sediment-export 0.1.0__py3-none-any.whl

sediment_export/__init__.py

@@ -0,0 +1,572 @@
+# SPDX-License-Identifier: AGPL-3.0-or-later
+"""Sediment export: thin, stateless projections over the canonical derived
+artifacts (ADR 0004) and their local destinations.
+
+Attributed completions (``attributed_completions.py``) are the first canonical
+artifact; the RLVR artifacts (``tasks.jsonl``, ``rollouts.jsonl``) project
+over rollouts, the second, with an optional taskset-level
+``environment.yaml`` sidecar (``environment_manifest.py``). DPO (``dpo.py``)
+and SFT (``sft.py``) project over attributed completions, sharing the baseline
+label-confidence policy (``label_confidence.py``). Diff-shaped SFT (``diff_sft.py``)
+reshapes the same attributed completions and policy into per-commit, per-file
+training rows; recovery (``recovery.py``) projects the one fact-derived
+training-row format (ADR 0004) into rows, stamping the eval split
+at projection time since ``RecoverySample`` carries none.
+
+The per-model outcome report (``outcome_report.py``) aggregates over attributed
+completions, and its stdlib-only two-proportion significance testing
+(``significance.py``) compares two such rows. Label-confidence inspection
+(``label_confidence_inspection.py``) stratify-samples attributed completions across the
+confidence ladder's real branches and surfaces ``resolve_confidence``'s
+factor breakdown per row, for ``sediment report label-confidence-inspection``.
+Inter-rater reliability (``interrater.py``) computes standalone Cohen's kappa
+over future multi-labeler review columns; label-confidence sensitivity
+(``label_confidence_sensitivity.py``) sweeps one ``LabelConfidencePolicy`` knob at a time over a
+fixed attributed-completion set; decision-latency diagnostics
+(``decision_latency.py``) bucket those same joined decisions by
+inference-call-to-decision latency.
+
+All export formats stay open-core (ADR 0006)."""
+
+from .calibration import (
+    BucketInversion,
+    CalibrationPair,
+    CalibrationRecord,
+    CalibrationMetrics,
+    CalibrationPolicyComparison,
+    CalibrationStratum,
+    ReliabilityBucket,
+    auroc,
+    brier_score,
+    bucket_inversions,
+    compare_calibration_policies,
+    expected_calibration_error,
+    reliability_diagram_data,
+    stratify_calibration,
+)
+from .accepted_work_lifecycle import (
+    AcceptedWorkLifecycleReport,
+    AcceptedWorkProgression,
+    EditRetentionPanel,
+    LifecycleCoverage,
+    LifecycleProvenance,
+    LifecycleReportPolicy,
+    LifecycleStage,
+    LifecycleStratum,
+    MergeDurabilityPanel,
+    ReworkComponent,
+    SessionAttritionPanel,
+    SessionExamples,
+    generate_accepted_work_lifecycle_report,
+)
+from .config import LabelConfidenceSettings
+from .dataset_diagnostics import (
+    DPOBucketSparsityReport,
+    ConfidenceDistributionRow,
+    ConfidenceFloorExclusionRow,
+    CrossSplitDuplicateExample,
+    CrossSplitDuplicateReport,
+    DEFAULT_DPO_NEAR_DUPLICATE_THRESHOLD,
+    DPONearDuplicateBucketExample,
+    DPONearDuplicateBucketReport,
+    DatasetDiagnostics,
+    DistributionStats,
+    FateDiagnostic,
+    ModelBalanceRow,
+    build_dataset_diagnostics,
+    build_dpo_bucket_sparsity,
+    generate_dataset_diagnostics,
+)
+from .decision_latency import (
+    DecisionLatencyBucket,
+    DecisionLatencyReport,
+    build_decision_latency_report,
+    generate_decision_latency_report,
+)
+from .derived_bundle import (
+    BundleFileMetadata,
+    BundleManifest,
+    BundleRecord,
+    BundleValidationError,
+    DerivedBundle,
+    DerivationPolicy,
+    DerivationScope,
+    build_derived_bundle,
+    build_derived_bundle_context,
+    load_derivation_policy,
+    read_derived_bundle,
+    open_derived_bundle,
+    BundleLimits,
+    BundleCapacityError,
+    BundleRecordStore,
+    validate_derived_bundle,
+    write_derived_bundle,
+)
+from .diff_sft import DiffSFTMetadata, DiffSFTSample, SourceIds, project_diff_sft
+from .diff_sft import DIFF_SFT_SKIP_REASONS, DiffSFTSkipReason
+from .diff_sft import Projection as DiffSFTProjection
+from .diff_sft import to_export_rows as diff_sft_to_export_rows
+from .dpo import (
+    DPO_RECIPE_VERSION,
+    DPO_SKIP_REASONS,
+    DPOChosenLabelSource as DPOChosenLabelSource,
+    DPOLabelSource,
+    DPOMetadata,
+    DPOPair,
+    DPOPolicy,
+    DPOProvenance,
+    DPORecipeId,
+    DPORecipeVersion,
+    DPORejectedLabelSource as DPORejectedLabelSource,
+    DPOSkipReason,
+    project_dpo,
+)
+from .dpo import Projection as DPOProjection
+from .dpo import to_export_rows as dpo_to_export_rows
