7 @@ AI 把"写代码"这件事的成本拉到了地板。但"在多个看似都对
- **编辑**：Jason 读、提观点、要求重写、调整声音、补判断
- **基础设施**：Astro + Starlight + GitHub Pages，由 Claude Code 搭
-1900+ 篇的产能不是单作者能写出来的——它是"AI 协作把人放大十倍"的活样本。藏着这件事反而会让站点失去价值；明说出来，你才能看到这种新工作方式怎么运转。
+2000+ 篇的产能不是单作者能写出来的——它是"AI 协作把人放大十倍"的活样本。藏着这件事反而会让站点失去价值；明说出来，你才能看到这种新工作方式怎么运转。
判断力部分必须由我贡献：**为什么是这个项目而不是它的同类**——这部分 AI 替不了。
diff --git a/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md
index be122b175..5ae2471f7 100644
--- a/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md
+++ b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-01.md
@@ -1,6 +1,6 @@
---
title: "智能体与 LLM 系统 · 论文 · 第 1 组"
-description: "68 条 智能体与 LLM 系统 Atlas 分块"
+description: "100 条 智能体与 LLM 系统 Atlas 分块"
sidebar:
hidden: true
---
@@ -9,7 +9,7 @@ sidebar:
[返回论文全景索引](/study/papers-atlas/)
-本分块共 68 条，稳定上限为 100 条。
+本分块共 100 条，稳定上限为 100 条。
| 论文 | Slug | 难度 | 可信状态 | 简介 |
|---|---|---|---|---|
@@ -22,6 +22,8 @@ sidebar:
| [AutoGen — 多智能体对话框架](/study/papers/autogen/) | `autogen` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [AWQ — 看激活脸色给权重打折](/study/papers/awq/) | `awq` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [AWQ 2023 — 把 70B 大模型权重压到 35GB](/study/papers/awq-2023/) | `awq-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [BIG-Bench Hard — 从大题库里挑出模型最头疼的 23 类题](/study/papers/big-bench-hard-2022/) | `big-bench-hard-2022` | intermediate | UNVERIFIED | 用 BBH 理解为什么 benchmark 需要难题子集和 CoT 对照 |
+| [BLOOM — 把 176B 多语种模型做成开放科学工程](/study/papers/bloom-2022/) | `bloom-2022` | intermediate | UNVERIFIED | 用 BLOOM 理解大模型也可以用社区协作、数据治理和开放发布来推进 |
| [Chain-of-Thought — 让大模型先写步骤再回答](/study/papers/chain-of-thought/) | `chain-of-thought` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [ClawTrace — 把 agent 每步操作的"成本账"先算清再蒸馏](/study/papers/clawtrace-cost-aware/) | `clawtrace-cost-aware` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Code as Agent Harness — 把代码当 agent 的"骨架"来重新看 agentic AI](/study/papers/code-as-agent-harness/) | `code-as-agent-harness` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
@@ -31,10 +33,17 @@ sidebar:
| [EVE-Agent — 自我训练前先把证据钉在桌上](/study/papers/eve-agent-evidence/) | `eve-agent-evidence` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Evo-Memory — 给"会自己长记性"的 agent 出一份统一考卷](/study/papers/evo-memory-2511/) | `evo-memory-2511` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [EXG 经验图 — 把 agent 的成败拼成一张可复用的关系图](/study/papers/exg-experience-graphs/) | `exg-experience-graphs` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Gorilla — 让 LLM 学会查 API 文档再调用](/study/papers/gorilla-2023/) | `gorilla-2023` | intermediate | UNVERIFIED | 用 Gorilla 理解 API grounding 如何降低工具调用幻觉 |
| [GPTQ — 把 175B 大模型压成 4-bit 还几乎不掉点](/study/papers/gptq-2023/) | `gptq-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [GraphRAG — 微软的知识图谱 + RAG](/study/papers/graphrag/) | `graphrag` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [GSM8K — 小学数学题把大模型算术短板照出来](/study/papers/gsm8k-2021/) | `gsm8k-2021` | intermediate | UNVERIFIED | 用 GSM8K 理解数学 word problem、verifier 和采样重排为什么重要 |
+| [HuggingGPT — 让 ChatGPT 当任务调度员，模型库当工具箱](/study/papers/hugginggpt-2023/) | `hugginggpt-2023` | intermediate | UNVERIFIED | 用 HuggingGPT 理解 LLM 如何规划并调用专用模型完成多模态任务 |
