# 论文元数据与证据升级阻断台账

本台账继续补全证据链，记录每个论文资源的已核验维度、缺失维度和升级阻断条件。

## 论文元数据

### molot_system_card · MOLOT System Card: Malicious Operational Logic Observation Transformer

- Primary URL: https://arxiv.org/abs/2606.07792
- arXiv ID: 2606.07792
- Directness: `direct_system_card`
- Promotion level: `L2_direct_public_evidence`

已核验维度：
- SAST setting
- static call graph
- behavior sequence
- SHAP source-location mapping
- Python/JavaScript package evaluation
- product constraints

缺失维度：
- independent Open Malicious-Code Bench URL
- full benchmark table extraction
- runtime/memory/FPR numeric extraction

### cerebro_behavior_sequence · Malicious Package Detection in NPM and PyPI using a Single Model of Malicious Behavior Sequence

- Primary URL: https://arxiv.org/abs/2309.02637
- arXiv ID: 2309.02637
- Directness: `method_background`
- Promotion level: `L1_public_method_background`

已核验维度：
- NPM/PyPI malicious package detection
- high-level malicious behavior abstraction
- behavior sequence
- BERT fine-tuning

缺失维度：
- mapping table to MOLOT activity taxonomy
- replication artifact URL

### python_malware_benchmark · A Benchmark Comparison of Python Malware Detection Approaches

- Primary URL: https://arxiv.org/abs/2209.13288
- arXiv ID: 2209.13288
- Directness: `baseline_benchmark`
- Promotion level: `L2_direct_baseline_evidence`

已核验维度：
- PyPI malware benchmark
- PyPI checks / Bandit4Mal / OSSGadget comparison
- false-positive constraint

缺失维度：
- exact benchmark dataset URL
- table extraction into structured rows

### npm_malicious_detection_benchmark · Understanding NPM Malicious Package Detection: A Benchmark-Driven Empirical Analysis

- Primary URL: https://arxiv.org/abs/2603.27549
- arXiv ID: 2603.27549
- Directness: `ecosystem_benchmark`
- Promotion level: `L2_direct_benchmark_evidence`

已核验维度：
- NPM malicious/benign package dataset
- 11 behavior categories
- 8 evasion techniques
- 8 tools / 13 variants
- behavioral-chain finding

缺失维度：
- public dataset/framework repository URL
- extract tool-combination table

### oss_malicious_package_analysis · OSS Malicious Package Analysis in the Wild

- Primary URL: https://arxiv.org/abs/2404.04991
- arXiv ID: 2404.04991
- Directness: `dataset_lifecycle_analysis`
- Promotion level: `L2_direct_dataset_evidence`

已核验维度：
- 23,425 malicious packages
- knowledge graph
- source overlap issue
- lifecycle changing-release-detection-removal

缺失维度：
- dataset download URL
- schema fields extraction

### backstabbers_knife_collection · Backstabber's Knife Collection

- Primary URL: https://arxiv.org/abs/2005.09535
- arXiv ID: 2005.09535
- Directness: `supply_chain_taxonomy`
- Promotion level: `L1_public_supply_chain_taxonomy`

已核验维度：
- 174 malicious packages
- npm/PyPI/RubyGems
- attack trees
- execution timing

缺失维度：
- machine-readable dataset URL
- execution phase mapping

### unveiling_malicious_logic · Unveiling Malicious Logic

- Primary URL: https://arxiv.org/abs/2512.12559
- arXiv ID: 2512.12559
- Directness: `statement_level_taxonomy`
- Promotion level: `L2_direct_statement_level_evidence`

已核验维度：
- statement-level dataset
- 370 malicious Python packages
- 47 indicators / 7 types
- sequential pattern mining

缺失维度：
- indicator taxonomy extraction
- dataset artifact URL

### llm_pypi_malicious_detection · Detecting Malicious Source Code in PyPI Packages with LLMs

- Primary URL: https://arxiv.org/abs/2504.13769
- arXiv ID: 2504.13769
- Directness: `llm_labeling_reference`
- Promotion level: `L1_public_llm_labeling_evidence`

