@researai/deepscientist 1.5.16 → 1.6.0

package/AISB/catalog/aisb.t3.085_neuralsurv.yaml

@@ -0,0 +1,183 @@
+id: aisb.t3.085_neuralsurv
+name: 'NeuralSurv: Deep Survival Analysis with Bayesian Uncertainty Quantification'
+version: 0.1.0
+one_line: Bayesian deep survival analysis with CAVI inference, sigmoidal hazard modeling,
+  and uncertainty-aware event-time predictions.
+task_description: 'NeuralSurv is the first deep survival model to incorporate full
+  Bayesian uncertainty quantification. The framework uses a sigmoidal hazard function
+  combining a Weibull-type baseline hazard with a Bayesian neural network modulation
+  component. Inference proceeds via coordinate-ascent variational inference (CAVI)
+  with closed-form updates derived through local linearization of the BNN and Pólya-Gamma
+  data augmentation. The current packaged snapshot preserves paper metadata, method
+  documentation, and metric definitions but does not contain executable source code.
+  This entry is intended for paper-backed planning, upstream restoration, or theoretical
+  evaluation of the survival analysis route.
+
+  '
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 2-6h
+cost_band: medium
+difficulty: medium
+data_access: public
+primary_outputs:
+- harrell_c_index
+- survival_calibration_metrics
+- posterior_credible_intervals
+launch_profiles:
+- id: analysis_only
+  label: Analysis Only
+  description: Review the preserved survival-analysis benchmark metadata, metric definitions,
+    and method documentation without execution.
+- id: restore_first
+  label: Restore First
+  description: Recover the missing NeuralSurv implementation from upstream before
+    treating this as a runnable benchmark. See GitHub repository for source restoration.
+dataset_download:
+  primary_method: restoration_required
+  sources:
+  - name: GitHub Repository
+    url: https://github.com/MLGlobalHealth/neuralsurv
+    type: upstream_source
+    notes: MIT Licensed; source code available for restoration
+  - name: Synthetic Benchmarks
+    url: null
+    type: unspecified
+    notes: Reproducible synthetic data described in paper; restoration required
+  - name: Real-World Datasets
+    url: null
+    type: unspecified
+    notes: Clinical datasets referenced in experiments section; restoration required
+  notes:
+  - The current local snapshot does not preserve an executable benchmark route; dataset
+    setup is part of upstream restoration.
+  - External datasets used in experiments may require separate access arrangements;
+    consult paper for provenance.
+credential_requirements:
+  mode: none
+  items: []
+  notes: []
+resources:
+  minimum:
+    cpu_cores: 8
+    ram_gb: 32
+    disk_gb: 50
+    gpu_count: 0
+    gpu_vram_gb: 0
+    notes: CPU-only execution is plausible for theoretical analysis and metadata review.
+  recommended:
+    cpu_cores: 16
+    ram_gb: 64
+    disk_gb: 100
+    gpu_count: 1
+    gpu_vram_gb: 16
+    notes: GPU recommended when source restoration provides runnable training code;
+      CAVI with BNN inference benefits from GPU acceleration.
+environment:
+  python: '3.10'
+  cuda: '11.8'
+  pytorch: 2.1.0
+  flash_attn: null
+  key_packages:
+  - numpy
+  - scipy
+  - pytorch
+  - pandas
+  notes:
+  - The current local package does not preserve executable setup files or runnable
+    source code; environment hints here should be treated as provisional until upstream
+    restoration.
+  - Core dependencies inferred from paper methodology: PyTorch for BNN implementation,
+      NumPy/SciPy for statistical computations.
+  - CPU-only execution is plausible for the minimum route; GPU acceleration would
+    benefit variational inference iterations.
+  - Consult bundled README/requirements for full dependency set after upstream restoration.
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- blocked_no_code
+risk_notes:
+- The current snapshot does not expose executable code files that can anchor the staged
+  metrics.
+- No benchmark execution was performed in this packaging pass.
+- Treat this entry as blocked until source restoration or a richer local snapshot
+  is recovered.
+- Top-level README is absent in the current snapshot.
