Datasets:

MatthewVaishnav
/

paired-acquisition-factorization-evidence

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
accelerator: string
captured_at_utc: timestamp[s]
environment_scope: string
git: string
packages: struct<numpy: string, pandas: string, scikit-learn: string, scipy: string>
  child 0, numpy: string
  child 1, pandas: string
  child 2, scikit-learn: string
  child 3, scipy: string
platform: string
python_implementation: string
python_version: string
schema_version: string
minimum_fit_samples_per_category_per_fold: int64
protocol: string
same_sample_neighbours_excluded: bool
uncertainty: string
fixed_categories: list<item: string>
  child 0, item: string
runtime_seconds: double
category_neighbour_pool: string
seeds: list<item: int64>
  child 0, item: int64
fit_only_probe_standardization: bool
minimum_test_samples_per_category_per_fold: int64
folds: list<item: int64>
  child 0, item: int64
historical_category_metrics_promoted: bool
excluded_categories: list<item: string>
  child 0, item: string
claim_boundary: string
oldstyle_centroid_qr_included: bool
same_region_neighbours_excluded: bool
status: string
to
{'category_neighbour_pool': Value('string'), 'claim_boundary': Value('string'), 'excluded_categories': List(Value('string')), 'fit_only_probe_standardization': Value('bool'), 'fixed_categories': List(Value('string')), 'folds': List(Value('int64')), 'historical_category_metrics_promoted': Value('bool'), 'minimum_fit_samples_per_category_per_fold': Value('int64'), 'minimum_test_samples_per_category_per_fold': Value('int64'), 'oldstyle_centroid_qr_included': Value('bool'), 'protocol': Value('string'), 'runtime_seconds': Value('float64'), 'same_region_neighbours_excluded': Value('bool'), 'same_sample_neighbours_excluded': Value('bool'), 'seeds': List(Value('int64')), 'status': Value('string'), 'uncertainty': Value('string')}
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
              accelerator: string
              captured_at_utc: timestamp[s]
              environment_scope: string
              git: string
              packages: struct<numpy: string, pandas: string, scikit-learn: string, scipy: string>
                child 0, numpy: string
                child 1, pandas: string
                child 2, scikit-learn: string
                child 3, scipy: string
              platform: string
              python_implementation: string
              python_version: string
              schema_version: string
              minimum_fit_samples_per_category_per_fold: int64
              protocol: string
              same_sample_neighbours_excluded: bool
              uncertainty: string
              fixed_categories: list<item: string>
                child 0, item: string
              runtime_seconds: double
              category_neighbour_pool: string
              seeds: list<item: int64>
                child 0, item: int64
              fit_only_probe_standardization: bool
              minimum_test_samples_per_category_per_fold: int64
              folds: list<item: int64>
                child 0, item: int64
              historical_category_metrics_promoted: bool
              excluded_categories: list<item: string>
                child 0, item: string
              claim_boundary: string
              oldstyle_centroid_qr_included: bool
              same_region_neighbours_excluded: bool
              status: string
              to
              {'category_neighbour_pool': Value('string'), 'claim_boundary': Value('string'), 'excluded_categories': List(Value('string')), 'fit_only_probe_standardization': Value('bool'), 'fixed_categories': List(Value('string')), 'folds': List(Value('int64')), 'historical_category_metrics_promoted': Value('bool'), 'minimum_fit_samples_per_category_per_fold': Value('int64'), 'minimum_test_samples_per_category_per_fold': Value('int64'), 'oldstyle_centroid_qr_included': Value('bool'), 'protocol': Value('string'), 'runtime_seconds': Value('float64'), 'same_region_neighbours_excluded': Value('bool'), 'same_sample_neighbours_excluded': Value('bool'), 'seeds': List(Value('int64')), 'status': Value('string'), 'uncertainty': Value('string')}
              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.

Paired-Acquisition Factorization Evidence

This dataset repository is a compact, machine-readable evidence release for two distinct Paired-Acquisition Neural Factorization (PA-NF) comparisons. It is an evidence dataset, not a trained model, frozen feature archive, or image dataset.

Data origin

Files are copied without reinterpretation from the authoritative program hub:

evidence/paired_acquisition/corrected-20260726;
evidence/paired_acquisition/scorpion-capacity-matched-20260726.

The source repository is matthewvaishnav/computational-pathology-research at immutable release-source commit edf8b2b96fbdb8b21fbecc03b8a19ac0351e1dce. The SCORPION campaign itself was executed at 0adea50f1ef22865969109f1834a3c175e3f8b43.

Unit of observation

Units differ by file and are explicitly named in the manifests: registered fit/cell, fold-level contrast, seed-averaged slide or biological-sample metric, and aggregate evidence record. Rows must not be treated as independent patients or clinical outcomes.