已核验维度：
- LLM/RAG/few-shot PyPI detection evaluation
- YARA/GitHub Security Advisories/snippet integration
- RAG modest, few-shot stronger

缺失维度：
- prompt artifact extraction
- dataset/code URL

### pdg_holmes_graph_siamese · Modeling Functional Similarity in Source Code with Graph-Based Siamese Networks

- Primary URL: https://arxiv.org/abs/2011.11228
- arXiv ID: 2011.11228
- Directness: `graph_clone_background`
- Promotion level: `L1_public_graph_clone_background`

已核验维度：
- program dependency graphs
- semantic clone detection
- graph-based Siamese HOLMES
- clone benchmark evaluation

缺失维度：
- HOLMES artifact URL
- relation to FCD table

### source_code_is_graph · Source Code is a Graph, Not a Sequence

- Primary URL: https://arxiv.org/abs/2312.16488
- arXiv ID: 2312.16488
- Directness: `graph_vs_sequence_clone_background`
- Promotion level: `L1_public_graph_clone_background`

已核验维度：
- graph-based code representation
- CodeGraph vs CodeBERT
- cross-lingual clone detection

缺失维度：
- benchmark artifact URLs
- relation to binary clone detection

### shap_unified_approach · A Unified Approach to Interpreting Model Predictions

- Primary URL: https://arxiv.org/abs/1705.07874
- arXiv ID: 1705.07874
- Directness: `explainability_foundation`
- Promotion level: `L1_public_explainability_foundation`

已核验维度：
- SHAP framework
- feature attribution
- additive feature importance

缺失维度：
- code-localization mapping, supplied by MOLOT not SHAP itself

### fcd_pdg_performance_unresolved · FCD / PDG C/C++ Performance Table from User JSON

- Primary URL: user_provided_json
- arXiv ID: None
- Directness: `unresolved_user_claim`
- Promotion level: `L0_user_input_only`

已核验维度：
- none_public

缺失维度：
- original paper/report URL
- definition of FCD
- dataset/task setup
- relation to PDG/HOLMES/CodeGraph/MOLOT

## 升级阻断条件

### block_fcd_pdg_table_upgrade

- Claim: FCD/PDG C/C++ table precision/recall/RMSE/FCD speed is verified.
- Status: `blocked`
- Reason: No direct public original source URL was found for the exact table/project-version rows.
- Allowed wording: 待核验用户线索；可能属于源码/二进制相似片段检测方向。

Required evidence:
- original paper/report URL
- table screenshot or text in source
- definition of FCD
- task/dataset description
- author/system linkage

Currently available:
- user_provided_json
- PDG/HOLMES graph-clone background
- CodeGraph clone-detection background

### block_open_malicious_code_bench_artifact_upgrade

- Claim: Open Malicious-Code Bench independent artifact URL is verified.
- Status: `blocked`
- Reason: Search found the MOLOT paper statement, but no stable independent project/GitHub URL in this iteration.
- Allowed wording: 论文声明释放 Open Malicious-Code Bench；独立 artifact URL 待补。

Required evidence:
- independent repository/project URL
- benchmark schema or README
- download/access instructions

Currently available:
- MOLOT arXiv statement that the benchmark is released

### block_agent_gate_performance_upgrade

- Claim: LRI Agent Gate has verified precision/latency/FPR.
- Status: `blocked`
- Reason: No LRI replayable tests, audit logs, latency or false-positive measurements have been produced yet.
- Allowed wording: 工程设计候选，等待 MVP 测试验证。

Required evidence:
- MVP test execution logs
- scanner version
- sample set
- false positive/false negative review
- latency/memory measurement

Currently available:
- security_gate_contract
- mvp_test_cases
- reviewer_evidence_schema