+recommended_when: 'Use this entry when you want paper-backed planning for the Bayesian
+  survival modeling route, upstream restoration of the NeuralSurv implementation,
+  or theoretical evaluation of the CAVI-based uncertainty quantification framework
+  for time-to-event data.
+
+  '
+not_recommended_when: 'Do not treat this local package as a ready-to-run survival
+  benchmark; the staged snapshot preserves metadata only. Do not attempt execution
+  without first restoring the upstream source code from the GitHub repository.
+
+  '
+paper:
+  title: 'NeuralSurv: Deep Survival Analysis with Bayesian Uncertainty Quantification'
+  authors:
+  - Mélodie Monod (Imperial College London)
+  - Alessandro Micheli (Imperial College London)
+  - Samir Bhatt (Imperial College London; University of Copenhagen)
+  venue: arXiv preprint
+  year: 2025
+  url: https://arxiv.org/abs/2505.11054
+  code_url: https://github.com/MLGlobalHealth/neuralsurv
+  license: MIT
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.085_neuralsurv.zip
+  archive_type: zip
+  local_dir_name: paper-85-NeuralSurv
+  notes:
+  - Download preserves paper metadata, method documentation, and metric definitions.
+  - Executable code and benchmark runner are not included in the current snapshot.
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.085_neuralsurv.zip
+  sha256: 99bb852c3f56a3596cdc9c1f5d1ae5c2941061585df48e407faa78f902aa8789
+  size_bytes: 51459
+display:
+  palette_seed: burgundy-ivory-survival
+  art_style: clinical-probability
+  accent_priority: medium
+image_path: ../image/085_aisb.t3.085_neuralsurv.jpg
+metric_definitions:
+- id: harrell_c_index
+  name: Harrell's C-index
+  type: discriminative_performance
+  description: 'Concordance index measuring the model''s ability to correctly rank
+    pairs of patients based on predicted risk scores. Higher values indicate better
+    discriminative ability. Range approximately 0.5 to 1.0; 0.5 indicates no discriminative
+    ability, 1.0 indicates perfect.
+
+    '
+  blocked: true
+  execution_note: Metric definition preserved; no executable code anchor in current
+    snapshot.
+- id: survival_calibration_metrics
+  name: Survival Calibration Metrics
+  type: calibration
+  description: 'Metrics assessing the alignment between predicted survival probabilities
+    and observed event rates. The paper emphasizes superior calibration compared to
+    state-of-the-art deep survival models as a key contribution.
+
+    '
+  blocked: true
+  execution_note: Calibration evaluation described in paper experiments; restoration
+    required.
+- id: posterior_credible_intervals
+  name: Posterior Credible Intervals
+  type: uncertainty_quantification
+  description: 'Bayesian credible intervals for individual survival functions derived
+    from the variational posterior. Communicate model confidence in predictions.
+
+    '
+  blocked: true
+  execution_note: Uncertainty quantification is a core contribution; implementation
+    restoration needed.

package/AISB/catalog/aisb.t3.085_neuralsurv.zh.yaml

@@ -0,0 +1,143 @@
+id: aisb.t3.085_neuralsurv
+name: 'NeuralSurv: Deep Survival Analysis with Bayesian Uncertainty Quantification'
+version: 0.1.0
+one_line: 采用 CAVI 推断、sigmoidal hazard 建模和不确定性感知事件时间预测的贝叶斯深度生存分析方法。
+task_description: NeuralSurv 是首个引入完整贝叶斯不确定性量化的深度生存模型。框架采用 sigmoidal hazard 函数，将 Weibull 型基线风险与贝叶斯神经网络调制项结合。推断通过坐标上升变分推断（CAVI）进行，并借助对 BNN 的局部线性化与 Pólya-Gamma 数据增广得到闭式更新。当前打包快照保留论文元数据、方法文档与指标定义，但不含可执行源码。该条目适用于基于论文的规划、上游恢复，或生存分析路线的理论评估。
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 2-6h
+cost_band: medium
+difficulty: medium
+data_access: public
+primary_outputs:
+- harrell_c_index
+- survival_calibration_metrics
+- posterior_credible_intervals
+launch_profiles:
+- id: analysis_only
+  label: Analysis Only
+  description: Review the preserved survival-analysis benchmark metadata, metric definitions, and method documentation without execution.