+| [Inner Monologue — 让机器人把观察结果说回计划里](/study/papers/inner-monologue-2022/) | `inner-monologue-2022` | intermediate | UNVERIFIED | 用 Inner Monologue 理解闭环反馈如何让语言计划接上真实环境变化 |
| [InstructGPT — RLHF 让 LLM 听话](/study/papers/instructgpt/) | `instructgpt` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Scaling Laws — 大模型训练不是玄学，是幂律预算题](/study/papers/kaplan-scaling-laws-2020/) | `kaplan-scaling-laws-2020` | advanced | UNVERIFIED | 用 Kaplan scaling laws 理解参数、数据和计算量怎样一起决定语言模型损失 |
| [KV-Fold — 把 KV cache 当成 fold 的累加器，一段一段读长文](/study/papers/kv-fold/) | `kv-fold` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [LaMDA — 聊天模型先学会有用、具体和不乱说](/study/papers/lamda-2022/) | `lamda-2022` | intermediate | UNVERIFIED | 用 LaMDA 理解开放域对话模型为什么需要质量、安全和 groundedness 三条线 |
+| [Least-to-Most — 先拆小题，再解大题](/study/papers/least-to-most-prompting-2022/) | `least-to-most-prompting-2022` | intermediate | UNVERIFIED | 用 Least-to-Most Prompting 理解复杂推理为什么要先分解再逐步求解 |
| [LLM.int8() — 大模型激活值里藏着几个超大异常通道](/study/papers/llm-int8-2022/) | `llm-int8-2022` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [LLM-Wiki — 把外部知识编译成 agent 自己的"维基"](/study/papers/llm-wiki-retrieval-reasoning/) | `llm-wiki-retrieval-reasoning` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [MCP-Bench — 用真实 MCP Server 测 agent 工具编排](/study/papers/mcp-bench-2025/) | `mcp-bench-2025` | intermediate | UNVERIFIED | MCP-Bench 通过 28 个 live MCP server 和 250 个工具评估多步工具编排 |
@@ -44,25 +53,41 @@ sidebar:
| [MemGym — 给长程 agent memory 做一间健身房](/study/papers/memgym/) | `memgym` | intermediate | UNVERIFIED | 用 MemGym 区分聊天记忆、执行记忆和可迁移的 agent 经验 |
| [MetaGPT — 多智能体软件公司](/study/papers/metagpt/) | `metagpt` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [MIND-Skill — 用归纳和演绎双 agent 抽 skill 并保证质量](/study/papers/mind-skill/) | `mind-skill` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Minerva — 把语言模型拉进数学草稿纸](/study/papers/minerva-2022/) | `minerva-2022` | advanced | UNVERIFIED | 用 Minerva 理解为什么数学推理需要专门的数据、逐步解题和采样验证 |
| [Misevolution — 自进化 agent 也会"越改越坏"，连顶配模型也躲不过](/study/papers/misevolution-2509/) | `misevolution-2509` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Mistral 7B — 小模型靠架构细节打出性价比](/study/papers/mistral-7b-2023/) | `mistral-7b-2023` | intermediate | UNVERIFIED | 用 Mistral 7B 理解 grouped-query attention 和 sliding-window attention 如何服务高效开源模型 |
| [MMSkills — 把视觉 agent 的"操作经验"做成多模态卡片](/study/papers/mmskills-multimodal/) | `mmskills-multimodal` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [MRKL — 给大模型配一组专家工具和路由器](/study/papers/mrkl-systems-2022/) | `mrkl-systems-2022` | intermediate | UNVERIFIED | 用 MRKL Systems 理解 neuro-symbolic agent 为什么要把 LLM、检索和计算模块拆开 |
+| [Super-NaturalInstructions — 1600+ 任务教模型读懂说明书](/study/papers/natural-instructions-v2-2022/) | `natural-instructions-v2-2022` | intermediate | UNVERIFIED | 用 Super-NaturalInstructions 理解 declarative instructions 如何评测任务泛化 |
| [NestedKV — 用三层记忆决定 KV cache 该留谁](/study/papers/nestedkv/) | `nestedkv` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [OpenHands — 开源 AI 软件工程师](/study/papers/openhands/) | `openhands` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [OPT — 把 GPT-3 级训练日志打开给研究社区](/study/papers/opt-2022/) | `opt-2022` | intermediate | UNVERIFIED | 用 OPT 理解开放权重、训练日志和复现实验对 LLM 研究的重要性 |
