Datasets:

jang1563
/

causalatlas-move1

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
primary_delta: double
B_boot: int64
combo: struct<combo_in_train: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<ite (... 961 chars omitted)
  child 0, combo_in_train: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<item: double>, ADD_minus_G (... 93 chars omitted)
      child 0, nP: int64
      child 1, nE: int64
      child 2, GEARS_acc: list<item: double>
          child 0, item: double
      child 3, ADD_acc: list<item: double>
          child 0, item: double
      child 4, ADD_minus_GEARS_acc: list<item: double>
          child 0, item: double
      child 5, GEARS_AUROC: list<item: double>
          child 0, item: double
      child 6, ADD_AUROC: list<item: double>
          child 0, item: double
  child 1, combo_seen1: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<item: double>, ADD_minus_G (... 93 chars omitted)
      child 0, nP: int64
      child 1, nE: int64
      child 2, GEARS_acc: list<item: double>
          child 0, item: double
      child 3, ADD_acc: list<item: double>
          child 0, item: double
      child 4, ADD_minus_GEARS_acc: list<item: double>
          child 0, item: double
      child 5, GEARS_AUROC: list<item: double>
          child 0, item: double
      child 6, ADD_AUROC: list<item: double>
          child 0, item: double
  child 2, combo_seen0: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<item: double>, ADD_minus_G (... 93 chars omit
...
ouble
          child 9, uniform_mapping_ci95: list<item: double>
              child 0, item: double
          child 10, hit_discovery_go_predicted_recall: double
          child 11, hit_discovery_random_recall: double
          child 12, hit_discovery_recall_ratio: double
      child 6, evidence_files: list<item: string>
          child 0, item: string
      child 7, code_files: list<item: string>
          child 0, item: string
      child 8, linked_benchmark: string
      child 9, linked_dataset: string
quality_upgrade_plan: list<item: null>
  child 0, item: null
schema_files: list<item: struct<path: string, type: string, description: string>>
  child 0, item: struct<path: string, type: string, description: string>
      child 0, path: string
      child 1, type: string
      child 2, description: string
project: struct<name: string, repository: string, default_branch: string, license: string, description: strin (... 2 chars omitted)
  child 0, name: string
  child 1, repository: string
  child 2, default_branch: string
  child 3, license: string
  child 4, description: string
result_artifacts: list<item: struct<path: string, type: string, schema: string, strict_json: bool, description: string (... 2 chars omitted)
  child 0, item: struct<path: string, type: string, schema: string, strict_json: bool, description: string>
      child 0, path: string
      child 1, type: string
      child 2, schema: string
      child 3, strict_json: bool
      child 4, description: string
to
{'schema_version': Value('string'), 'generated_date': Value('timestamp[s]'), 'project': {'name': Value('string'), 'repository': Value('string'), 'default_branch': Value('string'), 'license': Value('string'), 'description': Value('string')}, 'release': {'version': Value('string'), 'status': Value('string'), 'base_release': Value('string'), 'base_release_url': Value('string'), 'source_repository': Value('string')}, 'linked_public_artifacts': {'source_repository': Value('string'), 'verify_or_trust_repository': Value('string'), 'verify_or_trust_dataset': Value('string'), 'grounding_atlas_repository': Value('string')}, 'entrypoints': {'dataset_card': Value('string'), 'main_report': Value('string'), 'claim_map': Value('string'), 'retired_analyses': Value('string'), 'layer_a_report': Value('string'), 'three_tier_result': Value('string'), 'experiment_selection_result': Value('string'), 'reproducibility': Value('string'), 'data_provenance': Value('string'), 'archival_metadata': Value('string'), 'archival_release_notes': Value('string'), 'citation': Value('string'), 'license': Value('string'), 'one_page_arc': Value('string')}, 'data_policy': {'redistributes_third_party_raw_data': Value('bool'), 'small_result_artifacts_in_repo': Value('bool'), 'raw_data_regenerated_in_source_repository': Value('bool'), 'source_code_mirrored_here': Value('bool'), 'large_intermediate_predictions_released': Value('bool'), 'verify_or_trust_dataset': Value('string')}, 'claims': List({'id': Value('string'), '