+- id: restore_first
+  label: Restore First
+  description: Recover the missing NeuralSurv implementation from upstream before treating this as a runnable benchmark. See GitHub repository for source restoration.
+dataset_download:
+  primary_method: restoration_required
+  sources:
+  - name: GitHub Repository
+    url: https://github.com/MLGlobalHealth/neuralsurv
+    type: upstream_source
+    notes: MIT Licensed; source code available for restoration
+  - name: Synthetic Benchmarks
+    url: null
+    type: unspecified
+    notes: Reproducible synthetic data described in paper; restoration required
+  - name: Real-World Datasets
+    url: null
+    type: unspecified
+    notes: Clinical datasets referenced in experiments section; restoration required
+  notes:
+  - The current local snapshot does not preserve an executable benchmark route; dataset setup is part of upstream restoration.
+  - External datasets used in experiments may require separate access arrangements; consult paper for provenance.
+credential_requirements:
+  mode: none
+  items: []
+  notes: []
+resources:
+  minimum:
+    cpu_cores: 8
+    ram_gb: 32
+    disk_gb: 50
+    gpu_count: 0
+    gpu_vram_gb: 0
+    notes: CPU-only execution is plausible for theoretical analysis and metadata review.
+  recommended:
+    cpu_cores: 16
+    ram_gb: 64
+    disk_gb: 100
+    gpu_count: 1
+    gpu_vram_gb: 16
+    notes: GPU recommended when source restoration provides runnable training code; CAVI with BNN inference benefits from GPU acceleration.
+environment:
+  python: '3.10'
+  cuda: '11.8'
+  pytorch: 2.1.0
+  flash_attn: null
+  key_packages:
+  - numpy
+  - scipy
+  - pytorch
+  - pandas
+  notes:
+  - The current local package does not preserve executable setup files or runnable source code; environment hints here should be treated as provisional until upstream restoration.
+  - Core dependencies inferred from paper methodology: PyTorch for BNN implementation, NumPy/SciPy for statistical computations.
+  - CPU-only execution is plausible for the minimum route; GPU acceleration would benefit variational inference iterations.
+  - Consult bundled README/requirements for full dependency set after upstream restoration.
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- blocked_no_code
+risk_notes:
+- The current snapshot does not expose executable code files that can anchor the staged metrics.
+- No benchmark execution was performed in this packaging pass.
+- Treat this entry as blocked until source restoration or a richer local snapshot is recovered.
+- Top-level README is absent in the current snapshot.
+recommended_when: 当你需要对贝叶斯生存建模路线进行基于论文的规划、恢复 NeuralSurv 实现，或评估基于 CAVI 的生存不确定性量化框架时可使用本条目。
+not_recommended_when: 不要将当前本地包视为可直接运行的生存分析基准；当前分阶段快照仅有元数据。恢复上游源码前不应尝试执行。
+paper:
+  title: 'NeuralSurv: Deep Survival Analysis with Bayesian Uncertainty Quantification'
+  authors:
+  - Mélodie Monod (Imperial College London)
+  - Alessandro Micheli (Imperial College London)
+  - Samir Bhatt (Imperial College London; University of Copenhagen)
+  venue: arXiv preprint
+  year: 2025
+  url: https://arxiv.org/abs/2505.11054
+  code_url: https://github.com/MLGlobalHealth/neuralsurv
+  license: MIT
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.085_neuralsurv.zip
+  archive_type: zip
+  local_dir_name: paper-85-NeuralSurv
+  notes:
+  - Download preserves paper metadata, method documentation, and metric definitions.
+  - Executable code and benchmark runner are not included in the current snapshot.