| [Orca — Transformer 生成模型的分布式推理调度](/study/papers/orca-2022/) | `orca-2022` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Orca — 让一批 LLM 请求随到随走，不再排队等最长那个](/study/papers/orca-continuous-batching/) | `orca-continuous-batching` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Orca — 小模型不只抄答案，还学解释轨迹](/study/papers/orca-explanation-tuning-2023/) | `orca-explanation-tuning-2023` | intermediate | UNVERIFIED | 用 Orca 理解 explanation tuning 为什么比只蒸馏最终答案更像教学生 |
| [OSCAR — 离线转个方向，把 KV Cache 压到 2-bit](/study/papers/oscar-int2-kv/) | `oscar-int2-kv` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [OSWorld — 把 GUI agent 放进真正的电脑里考试](/study/papers/osworld/) | `osworld` | intermediate | UNVERIFIED | 用 OSWorld 理解为什么电脑操作 agent 不能只在网页或脚本环境里评测 |
+| [P-Tuning v2 — 把 prompt tuning 深插到每一层](/study/papers/p-tuning-v2-2021/) | `p-tuning-v2-2021` | intermediate | UNVERIFIED | 用 P-Tuning v2 理解深层连续提示为什么能跨规模和任务接近 fine-tuning |
| [PagedAttention — 把 KV cache 当虚拟内存页来管理](/study/papers/paged-attention/) | `paged-attention` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [PagedAttention — 以页替代整段内存的显存管理](/study/papers/paged-attention-vllm/) | `paged-attention-vllm` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [PAL — 让 Python 成为语言模型的草稿纸](/study/papers/pal-code-reasoning-2022/) | `pal-code-reasoning-2022` | intermediate | UNVERIFIED | 用 PAL 理解 Program-aided Language Models 如何把推理转成可运行代码 |
+| [PaLM — Pathways 把 540B LLM 扩成统一底座](/study/papers/palm-2022/) | `palm-2022` | advanced | UNVERIFIED | 用 PaLM 理解大规模 dense decoder 如何在多任务、推理和代码能力上同时冒头 |
+| [Plan-and-Solve — 零样本推理先写计划再执行](/study/papers/plan-and-solve-prompting-2023/) | `plan-and-solve-prompting-2023` | intermediate | UNVERIFIED | 用 Plan-and-Solve 理解为什么 prompt 可以显式拆成 plan 和 solve 两段 |
+| [Prefix-Tuning — 不改模型，只给每层塞一段可训练前缀](/study/papers/prefix-tuning-2021/) | `prefix-tuning-2021` | intermediate | UNVERIFIED | 用 Prefix-Tuning 理解连续 prompt 如何成为参数高效微调方法 |
+| [Program of Thoughts — 让模型写程序，把计算交给解释器](/study/papers/program-of-thoughts-2022/) | `program-of-thoughts-2022` | intermediate | UNVERIFIED | 用 Program of Thoughts 理解自然语言推理和精确计算为什么要分工 |
+| [Prompt Tuning — 规模变大后，软提示也能接近微调](/study/papers/prompt-tuning-2021/) | `prompt-tuning-2021` | intermediate | UNVERIFIED | 用 Prompt Tuning 理解为什么 soft prompt 在大模型上突然变得有效 |
+| [QLoRA — 4-bit 量化底座上贴 LoRA 也能微调](/study/papers/qlora-2023/) | `qlora-2023` | advanced | UNVERIFIED | 用 QLoRA 理解 NF4、double quantization 和 paged optimizers 如何降低微调门槛 |
| [RAG (Lewis 2020) — 检索增强生成奠基](/study/papers/rag-lewis-2020/) | `rag-lewis-2020` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [ReAct — Reasoning and Acting](/study/papers/react/) | `react` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [ReAct Agent — 推理和行动交替的工具使用范式](/study/papers/react-agent/) | `react-agent` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Reflexion — 让 LLM 自我反思](/study/papers/reflexion/) | `reflexion` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [RETRO — DeepMind 的检索增强 LLM](/study/papers/retro/) | `retro` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [SayCan — 机器人不只问“想做什么”，还问“我能做什么”](/study/papers/saycan-2022/) | `saycan-2022` | intermediate | UNVERIFIED | 用 SayCan 理解语言模型和机器人 affordance 如何合成可执行动作 |
