1 datasets.bundles.corpus

datasets.bundles.corpus

Generation-bound composition of rollout stores and their VIN dependencies.

Corpus membership is independent of dependency availability. Only aliases of one physical target are equivalent; matching manifests in independent stores require an explicit override because manifests do not seal tensor contents. The established one-root bundle API and its V1 identities remain unchanged.

1.1 Classes

Name Description
DatasetCorpusEvidence Frozen evidence retaining unresolved members and attributable findings.
DatasetCorpusGeneration V2 bounded identity covering every selected entry and binding override.
DatasetCorpusSelection Explicit rollout membership, root candidates and exact path overrides.

1.1.1 DatasetCorpusEvidence

datasets.bundles.corpus.DatasetCorpusEvidence(
    generation,
    roots,
    rollouts,
    aggregate,
    findings,
    verdict,
)

Frozen evidence retaining unresolved members and attributable findings.

1.1.1.1 Methods

Name Description
to_jsonable Export the same population, bindings and findings shown by clients.
1.1.1.1.1 to_jsonable
datasets.bundles.corpus.DatasetCorpusEvidence.to_jsonable()

Export the same population, bindings and findings shown by clients.

1.1.2 DatasetCorpusGeneration

datasets.bundles.corpus.DatasetCorpusGeneration(
    identity_version,
    selection,
    roots,
    rollouts,
    generation_digest,
)

V2 bounded identity covering every selected entry and binding override.

1.1.2.1 Methods

Name Description
to_jsonable Export the exact replacement-sensitive generation preimage.
1.1.2.1.1 to_jsonable
datasets.bundles.corpus.DatasetCorpusGeneration.to_jsonable()

Export the exact replacement-sensitive generation preimage.

1.1.3 DatasetCorpusSelection

datasets.bundles.corpus.DatasetCorpusSelection(
    root_stores,
    rollout_stores,
    root_overrides=(),
)

Explicit rollout membership, root candidates and exact path overrides.

Selected entry paths retain aliases until generation capture. Override pairs are (manifest_hash, selected_root_path) and cannot introduce a root outside the candidate set.

1.2 Functions

Name Description
assert_dataset_corpus_generation_current Fail if any root, shard, alias or override changed since acquisition.
capture_dataset_corpus_generation Fingerprint bounded authoritative metadata without opening arrays.
inspect_dataset_corpus Resolve each member’s exact root, keeping missing/ambiguous members.
referenced_root_overrides Scope retained bindings to exact source manifests of selected rollouts.

1.2.1 assert_dataset_corpus_generation_current

datasets.bundles.corpus.assert_dataset_corpus_generation_current(generation)

Fail if any root, shard, alias or override changed since acquisition.

1.2.2 capture_dataset_corpus_generation

datasets.bundles.corpus.capture_dataset_corpus_generation(selection)

Fingerprint bounded authoritative metadata without opening arrays.

1.2.3 inspect_dataset_corpus

datasets.bundles.corpus.inspect_dataset_corpus(
    selection,
    *,
    generation=None,
    validate_rollouts=False,
)

Resolve each member’s exact root, keeping missing/ambiguous members.

This explicit inspection reads manifests, source indices and splits. Full rollout payload validation is optional and never follows from discovery.

1.2.4 referenced_root_overrides

datasets.bundles.corpus.referenced_root_overrides(
    rollout_stores,
    root_overrides,
)

Scope retained bindings to exact source manifests of selected rollouts.

Unrelated retained bindings are inactive. Required bindings still pass the selection’s configured-candidate validation, so stale required paths block acquisition rather than selecting another root implicitly.

Back to top