...
ccuracy': Value('float64'), 'STATE_edge_AUROC': Value('float64'), 'STATE_cosine_PDS': Value('float64'), 'lambda_0_5': {'always_additive': Value('float64'), 'haiku_4_5': Value('float64'), 'sonnet_4_6': Value('float64'), 'opus_4_8': Value('float64')}, 'lambda_0_2': {'always_additive': Value('float64'), 'haiku_4_5': Value('float64'), 'sonnet_4_6': Value('float64'), 'opus_4_8': Value('float64')}, 'uniform_mapping_uncertainty_minus_random_AULC': Value('float64'), 'uniform_mapping_ci95': List(Value('float64')), 'hit_discovery_go_predicted_recall': Value('float64'), 'hit_discovery_random_recall': Value('float64'), 'hit_discovery_recall_ratio': Value('float64')}, 'evidence_files': List(Value('string')), 'code_files': List(Value('string')), 'linked_benchmark': Value('string'), 'linked_dataset': Value('string')}), 'retired_analyses': List({'id': Value('string'), 'status': Value('string'), 'name': Value('string'), 'reason': Value('string'), 'documentation': Value('string'), 'code_files': List(Value('string'))}), 'result_artifacts': List({'path': Value('string'), 'type': Value('string'), 'schema': Value('string'), 'strict_json': Value('bool'), 'description': Value('string')}), 'schema_files': List({'path': Value('string'), 'type': Value('string'), 'description': Value('string')}), 'validation': {'local_commands': List(Value('string')), 'source_validator': Value('string'), 'checks': List(Value('string'))}, 'pipeline_files': List(Value('null')), 'quality_upgrade_plan': List(Value('null'))}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 489, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2818, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2355, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2380, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 536, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 419, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 343, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2369, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2297, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              primary_delta: double
              B_boot: int64
              combo: struct<combo_in_train: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<ite (... 961 chars omitted)
                child 0, combo_in_train: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<item: double>, ADD_minus_G (... 93 chars omitted)
                    child 0, nP: int64
                    child 1, nE: int64
                    child 2, GEARS_acc: list<item: double>
                        child 0, item: double
                    child 3, ADD_acc: list<item: double>
                        child 0, item: double
                    child 4, ADD_minus_GEARS_acc: list<item: double>
                        child 0, item: double
                    child 5, GEARS_AUROC: list<item: double>
                        child 0, item: double
                    child 6, ADD_AUROC: list<item: double>
                        child 0, item: double
                child 1, combo_seen1: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<item: double>, ADD_minus_G (... 93 chars omitted)
                    child 0, nP: int64
                    child 1, nE: int64
                    child 2, GEARS_acc: list<item: double>
                        child 0, item: double
                    child 3, ADD_acc: list<item: double>
                        child 0, item: double
                    child 4, ADD_minus_GEARS_acc: list<item: double>
                        child 0, item: double
                    child 5, GEARS_AUROC: list<item: double>
                        child 0, item: double
                    child 6, ADD_AUROC: list<item: double>
                        child 0, item: double
                child 2, combo_seen0: struct<nP: int64, nE: int64, GEARS_acc: list<item: double>, ADD_acc: list<item: double>, ADD_minus_G (... 93 chars omit
              ...
              ouble
                        child 9, uniform_mapping_ci95: list<item: double>
                            child 0, item: double
                        child 10, hit_discovery_go_predicted_recall: double
                        child 11, hit_discovery_random_recall: double
                        child 12, hit_discovery_recall_ratio: double
                    child 6, evidence_files: list<item: string>
                        child 0, item: string
                    child 7, code_files: list<item: string>
                        child 0, item: string