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.085_neuralsurv.zip
+  sha256: 99bb852c3f56a3596cdc9c1f5d1ae5c2941061585df48e407faa78f902aa8789
+  size_bytes: 51459
+display:
+  palette_seed: burgundy-ivory-survival
+  art_style: clinical-probability
+  accent_priority: medium
+image_path: ../image/085_aisb.t3.085_neuralsurv.jpg
+metric_definitions:
+- id: harrell_c_index
+  name: Harrell's C-index
+  type: discriminative_performance
+  description: 'Concordance index measuring the model''s ability to correctly rank pairs of patients based on predicted risk scores. Higher values indicate better discriminative ability. Range approximately 0.5 to 1.0; 0.5 indicates no discriminative ability, 1.0 indicates perfect.
+
+    '
+  blocked: true
+  execution_note: Metric definition preserved; no executable code anchor in current snapshot.
+- id: survival_calibration_metrics
+  name: Survival Calibration Metrics
+  type: calibration
+  description: 'Metrics assessing the alignment between predicted survival probabilities and observed event rates. The paper emphasizes superior calibration compared to state-of-the-art deep survival models as a key contribution.
+
+    '
+  blocked: true
+  execution_note: Calibration evaluation described in paper experiments; restoration required.
+- id: posterior_credible_intervals
+  name: Posterior Credible Intervals
+  type: uncertainty_quantification
+  description: 'Bayesian credible intervals for individual survival functions derived from the variational posterior. Communicate model confidence in predictions.
+
+    '
+  blocked: true
+  execution_note: Uncertainty quantification is a core contribution; implementation restoration needed.

package/AISB/catalog/aisb.t3.086_stella.yaml

@@ -0,0 +1,197 @@
+id: aisb.t3.086_stella
+name: 'STELLA: Spatial-Temporal Knowledge Embedded Lightweight Model for Atmospheric
+  Time Series Forecasting'
+version: 0.1.0
+one_line: Lightweight atmospheric time-series forecasting via spatial-temporal position
+  embeddings and MLP, challenging Transformer-complexity assumptions.
+task_description: 'This packaged benchmark covers STELLA, a lightweight atmospheric
+  time-series forecasting model that replaces Transformer attention stacks with spatial-temporal
+  position embeddings (STPE) combined with a simple MLP. The core insight is that
+  geographical coordinates and temporal features—intrinsically linked to atmospheric
+  dynamics—are sufficient to capture spatial-temporal correlations without heavy attention
+  mechanisms. The current local snapshot preserves paper metadata and references but
+  lacks executable training and evaluation code; full restoration from the upstream
+  GitHub repository is required before benchmark execution is possible.
+
+  '
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 2-6h
+cost_band: medium
+difficulty: medium
+data_access: public
+primary_outputs:
+- test_mae
+- test_rmse
+launch_profiles:
+- id: analysis_only
+  label: Analysis Only
+  description: Review preserved atmospheric forecasting benchmark metadata, paper
+    references, and architecture documentation without executing code.
+- id: restore_first
+  label: Restore Then Evaluate
+  description: Restore STELLA source implementation from upstream GitHub repository
+    before attempting training or evaluation runs. Code restoration from https://github.com/GestaltCogTeam/STELLA
+    is a prerequisite.
+dataset_download:
+  primary_method: pending_restoration
+  sources:
+  - type: upstream_repository
+    url: https://github.com/GestaltCogTeam/STELLA
+    notes: Source code and datasets referenced in paper require restoration from upstream;
+      not bundled in current snapshot.
+  - type: paper_reference
+    url: https://arxiv.org/abs/2408.09695
+    notes: Full paper with dataset descriptions for GlobalWind, PEMS-BAY, and three
+      additional atmospheric datasets.
+  notes:
+  - Current local snapshot does not preserve executable benchmark route; dataset setup
+    is part of upstream restoration.
+  - Five datasets referenced: one synthetic and four real-world atmospheric measurement
+      collections from weather station networks.
+  - Treat dataset acquisition as blocked until source restoration completes.
+credential_requirements:
+  mode: none
+  items: []
+  notes:
+  - Public benchmark; no authentication required for paper access.
+  - GitHub access required for source restoration phase.
