1 oracle.pipelines.online_qh

oracle.pipelines.online_qh

Pipeline-local Q_H inference adapters and immutable round receipts.

1.1 Classes

Name Description
OnlineQhPopulationManifestRef Content-bound reference to the immutable collection population.
OnlineQhRoundCounts Explicit non-derived counters for one collection round.
OnlineQhRoundRequest Frozen behavior and query policy for one synchronous collection round.
OnlineQhRoundResult Validated immutable shard and provenance receipt for one round.

1.1.1 OnlineQhPopulationManifestRef

oracle.pipelines.online_qh.OnlineQhPopulationManifestRef(
    manifest_path,
    schema_version,
    manifest_sha256,
)

Content-bound reference to the immutable collection population.

1.1.1.1 Attributes

Name Description
manifest_path Path to the exact JSON population manifest.
schema_version Closed schema expected inside the manifest payload.
manifest_sha256 SHA-256 of the exact manifest bytes.

1.1.1.2 Methods

Name Description
read_verified Read the manifest after verifying its exact bytes and declared schema.
1.1.1.2.1 read_verified
oracle.pipelines.online_qh.OnlineQhPopulationManifestRef.read_verified()

Read the manifest after verifying its exact bytes and declared schema.

1.1.2 OnlineQhRoundCounts

oracle.pipelines.online_qh.OnlineQhRoundCounts(
    proposed,
    valid,
    queried,
    labeled,
    selected,
    persisted,
    rejected,
)

Explicit non-derived counters for one collection round.

proposed through labeled count candidate rows. selected and persisted count decision transitions, so no cross-unit inequality is inferred between labeled and selected. rejected counts proposal tables that yield no transition because support is empty, labels are unavailable, or the episode terminates before selection.

1.1.2.1 Attributes

Name Description
proposed All attempted candidate rows before hard validity.
valid Hard-valid candidate rows presented to a selection policy.
queried Candidate rows submitted to the hard oracle.
labeled Candidate rows with admitted finite oracle labels.
selected Decision transitions with one action selected from labeled support.
persisted Transitions committed to the immutable shard.
rejected Proposal tables rejected or terminated before an action is selected.

1.1.3 OnlineQhRoundRequest

oracle.pipelines.online_qh.OnlineQhRoundRequest(
    behavior_bundle,
    training_population_manifest,
    acquisition_round,
    proposal_policy_manifest_sha256,
    selection_policy,
    oracle_query_budget,
    decision_cohort,
    query_mode='dense_valid',
)

Frozen behavior and query policy for one synchronous collection round.

1.1.3.1 Attributes

Name Description
behavior_bundle Verified immutable model used for every decision in the round.
training_population_manifest Content-bound train-only scene/acquisition population manifest.
acquisition_round Non-negative sequential round identifier.
proposal_policy_manifest_sha256 Candidate proposal-policy manifest identity, distinct from selection.
selection_policy Existing replay selection policy applied to Q_H scores.
oracle_query_budget Positive hard-oracle candidate-row budget.
decision_cohort Explicit role of every decision collected in this round.
query_mode Closed MVP collection mode admitted to fitted-Q learning.

1.1.3.2 Methods

Name Description
verify_training_population Return the population only after exact content and schema verification.
1.1.3.2.1 verify_training_population
oracle.pipelines.online_qh.OnlineQhRoundRequest.verify_training_population()

Return the population only after exact content and schema verification.

1.1.4 OnlineQhRoundResult

oracle.pipelines.online_qh.OnlineQhRoundResult(
    shard_dir,
    shard_manifest_sha256,
    behavior_bundle_manifest_sha256,
    proposal_policy_manifest_sha256,
    oracle_query_policy_id,
    selected_action_policy_sha256,
    round_receipt_path,
    round_receipt_sha256,
    counts,
    training_population_manifest_schema_version,
    training_population_manifest_sha256,
    training_population_manifest_path,
    decision_cohort,
)

Validated immutable shard and provenance receipt for one round.

1.1.4.1 Attributes

Name Description
shard_dir New immutable dense-valid rollout shard.
shard_manifest_sha256 Validated shard-manifest content identity.
behavior_bundle_manifest_sha256 Frozen behavior bundle used throughout collection.
proposal_policy_manifest_sha256 Candidate-proposal policy identity, separate from selection.
oracle_query_policy_id Closed hard-oracle query policy admitted to the MVP objective.
selected_action_policy_sha256 Existing replay selection-policy identity.
round_receipt_path Canonical collection receipt path.
round_receipt_sha256 Content hash of round_receipt_path.
counts Explicit collection counters; readers must not derive one from another.
training_population_manifest_schema_version Schema of the exact population manifest used by this round.
training_population_manifest_sha256 Content digest of the exact population manifest used by this round.
training_population_manifest_path Recorded path of the exact content-verified population manifest.
decision_cohort Explicit evaluation role of every decision in the promoted shard.
Back to top