| [Self-Consistency — 让模型把同一道题做 40 遍再投票](/study/papers/self-consistency-2022/) | `self-consistency-2022` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [自进化 AI agent 综述 — 给"会自己升级"的 agent 画一张统一地图](/study/papers/self-evolving-agents-survey/) | `self-evolving-agents-survey` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Self-Evolving RecSys — 让 LLM agent 自己跑超参实验上线](/study/papers/self-evolving-recsys-2602/) | `self-evolving-recsys-2602` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [BDI-LLM Self-Evolving Agents — 让 agent 自己改自己源代码](/study/papers/self-evolving-software-agents/) | `self-evolving-software-agents` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Self-Instruct — 让模型自己造指令数据再学习](/study/papers/self-instruct-2022/) | `self-instruct-2022` | intermediate | UNVERIFIED | 用 Self-Instruct 理解指令微调数据如何从少量种子任务扩展出来 |
| [SkCC — 给 LLM agent 写一个真正的 skill 编译器](/study/papers/skcc-skill-compiler/) | `skcc-skill-compiler` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Skill-as-Pseudocode — 把 agent 笔记本写成可校验的伪代码](/study/papers/skill-as-pseudocode/) | `skill-as-pseudocode` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Skill-Pro — 不动权重学可复用 skill 的非参数 PPO](/study/papers/skill-pro-nonparametric-ppo/) | `skill-pro-nonparametric-ppo` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
@@ -70,14 +95,23 @@ sidebar:
| [SmoothQuant 2023 — 把激活的烫手山芋扔给权重](/study/papers/smoothquant-2023/) | `smoothquant-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [SparseGPT — 175B 大模型一次过剪 50%，不重训](/study/papers/sparsegpt-2023/) | `sparsegpt-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [SpecInfer — 让大模型一次"猜一棵树"再并行验证](/study/papers/specinfer-2023/) | `specinfer-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Speculative Decoding — 小模型先猜，大模型只验收](/study/papers/speculative-decoding-2022/) | `speculative-decoding-2022` | advanced | UNVERIFIED | 用 Speculative Decoding 理解如何不改变分布地加速自回归生成 |
+| [STaR — 模型先试着讲理由，再用对的理由训练自己](/study/papers/star-self-taught-reasoner-2022/) | `star-self-taught-reasoner-2022` | intermediate | UNVERIFIED | 用 STaR 理解 rationale bootstrapping 怎样减少人工推理标注 |
| [SWE-Agent — Princeton SWE-bench 解法](/study/papers/swe-agent/) | `swe-agent` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [SWE-bench — 真实 GitHub Issue 评测](/study/papers/swe-bench/) | `swe-bench` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [SWE-Bench-CL — coding agent 不能只刷静态题](/study/papers/swe-bench-cl/) | `swe-bench-cl` | intermediate | UNVERIFIED | 用 SWE-Bench-CL 理解软件工程 agent 的持续学习、迁移和灾难性遗忘 |
| [SWE-Skills-Bench — Agent 技能真的帮得上软件工程吗](/study/papers/swe-skills-bench-2026/) | `swe-skills-bench-2026` | intermediate | UNVERIFIED | 用 paired evaluation 衡量 SWE skills 对真实软件工程 agent 的边际收益和 token 成本 |
| [ToolBench-X — 工具会坏时，agent 还能不能把事做完](/study/papers/toolbench-x/) | `toolbench-x` | intermediate | UNVERIFIED | 用 ToolBench-X 理解 tool-use benchmark 为什么要模拟规格漂移、调用错误、执行失败和结果冲突 |
| [Toolformer — 教 LLM 自主调用 API](/study/papers/toolformer/) | `toolformer` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [ToolLLM — 用 16000+ API 训练模型进入真实工具世界](/study/papers/toolllm-2023/) | `toolllm-2023` | intermediate | UNVERIFIED | 用 ToolLLM 理解大规模 API 数据集、工具检索和工具评测如何支撑 agent |