                    child 8, linked_benchmark: string
                    child 9, linked_dataset: string
              quality_upgrade_plan: list<item: null>
                child 0, item: null
              schema_files: list<item: struct<path: string, type: string, description: string>>
                child 0, item: struct<path: string, type: string, description: string>
                    child 0, path: string
                    child 1, type: string
                    child 2, description: string
              project: struct<name: string, repository: string, default_branch: string, license: string, description: strin (... 2 chars omitted)
                child 0, name: string
                child 1, repository: string
                child 2, default_branch: string
                child 3, license: string
                child 4, description: string
              result_artifacts: list<item: struct<path: string, type: string, schema: string, strict_json: bool, description: string (... 2 chars omitted)
                child 0, item: struct<path: string, type: string, schema: string, strict_json: bool, description: string>
                    child 0, path: string
                    child 1, type: string
                    child 2, schema: string
                    child 3, strict_json: bool
                    child 4, description: string
              to
              {'schema_version': Value('string'), 'generated_date': Value('timestamp[s]'), 'project': {'name': Value('string'), 'repository': Value('string'), 'default_branch': Value('string'), 'license': Value('string'), 'description': Value('string')}, 'release': {'version': Value('string'), 'status': Value('string'), 'base_release': Value('string'), 'base_release_url': Value('string'), 'source_repository': Value('string')}, 'linked_public_artifacts': {'source_repository': Value('string'), 'verify_or_trust_repository': Value('string'), 'verify_or_trust_dataset': Value('string'), 'grounding_atlas_repository': Value('string')}, 'entrypoints': {'dataset_card': Value('string'), 'main_report': Value('string'), 'claim_map': Value('string'), 'retired_analyses': Value('string'), 'layer_a_report': Value('string'), 'three_tier_result': Value('string'), 'experiment_selection_result': Value('string'), 'reproducibility': Value('string'), 'data_provenance': Value('string'), 'archival_metadata': Value('string'), 'archival_release_notes': Value('string'), 'citation': Value('string'), 'license': Value('string'), 'one_page_arc': Value('string')}, 'data_policy': {'redistributes_third_party_raw_data': Value('bool'), 'small_result_artifacts_in_repo': Value('bool'), 'raw_data_regenerated_in_source_repository': Value('bool'), 'source_code_mirrored_here': Value('bool'), 'large_intermediate_predictions_released': Value('bool'), 'verify_or_trust_dataset': Value('string')}, 'claims': List({'id': Value('string'), '
              ...
              ccuracy': Value('float64'), 'STATE_edge_AUROC': Value('float64'), 'STATE_cosine_PDS': Value('float64'), 'lambda_0_5': {'always_additive': Value('float64'), 'haiku_4_5': Value('float64'), 'sonnet_4_6': Value('float64'), 'opus_4_8': Value('float64')}, 'lambda_0_2': {'always_additive': Value('float64'), 'haiku_4_5': Value('float64'), 'sonnet_4_6': Value('float64'), 'opus_4_8': Value('float64')}, 'uniform_mapping_uncertainty_minus_random_AULC': Value('float64'), 'uniform_mapping_ci95': List(Value('float64')), 'hit_discovery_go_predicted_recall': Value('float64'), 'hit_discovery_random_recall': Value('float64'), 'hit_discovery_recall_ratio': Value('float64')}, 'evidence_files': List(Value('string')), 'code_files': List(Value('string')), 'linked_benchmark': Value('string'), 'linked_dataset': Value('string')}), 'retired_analyses': List({'id': Value('string'), 'status': Value('string'), 'name': Value('string'), 'reason': Value('string'), 'documentation': Value('string'), 'code_files': List(Value('string'))}), 'result_artifacts': List({'path': Value('string'), 'type': Value('string'), 'schema': Value('string'), 'strict_json': Value('bool'), 'description': Value('string')}), 'schema_files': List({'path': Value('string'), 'type': Value('string'), 'description': Value('string')}), 'validation': {'local_commands': List(Value('string')), 'source_validator': Value('string'), 'checks': List(Value('string'))}, 'pipeline_files': List(Value('null')), 'quality_upgrade_plan': List(Value('null'))}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