+resources:
+  minimum:
+    cpu_cores: 8
+    ram_gb: 32
+    disk_gb: 80
+    gpu_count: 1
+    gpu_vram_gb: 12
+    gpu_arch: pascal+
+    network: required_for_restore
+  recommended:
+    cpu_cores: 16
+    ram_gb: 64
+    disk_gb: 150
+    gpu_count: 1
+    gpu_vram_gb: 24
+    gpu_arch: ampere+
+    network: required_for_restore
+environment:
+  python: '3.9'
+  cuda: '11.1'
+  pytorch: 1.10.0
+  flash_attn: null
+  key_packages:
+  - pytorch
+  - numpy
+  - pandas
+  - scipy
+  notes:
+  - Environment specifications are provisional; derived from paper dependencies and
+    not anchored by preserved setup files.
+  - Current local snapshot lacks executable environment configuration; treat Python/pytorch/cuda
+    hints as reference for restoration target.
+  - Confirm exact dependency versions against upstream requirements.txt after restoration.
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- no_execution_anchor
+risk_notes:
+- The current snapshot preserves no executable code files; both primary metrics (test_mae,
+  test_rmse) are blocked-no-code.
+- No benchmark execution was performed in this packaging pass.
+- Top-level README absent in current snapshot; restoration from GitHub required.
+- Code-bearing roots: none detected in current snapshot.
+- Static code-audit status: blocked-no-code.
+- Prefer existing code files over README-only claims when restoring evaluation protocol.
+recommended_when: 'Use this entry for paper-backed analysis and planning of the STELLA
+  atmospheric forecasting route, or when evaluating lightweight model architectures
+  for spatial-temporal prediction tasks that challenge Transformer-complexity assumptions.
+  Appropriate for pre-execution analysis when the source restoration workflow is the
+  next action.
+
+  '
+not_recommended_when: 'Do not treat this local package as a ready-to-run forecasting
+  benchmark. Do not attempt benchmark execution without first completing source restoration
+  from the upstream GitHub repository. The executable training and evaluation code
+  is absent from the staged snapshot; execution attempts will fail.
+
+  '
+paper:
+  title: 'On the Integration of Spatial-Temporal Knowledge: A Lightweight Approach
+    to Atmospheric Time Series Forecasting'
+  venue: NeurIPS 2025
+  year: 2025
+  url: https://arxiv.org/abs/2408.09695
+  github: https://github.com/GestaltCogTeam/STELLA
+  authors:
+  - Yisong Fu
+  - Fei Wang
+  - Zezhi Shao
+  - Boyu Diao
+  - Lin Wu
+  - Zhulin An
+  - Chengqing Yu
+  - Yujie Li
+  - Yongjun Xu
+  affiliations:
+  - State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese
+    Academy of Sciences
+  - University of Chinese Academy of Sciences
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.086_stella.zip
+  archive_type: zip
+  local_dir_name: paper-86-STELLA
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.086_stella.zip
+  sha256: ac2af82678fd2caab746e97744911de757709b1bc85d89ff8682b9f0d25548ed
+  size_bytes: 32873
+display:
+  palette_seed: sky-gold-atmosphere
+  art_style: climate-signal
+  accent_priority: high
+image_path: ../image/086_aisb.t3.086_stella.jpg
+capability_tags:
+- research_code_optimization
+- atmospheric_forecasting
+- time_series_forecasting
+- lightweight_models
+- spatial_temporal_modeling
+track_fit:
+- paper_track
+- benchmark_track
+aisb_direction: T3
+operational_notes:
+- name: restoration_required
+  description: 'Source code and evaluation pipelines must be restored from https://github.com/GestaltCogTeam/STELLA
+    before any execution. Current snapshot contains no executable anchors.
+
+    '
+- name: metric_contract
+  description: 'Metrics test_mae and test_rmse are declared in the catalog but have
+    no executable code anchor in the current snapshot. Mark metrics as provisional
+    until trusted execution outputs exist.
+
+    '
+- name: architecture_summary
+  description: 'STELLA uses spatial-temporal position embedding (STPE) combined with
+    MLP layers, replacing Transformer attention mechanisms. Model size is approximately
+    10k parameters. Training time reported as approximately 1 hour on single GPU in
+    paper claims.