+| [ToxiGen — 用生成模型造隐性仇恨测试集](/study/papers/toxigen-2022/) | `toxigen-2022` | intermediate | UNVERIFIED | 用 ToxiGen 理解安全评测为什么要覆盖隐性、对抗性和群体相关文本 |
| [Tree of Thoughts — 让 LLM 像下棋一样多想几步再答](/study/papers/tree-of-thoughts-2023/) | `tree-of-thoughts-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [TruthfulQA — 专门问模型容易学人类谬误的问题](/study/papers/truthfulqa-2021/) | `truthfulqa-2021` | intermediate | UNVERIFIED | 用 TruthfulQA 理解语言模型为什么会模仿常见假话而不是坚持事实 |
+| [UL2 — 一个模型同时练完补空、续写和长文本](/study/papers/ul2-2022/) | `ul2-2022` | advanced | UNVERIFIED | 用 UL2 理解 mixture-of-denoisers 如何统一不同语言模型训练范式 |
| [VeriCache: Turning Lossy KV Cache into Lossless LLM Inference — 有损压缩草稿，无损输出验收](/study/papers/vericache/) | `vericache` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Voyager — LLM 终身学习智能体](/study/papers/voyager/) | `voyager` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [WebGPT — 让模型带着浏览器回答问题](/study/papers/webgpt-2021/) | `webgpt-2021` | intermediate | UNVERIFIED | 用 WebGPT 理解检索、引用和人类偏好如何组合成可追溯问答 |
| [WebXSkill — 给 Web agent 的可执行 skill 是参数化代码 + URL 图索引](/study/papers/webxskill/) | `webxskill` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
-| [Zombie Agents — 自进化 agent 的长期记忆能被持久化"借尸还魂"](/study/papers/zombie-agents-2602/) | `zombie-agents-2602` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [WizardLM — 用 Evol-Instruct 自动变难训练题](/study/papers/wizardlm-2023/) | `wizardlm-2023` | intermediate | UNVERIFIED | 用 WizardLM 理解 instruction 数据不只要多，还要逐步变复杂 |
+
+[下一组](/study/atlas/papers/topic-papers-agents-and-llm-systems-02/)
diff --git a/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-02.md b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-02.md
new file mode 100644
index 000000000..1b9b9b8af
--- /dev/null
+++ b/src/content/docs/atlas/papers/topic-papers-agents-and-llm-systems-02.md
@@ -0,0 +1,18 @@
+---
+title: "智能体与 LLM 系统 · 论文 · 第 2 组"
+description: "1 条 智能体与 LLM 系统 Atlas 分块"
+sidebar:
+ hidden: true
+---
+
+
+
+[返回论文全景索引](/study/papers-atlas/)
+
+本分块共 1 条，稳定上限为 100 条。
+
+| 论文 | Slug | 难度 | 可信状态 | 简介 |
+|---|---|---|---|---|
+| [Zombie Agents — 自进化 agent 的长期记忆能被持久化"借尸还魂"](/study/papers/zombie-agents-2602/) | `zombie-agents-2602` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+
+[上一组](/study/atlas/papers/topic-papers-agents-and-llm-systems-01/)
diff --git a/src/content/docs/atlas/papers/topic-papers-generative-models-and-diffusion-01.md b/src/content/docs/atlas/papers/topic-papers-generative-models-and-diffusion-01.md
index b42002355..9a11b25bc 100644
--- a/src/content/docs/atlas/papers/topic-papers-generative-models-and-diffusion-01.md
+++ b/src/content/docs/atlas/papers/topic-papers-generative-models-and-diffusion-01.md
@@ -1,6 +1,6 @@
---
title: "生成模型 / 扩散 · 论文 · 第 1 组"
-description: "10 条 生成模型 / 扩散 Atlas 分块"
+description: "14 条 生成模型 / 扩散 Atlas 分块"
sidebar:
hidden: true
---
@@ -9,17 +9,21 @@ sidebar:
[返回论文全景索引](/study/papers-atlas/)
-本分块共 10 条，稳定上限为 100 条。
+本分块共 14 条，稳定上限为 100 条。
| 论文 | Slug | 难度 | 可信状态 | 简介 |
|---|---|---|---|---|
+| [ControlNet — 给扩散模型加一条可控条件支路](/study/papers/controlnet-2023/) | `controlnet-2023` | intermediate | UNVERIFIED | 用 ControlNet 理解边缘、姿态和深度图如何稳定控制图像生成 |