CausalAtlas Move 1 Fact-Locked Artifacts

This dataset is the lightweight artifact mirror for CausalAtlas. It contains final public reports, small strict-JSON results, schemas, provenance, and citation metadata. It does not contain raw third-party single-cell matrices or large intermediate predictions.

Supported results

Question	Fact-locked result	Scope
Does GEARS beat the observed-additive baseline?	The additive baseline wins in aggregate on held-out Norman combinations; GEARS wins on the held-out non-additive subset.	Norman K562 CRISPRa, one GEARS configuration, edge AUROC.
Does Arc STATE beat the no-change baseline?	STATE call accuracy is 0.732 versus a 0.805 no-change baseline; cosine-PDS is 0.793.	Tahoe cancer-line substrate; metrics must be read together.
Does a free baseline change LLM orchestration?	At λ = 0.5, the always-additive policy achieves a net value of 0.802 and exceeds three tested Claude policies; at λ = 0.2, those model policies exceed the additive baseline.	52 panels, one neutral prompt family, three model versions.
Does informed experiment selection always help?	Uncertainty-based selection loses to random for uniform reconstruction, while GO-predicted effect-magnitude selection raises tested hit-discovery recall from 0.07 to 0.19.	Offline Replogle K562 replay; no LLM/RL selector claim.

The canonical account is results/move1/MOVE1_REPORT.md. Removed claims and their reasons are recorded in results/move1/RETIRED_ANALYSES.md.

Key numbers

Norman five-fold leave-combination-out evaluation:

subset	GEARS AUROC	additive AUROC	GEARS − additive (95% CI)
all held-out combinations	0.790	0.897	-0.107 [-0.132, -0.083]
additive-residual majority	0.755	0.945	-0.190 [-0.217, -0.162]
non-additive subset	0.749	0.606	+0.142 [+0.050, +0.233]

Three-tier cost sweep:

policy	net at λ = 0.2	net at λ = 0.5
always additive	0.802	0.802
Haiku 4.5	0.821	0.725
Sonnet 4.6	0.833	0.722
Opus 4.8	0.851	0.730

These results are conditioned on cost, prompt, model, and dataset. They do not define a general model capability scale.

artifact_manifest.json — HF-specific map of local artifacts, supported claims, source-code links, and retired analyses;
results/move1/MOVE1_REPORT.md — final fact-locked report;
results/move1/RESULT_layerA.md — Norman and Tahoe result details;
results/move1/MOVE1_3TIER_RESULT.md — scoped cost sweep;
results/move1/EXPERIMENT_SELECTION_RESULT.md — objective-dependent result;
results/move1/RETIRED_ANALYSES.md — retired interpretations and reasons;
results/move1/layerA_norman.json and layerA_tahoe.json — strict-JSON result artifacts;
schemas/*.json — JSON Schemas;
docs/*.md — claims, provenance, reproducibility, and archival status.

What is not included

raw Norman, Replogle, or Tahoe matrices;
GEARS checkpoints and large prediction tables;
LLM run logs;
source evaluation scripts, which remain in the GitHub repository;
superseded draft, synthesis, or experiment-design documents.

Load an artifact

import json
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="jang1563/causalatlas-move1",
    filename="results/move1/layerA_norman.json",
    repo_type="dataset",
)

with open(path) as handle:
    layer_a_norman = json.load(handle)

Validation

Validate the local JSON artifacts:

python3 -m json.tool artifact_manifest.json >/dev/null
python3 -m json.tool results/move1/layerA_norman.json >/dev/null
python3 -m json.tool results/move1/layerA_tahoe.json >/dev/null

Full source-release validation, including JSON Schema and link checks, is documented in docs/REPRODUCIBILITY.md.

Release status

This mirror corresponds to the stable v0.1.6 GitHub release.

Citation

Downloads last month: 68

Total file size:

93.9 kB