+
+    '
+- name: dataset_summary
+  description: 'Paper evaluates on five datasets including GlobalWind and PEMS-BAY
+    for atmospheric variable prediction across N weather station networks. Exact dataset
+    files require restoration.
+
+    '
+recovery_rule: 'Prefer existing code files over README-only claims when choosing origin_path,
+  source_ref, and evaluation_protocol.code_paths. Keep metrics_summary empty until
+  trusted execution outputs exist. If the current snapshot has no executable anchor
+  for a metric, mark that metric as provisional or blocked rather than claiming support.
+
+  '

package/AISB/catalog/aisb.t3.086_stella.zh.yaml

@@ -0,0 +1,142 @@
+id: aisb.t3.086_stella
+name: STELLA：用于大气时间序列预测的时空知识嵌入式轻量级模型
+version: 0.1.0
+one_line: 通过时空位置嵌入和 MLP 进行轻量级大气时间序列预测，挑战 Transformer 复杂性假设。
+task_description: 这个打包的基准测试涵盖了 STELLA，这是一种轻量级大气时间序列预测模型，它用时空位置嵌入 (STPE) 与简单的 MLP 相结合取代了 Transformer 注意力堆栈。核心见解是地理坐标和时间特征（与大气动力学本质上相关）足以捕获时空相关性，而无需大量关注机制。当前的本地快照保留了论文元数据和参考文献，但缺乏可执行的培训和评估代码；在执行基准测试之前，需要从上游 GitHub 存储库进行完全恢复。
+task_mode: analysis_driven
+requires_execution: true
+requires_paper: true
+integrity_level: cas_plus_canary
+snapshot_status: restore_needed
+support_level: recovery
+time_band: 2-6h
+cost_band: medium
+difficulty: medium
+data_access: public
+primary_outputs:
+- test_mae
+- test_rmse
+launch_profiles:
+- id: analysis_only
+  label: 仅分析
+  description: 在不执行代码的情况下查看保存的大气预报基准元数据、论文参考文献和架构文档。
+- id: restore_first
+  label: 恢复然后评估
+  description: 在尝试训练或评估运行之前，从上游 GitHub 存储库恢复 STELLA 源实现。从 https://github.com/GestaltCogTeam/STELLA 恢复代码是先决条件。
+dataset_download:
+  primary_method: pending_restoration
+  sources:
+  - type: upstream_repository
+    url: https://github.com/GestaltCogTeam/STELLA
+    notes: Source code and datasets referenced in paper require restoration from upstream; not bundled in current snapshot.
+  - type: paper_reference
+    url: https://arxiv.org/abs/2408.09695
+    notes: Full paper with dataset descriptions for GlobalWind, PEMS-BAY, and three additional atmospheric datasets.
+  notes:
+  - 当前本地快照不保留可执行的基准路由；数据集设置是上游恢复的一部分。
+  - Five datasets referenced: 来自气象站网络的一份合成数据和四份真实世界大气测量数据集。
+  - 将数据集获取视为已阻止，直到源恢复完成。
+credential_requirements:
+  mode: none
+  items: []
+  notes:
+  - 公共基准；纸质访问无需身份验证。
+  - 源恢复阶段需要 GitHub 访问权限。
+resources:
+  minimum:
+    cpu_cores: 8
+    ram_gb: 32
+    disk_gb: 80
+    gpu_count: 1
+    gpu_vram_gb: 12
+    gpu_arch: pascal+
+    network: required_for_restore
+  recommended:
+    cpu_cores: 16
+    ram_gb: 64
+    disk_gb: 150
+    gpu_count: 1
+    gpu_vram_gb: 24
+    gpu_arch: ampere+
+    network: required_for_restore
+environment:
+  python: '3.9'
+  cuda: '11.1'
+  pytorch: 1.10.0
+  flash_attn: null
+  key_packages:
+  - pytorch
+  - numpy
+  - pandas
+  - scipy
+  notes:
+  - 环境规范是暂定的；源自纸张依赖性，并且不由保留的设置文件锚定。
+  - 当前本地快照缺少可执行环境配置；将 Python/pytorch/cuda 提示作为恢复目标的参考。
+  - 恢复后确认上游 requirements.txt 的确切依赖版本。
+risk_flags:
+- source_snapshot_incomplete
+- route_caveat
+- no_execution_anchor
+risk_notes:
+- 当前快照不保留任何可执行代码文件；两个主要指标（test_mae、test_rmse）都是无代码阻止的。
+- 在此打包过程中未执行任何基准测试。
+- 当前快照中缺少顶级自述文件；需要从 GitHub 恢复。
+- Code-bearing roots: 当前快照中未检测到任何内容。
+- Static code-audit status: 阻止无代码。
+- 恢复评估协议时，优先选择现有代码文件而不是仅自述文件声明。
+recommended_when: 使用此条目进行 STELLA 大气预报路线的纸质分析和规划，或者在评估挑战 Transformer 复杂性假设的时空预测任务的轻量级模型架构时。当源恢复工作流程是下一个操作时，适用于执行前分析。
+not_recommended_when: 不要将此本地包视为准备运行的预测基准。在未先从上游 GitHub 存储库完成源恢复之前，请勿尝试执行基准测试。分阶段快照中缺少可执行的训练和评估代码；执行尝试将会失败。
+paper:
+  title: 'On the Integration of Spatial-Temporal Knowledge: A Lightweight Approach to Atmospheric Time Series Forecasting'
+  venue: NeurIPS 2025
+  year: 2025
+  url: https://arxiv.org/abs/2408.09695
+  github: https://github.com/GestaltCogTeam/STELLA
+  authors:
+  - Yisong Fu
+  - Fei Wang
+  - Zezhi Shao
+  - Boyu Diao
+  - Lin Wu
+  - Zhulin An
+  - Chengqing Yu
+  - Yujie Li
+  - Yongjun Xu
+  affiliations:
+  - State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences
+  - University of Chinese Academy of Sciences
+download:
+  url: https://github.com/ResearAI/DeepScientist/releases/download/aisb-v0.0.1/aisb.t3.086_stella.zip
+  archive_type: zip
+  local_dir_name: paper-86-STELLA
+  provider: github_release
+  repo: ResearAI/DeepScientist
+  tag: aisb-v0.0.1
+  asset_name: aisb.t3.086_stella.zip
+  sha256: ac2af82678fd2caab746e97744911de757709b1bc85d89ff8682b9f0d25548ed
+  size_bytes: 32873
+display:
+  palette_seed: sky-gold-atmosphere
+  art_style: climate-signal
+  accent_priority: high
+image_path: ../image/086_aisb.t3.086_stella.jpg
+capability_tags:
+- research_code_optimization
+- atmospheric_forecasting
+- time_series_forecasting
+- lightweight_models
+- spatial_temporal_modeling
+track_fit:
+- paper_track
+- benchmark_track
+aisb_direction: T3
+operational_notes:
+- name: restoration_required
+  description: 在执行任何操作之前，必须从 https://github.com/GestaltCogTeam/STELLA 恢复源代码和评估管道。当前快照不包含可执行锚点。
+- name: metric_contract
+  description: Error 500 (Server Error)!!1500.That’s an error.There was an error. Please try again later.That’s all we know.
+- name: architecture_summary
+  description: STELLA 使用时空位置嵌入（STPE）与 MLP 层相结合，取代了 Transformer 注意力机制。模型大小约为 10k 个参数。论文声称，在单个 GPU 上的训练时间约为 1 小时。
+- name: dataset_summary
+  description: 论文对包括 GlobalWind 和 PEMS-BAY 在内的五个数据集进行了评估，用于 N 个气象站网络的大气变量预测。精确的数据集文件需要恢复。
+recovery_rule: 选择 origin_path、source_ref 和 evaluation_protocol.code_paths 时，优先选择现有代码文件而不是仅自述文件声明。保持 metrics_summary 为空，直到存在可信执行输出。如果当前快照没有某个指标的可执行锚点，请将该指标标记为临时或已阻止，而不是声明支持。