| [DALL-E 2 — 基于 CLIP + 扩散的图像生成](/study/papers/dalle-2/) | `dalle-2` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [DDIM — 把扩散模型 1000 步采样压到 50 步](/study/papers/ddim-2020/) | `ddim-2020` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [DDPM — Denoising Diffusion Probabilistic Models](/study/papers/ddpm/) | `ddpm` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [DiT — Diffusion Transformer](/study/papers/dit/) | `dit` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [DreamBooth — 用几张图把一个新主体塞进生成模型](/study/papers/dreambooth-2022/) | `dreambooth-2022` | intermediate | UNVERIFIED | 用 DreamBooth 理解 subject-driven generation 怎样让扩散模型记住特定对象 |
| [DreamFusion — 用 2D 扩散模型当老师，把 NeRF 教成 3D](/study/papers/dreamfusion-2022/) | `dreamfusion-2022` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [EDM — 把扩散模型的训练配方一次拆清楚](/study/papers/edm-2022/) | `edm-2022` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [LLaVA — 开源多模态对话模型](/study/papers/llava/) | `llava` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Magic3D — 把 DreamFusion 的 NeRF 拆成"先粗后精"两阶段](/study/papers/magic3d-2023/) | `magic3d-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Parti — 把文生图当作翻译，用自回归 Transformer 一像素接一像素地写](/study/papers/parti-2022/) | `parti-2022` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Prompt-to-Prompt — 改词不改构图的 cross-attention 编辑](/study/papers/prompt-to-prompt-2022/) | `prompt-to-prompt-2022` | intermediate | UNVERIFIED | 用 Prompt-to-Prompt 理解扩散模型里文本 token 和图像布局如何对齐 |
| [Stable Diffusion — 开源文生图引爆](/study/papers/stable-diffusion/) | `stable-diffusion` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Textual Inversion — 给新概念学一个专属 token](/study/papers/textual-inversion-2022/) | `textual-inversion-2022` | intermediate | UNVERIFIED | 用 Textual Inversion 理解冻结扩散模型时如何只学习概念 embedding |
diff --git a/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-01.md b/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-01.md
index 4d26f8d09..2151f1d88 100644
--- a/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-01.md
+++ b/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-01.md
@@ -21,6 +21,7 @@ sidebar:
| [Attention Is All You Need](/study/papers/attention/) | `attention` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Batch Normalization — 把每层激活值规整到 0 均值 1 方差，深网训练时间砍成 1/14](/study/papers/batchnorm-2015/) | `batchnorm-2015` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [BERT — 双向 Transformer 预训练](/study/papers/bert/) | `bert` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [BigBird — 用稀疏 attention 拉长 Transformer 视野](/study/papers/bigbird-2020/) | `bigbird-2020` | advanced | UNVERIFIED | 用 BigBird 理解局部、全局和随机 attention 怎样组成长序列模式 |
| [BigGAN — 把 GAN 暴力放大到 ImageNet 512×512](/study/papers/biggan-2018/) | `biggan-2018` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [BLIP-2 — 用 188M 小桥接器把冻结的视觉模型和大语言模型拼起来](/study/papers/blip2-2023/) | `blip2-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [CCOPD — 让多轮对话别被自己的旧话带偏](/study/papers/ccopd-distillation/) | `ccopd-distillation` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
@@ -64,11 +65,13 @@ sidebar:
| [Label Smoothing — 别让模型对正确答案过度自信](/study/papers/label-smoothing-2016/) | `label-smoothing-2016` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Layer Normalization — 把归一化方向从 batch 转到 feature，让 RNN/Transformer 也能稳定训](/study/papers/layernorm-2016/) | `layernorm-2016` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Linear Attention, Still: Why Mamba-style Models Plateau](/study/papers/linear-attention-still-2026/) | `linear-attention-still-2026` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [Linformer — 把 attention 矩阵投影成线性复杂度](/study/papers/linformer-2020/) | `linformer-2020` | advanced | UNVERIFIED | 用 Linformer 理解低秩假设如何压缩 self-attention |
| [Lion — 让程序自己搜出来的优化器，比 AdamW 内存少一半](/study/papers/lion-2023/) | `lion-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [LLaMA — Meta 开源大语言模型](/study/papers/llama/) | `llama` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [LLMSurgeon — 从模型回答反推训练数据配方](/study/papers/llmsurgeon-data-mixture/) | `llmsurgeon-data-mixture` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [LoMo — 把同一句话换成图片也要看懂](/study/papers/lomo-modality/) | `lomo-modality` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Longformer — 滑窗加少数全局 token，把长文档喂进 Transformer](/study/papers/longformer-2020/) | `longformer-2020` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [LongNet — 用 dilated attention 把上下文推到十亿 token 想象空间](/study/papers/longnet-2023/) | `longnet-2023` | advanced | UNVERIFIED | 用 LongNet 理解扩张式 attention 如何在多尺度上连接超长序列 |
| [Loong DocMT — 长文档翻译里的会挑上下文的代理](/study/papers/loong-doc-mt/) | `loong-doc-mt` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [LoRA — 给冻结大模型贴低秩便签](/study/papers/lora/) | `lora` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [彩票假设 — 大网里藏着一张能独立训出来的小网](/study/papers/lottery-ticket-2019/) | `lottery-ticket-2019` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
@@ -110,8 +113,5 @@ sidebar:
| [SoundnessBench — 判断 AI 科学家会不会把坏点子当好点子](/study/papers/soundness-bench/) | `soundness-bench` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [StarCoder — 把训练数据完整公开的 15B 代码模型](/study/papers/starcoder-2023/) | `starcoder-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [StyleGAN2 — 把 StyleGAN 的水滴瑕疵和潜空间纠葛一起修掉](/study/papers/stylegan2-2020/) | `stylegan2-2020` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
-| [Sycophancy 2023 — RLHF 模型为什么爱顺着用户说](/study/papers/sycophancy-2023/) | `sycophancy-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
-| [T0 — 让 50 个人各写各的提示词，模型反而更会听新指令](/study/papers/t0-2021/) | `t0-2021` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
-| [T5 — Text-to-Text Transfer Transformer](/study/papers/t5/) | `t5` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
[下一组](/study/atlas/papers/topic-papers-nlp-foundations-and-scaling-02/)
diff --git a/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-02.md b/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-02.md
index 7ccac8dbc..f8b3f7e0d 100644
--- a/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-02.md
+++ b/src/content/docs/atlas/papers/topic-papers-nlp-foundations-and-scaling-02.md
@@ -1,6 +1,6 @@
---
title: "NLP 基础与 Scaling · 论文 · 第 2 组"
-description: "11 条 NLP 基础与 Scaling Atlas 分块"
+description: "14 条 NLP 基础与 Scaling Atlas 分块"