Schema

Path family	Format	Meaning
`corrected-20260726/`	JSON, CSV, Markdown	Corrected SCORPION and canine fixed-estimand evidence, environment, manifests, and bounded interpretation.
`scorpion-capacity-matched-20260726/campaign/`	JSON, JSONL, CSV	Registered campaign design, complete append-only run ledger, inventory, and run metrics.
`scorpion-capacity-matched-20260726/analysis/`	JSON, CSV	Fold-aware analysis specification, summaries, and contrasts.
`release-provenance.json`	JSON	HF preparation time, exact Git source, source paths, and release-spec hash.
`checksums.sha256`	SHA256 text	Complete integrity inventory for the HF release folder.

Each evidence family includes its own schema version and release manifest. Those machine-readable manifests are authoritative over this overview.

Sample counts

SCORPION capacity-matched campaign: 7 variants × 5 folds × 5 seeds = 175/175 valid registered fits, 36 aggregate contrast rows, and no failed, invalid, or mixed-configuration cells.
Corrected canine dimensionality × cross-covariance campaign: 450/450 registered cells in the promoted repository record.
SCORPION source design: 2,400 patches from 480 aligned regions on 48 original H&E slides scanned by five devices.

These counts describe the evidence designs; raw images and external feature arrays are not included.

Preprocessing

This release does not re-run, normalize, or transform scientific results. The pre-registered analysis outputs, environment records, command records, and hashes are copied from Git. Producer and analysis code paths are recorded in each release manifest. Reproduction must use those manifests and immutable commits.

Coordinates

No pixel or WSI coordinates are distributed. Same-region correspondence and slide/sample blocking are represented through the tracked study manifests and analysis design, not through image data in this HF repository.

Exclusions

The compact release intentionally excludes:

raw SCORPION or canine images;
frozen foundation-model feature archives;
projected feature arrays;
trained checkpoints;
durable terminal logs and local run directories;
superseded canine analyses;
historical slide-independent SCORPION inference;
unified cross-protocol leaderboard interpretations.

External artifacts referenced by hash in a release manifest are not silently reconstructed or represented as included files.

Checksums

checksums.sha256 is generated after the release folder is assembled. The publishing tool verifies every local file, uploads in one commit, downloads every released file at the returned immutable HF revision, recomputes SHA256, and only then records that revision in the GitHub release registry.

Licensing

The researcher-authored aggregate evidence package is distributed under the MIT license inherited from the source repository. Underlying images, external features, third-party model weights, and their licenses are not redistributed or relicensed by this release.

Intended use

reproduce the promoted fold-aware aggregate analyses;
audit registered campaign completeness and provenance;
inspect the distinct positive SCORPION and negative canine boundaries;
build evidence-aware comparisons without downloading raw pathology images.

Limitations

This release is not evidence of pure biological factors, complete scanner invariance, information-theoretic independence, universal harmonization superiority, diagnostic improvement, patient benefit, or clinical readiness. Probe and retrieval metrics are representation diagnostics. The canine public source is bounded by its documented coarse 4 µm/pixel release resolution.

Provenance

Program hub: https://github.com/matthewvaishnav/computational-pathology-research
Release-source commit: edf8b2b96fbdb8b21fbecc03b8a19ac0351e1dce
SCORPION execution commit: 0adea50f1ef22865969109f1834a3c175e3f8b43
Corrected evidence manifest: evidence/paired_acquisition/corrected-20260726/release_manifest.json
Capacity-matched evidence manifest: evidence/paired_acquisition/scorpion-capacity-matched-20260726/release_manifest.json
Current claim boundary: https://github.com/matthewvaishnav/computational-pathology-research/blob/main/CLAIM_BOUNDARY.md

Citation

Use the repository CITATION.cff and cite the primary foundations manuscript:

Matthew Vaishnav. Accountable Neural Aggregation in Computational Pathology: From Paired-Acquisition Representations to Whole-Slide and Institutional Learning (2026).

Also cite the original SCORPION and canine dataset publications when using the corresponding evidence family.

Claim boundary

Supported SCORPION statement:

On the registered SCORPION structured-separation objective, PA-NF has a controlled comparative advantage over the equal-capacity two-branch neural control: tissue-branch scanner balanced accuracy is reduced by 0.3108 with a fold-aware 95% interval of [-0.3346, -0.2858], registered same-region retrieval noninferiority is preserved, and acquisition-branch scanner information remains strong.

Separate supported canine statement:

Under the corrected five-category canine fixed-estimand comparison, no additional neural feature-space improvement over the strongest simple scanner-removal baselines was established.

The canine result does not negate the SCORPION controlled advantage, and the SCORPION result does not establish superiority to every simple scanner-removal method. Both remain endpoint-, comparator-, dataset-, and protocol-specific.

Downloads last month: 13

Total file size:

1.31 MB