sidebar:
hidden: true
---
@@ -9,10 +9,13 @@ sidebar:
[返回论文全景索引](/study/papers-atlas/)
-本分块共 11 条，稳定上限为 100 条。
+本分块共 14 条，稳定上限为 100 条。
| 论文 | Slug | 难度 | 可信状态 | 简介 |
|---|---|---|---|---|
+| [Sycophancy 2023 — RLHF 模型为什么爱顺着用户说](/study/papers/sycophancy-2023/) | `sycophancy-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [T0 — 让 50 个人各写各的提示词，模型反而更会听新指令](/study/papers/t0-2021/) | `t0-2021` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
+| [T5 — Text-to-Text Transfer Transformer](/study/papers/t5/) | `t5` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [TabPFN — 一秒解决小表格分类的 Transformer](/study/papers/tabpfn-2023/) | `tabpfn-2023` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [TD3 — 给 DDPG 装两副刹车，连续控制终于稳了](/study/papers/td3-2018/) | `td3-2018` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
| [Transformer — 让每个词一次看完整句话](/study/papers/transformer/) | `transformer` | unknown | UNVERIFIED | 暂无独立描述；可先从标题与正文定位开始。 |
diff --git a/src/content/docs/career-plan.md b/src/content/docs/career-plan.md
index acb2d2eae..ca5292e40 100644
--- a/src/content/docs/career-plan.md
+++ b/src/content/docs/career-plan.md
@@ -5,7 +5,7 @@ sidebar:
order: 1
---
-> 本页是路径说明。具体笔记见左侧分组；当前规模 1900+ 篇（论文 1023 + 项目 961）。
+> 本页是路径说明。具体笔记见左侧分组；当前规模 2000+ 篇（论文 1063 + 项目 961）。
## 1. 路径模型的演化
@@ -114,7 +114,7 @@ sidebar:
## 5. 当前优势与短板
-强项（截至当前 1900+ 篇规模）：
+强项（截至当前 2000+ 篇规模）：
- 编程语言与类型理论：完整覆盖 HM / λ / Hoare 链
- 分布式共识：Paxos / Raft / Lamport 主线齐全
diff --git a/src/content/docs/index.md b/src/content/docs/index.md
index e201f8521..444efb49e 100644
--- a/src/content/docs/index.md
+++ b/src/content/docs/index.md
@@ -144,7 +144,7 @@ head:
- 当前规模：1023 篇论文 + 961 个项目 = 1984 篇笔记，按 19 个主题组织。数量已移出首屏，只作为覆盖面证据。
+ 当前规模：1063 篇论文 + 961 个项目 = 2024 篇笔记，按 19 个主题组织。数量已移出首屏，只作为覆盖面证据。
diff --git a/src/content/docs/method.md b/src/content/docs/method.md
index 0ee10015b..f146d5427 100644
--- a/src/content/docs/method.md
+++ b/src/content/docs/method.md
@@ -1,6 +1,6 @@
---
title: 怎么消化一个 GitHub 项目
-description: 7 层方法论 + 1900 篇生产实践后的回看
+description: 7 层方法论 + 2000 篇生产实践后的回看
sidebar:
order: 0
---
@@ -11,7 +11,7 @@ sidebar:
## 顶层结论（先看）
-- 这套方法跑过 **961 篇项目 + 1023 篇论文 = 1984 篇**笔记，跨 19 个一级主题、约 1984 行写作密度
+- 这套方法跑过 **961 篇项目 + 1063 篇论文 = 2024 篇**笔记，跨 19 个一级主题、约 2024 行写作密度
- 最常被跳过的层是 **Layer 4 改一处**——但每次跳过都让整篇笔记从"机制"退回"翻译"
- 真正变成"门面级"反向引用枢纽的笔记（[React](/study/projects/react/) 68 / [[pytorch]] 67 / [[kubernetes]] 66 / [[postgresql]] 66），无一例外都做过 L3+L4 双层
- L0 / L1 / L2 / L7 即使做得平庸也不致命；L3+L4 任一项缺失 = 整篇笔记掉档
@@ -238,7 +238,7 @@ sidebar:
## L4 改一处：从抽象到肌肉记忆的桥
-L4 是这套方法最被低估的一层。1900+ 篇后的具体观察：
+L4 是这套方法最被低估的一层。2000+ 篇后的具体观察：
- **改一处不是"做实验"，是"破除幻觉"**：READMEs 经常隐藏耦合点，改一行就暴露
- **3 类最高 ROI 的改动**：
diff --git a/src/content/docs/papers-atlas.md b/src/content/docs/papers-atlas.md
index 55cb70b7d..eb5279d9d 100644
--- a/src/content/docs/papers-atlas.md
+++ b/src/content/docs/papers-atlas.md
@@ -1,6 +1,6 @@
---
title: 论文全景索引
-description: 1023 篇论文的分块地图 · 稳定 taxonomy · 自动生成
+description: 1063 篇论文的分块地图 · 稳定 taxonomy · 自动生成
sidebar:
order: 5
label: 论文全景索引
@@ -12,10 +12,10 @@ sidebar:
## 总览
-
1023论文总数
-
1004已有规范主题
+
1063论文总数
+
1044已有规范主题
19暂未收纳进主题路线
-
98.1%分类覆盖率（1004 / 1023，已分类 / 总数）
+
98.2%分类覆盖率（1044 / 1063，已分类 / 总数）