+from .environment_manifest import (
+    NemoGymRuntimeSettings,
+    OpenEnvRuntimeSettings,
+    build_manifest,
+    write_manifest,
+)
+from .interrater import (
+    InterRaterAgreement,
+    Label,
+    cohens_kappa,
+    percent_agreement,
+)
+from .jsonl import ExportRow, WriteResult, write_jsonl
+from .outcome_report import (
+    CIGrain,
+    MantelHaenszelRiskDifference,
+    DiffSizeStats,
+    MannKendallTrendTest,
+    ModelOutcomeReport,
+    ModelOutcomeReportResult,
+    RepositoryShrunkRates,
+    ModelTemporalTrend,
+    OutcomeReportPolicy,
+    RecoveryDiffSizeDistribution,
+    RecoveryYieldReport,
+    SignalFunnelReport,
+    StratificationCheck,
+    StratificationComparison,
+    StratificationMetric,
+    StratificationReversal,
+    StratificationStatus,
+    StratifiedRate,
+    TemporalTrendWindow,
+    TrendMetric,
+    TrendStatus,
+    ShrunkStratifiedRate,
+    build_model_report,
+    build_model_report_result,
+    build_recovery_yield_report,
+    build_signal_funnel_report,
+    build_stratification_checks,
+    build_temporal_trend_report,
+    derive_model_report_attribution_share,
+    generate_model_report,
+    generate_model_report_result,
+    generate_recovery_yield_report,
+    mann_kendall_trend_test,
+    shrink_stratified_rates,
+    scope_model_report_evidence,
+    wilson_score_interval,
+)
+from .operational_scope import OperationalReportScope
+from .price_manifest import ModelPrice as ModelPrice
+from .price_manifest import PriceManifest as PriceManifest
+from .recovery import (
+    RECOVERY_PROJECTION_SKIP_REASONS as RECOVERY_PROJECTION_SKIP_REASONS,
+    RECOVERY_RECIPE_ID,
+    RECOVERY_RECIPE_VERSION,
+    RecoveryProjectionSkipReason as RecoveryProjectionSkipReason,
+    RecoveryRow,
+    RecoveryRecipeId,
+    RecoveryRecipeVersion,
+    project_recovery,
+)
+from .recovery import Projection as RecoveryProjection
+from .recovery import to_export_rows as recovery_to_export_rows
+from .label_confidence import (
+    ConfidenceBreakdown,
+    DecisionBranch,
+    LabelConfidencePolicy,
+    ci_failed,
+    ci_passed,
+    decision_branch,
+    resolve_confidence,
+    resolve_confidence_breakdown,
+    resolve_ci_resolution,
+    resolve_ci_resolution_result,
+    resolve_policy_v3_confidence,
+)
+from .verifier_commands import VerifierCommands, VerifierCommandsSettings
+from .label_confidence_inspection import (
+    CIOutcomeSummary,
+    DecisionSummary,
+    InspectionRow,
+    build_label_confidence_inspection,
+    generate_label_confidence_inspection,
+    stratified_sample,
+)
+from .label_confidence_sensitivity import (
+    DEFAULT_SWEEP_GRIDS,
+    LABEL_CONFIDENCE_POLICY_KNOBS,
+    ConfidenceStats,
+    LabelConfidencePolicyKnob,
+    LabelConfidenceSensitivityRow,
+    build_label_confidence_sensitivity,
+    generate_label_confidence_sensitivity,
+)
+from .rlvr import (
+    NemoGymMetadata,
+    NemoGymResponse,
+    NemoGymResponsesCreateParams,
+    NemoGymRolloutRow,
+    Projection,
+    RLVRDecisionRow,
+    RLVRInferenceMessageRow,
+    RLVREcipeId,
+    RLVRRecipeVersion,
+    RLVRRewardSource,
+    RLVRTarget,
+    RLVRTurnRow,
+    SWEBenchMetadata,
+    SWEBenchTaskRow,
+    SedimentRolloutRow,
+    SedimentTaskRow,
+    Verification,
+    export_rlvr,
+    export_rlvr_from_rollouts,
+    export_rlvr_from_bundle,
+    project_nemo_gym_rollouts,
+    project_sediment_rollouts,
+    project_sediment_tasks,
+    project_swe_bench_tasks,
+)
+from .sft import Projection as SFTProjection
+from .sft import (
+    SFT_CURATED_V1_STRONG_RETENTION_THRESHOLD,
+    SFT_RECIPE_VERSION,
+    SFT_SKIP_REASONS,
+    SFTEligibilitySource,
+    SFTMetadata,
+    SFTPolicy,
+    SFTRecipeId,
+    SFTRecipeVersion,
+    SFTSkipReason,
+    SFTSample,
+    project_sft,
+)
+from .sft import to_export_rows as sft_to_export_rows
+from .significance import (
+    DEFAULT_BOOTSTRAP_ITERATIONS,
+    DEFAULT_BOOTSTRAP_SEED,
+    BayesianProportionComparison,
+    BootstrapIntervalDiagnostic,
+    EffectDecayCheckpoint,
+    EffectDecayReport,
+    ModelComparison,
+    MultiModelComparison,
+    NonInferiorityResult,
+    PowerSimulationResult,
+    ProportionComparison,
+    RegretAlternative,
+    RegretReport,
+    SequentialBoundary,
+    beta_binomial_probability_of_improvement,
+    bootstrap_two_proportion_interval_diagnostic,
+    cohens_h,
+    compare_all_models,
+    compare_models,
+    effect_decay_check,
+    expected_regret,
+    minimum_detectable_effect,
+    non_inferiority_test,
+    obrien_fleming_alpha_spent,
+    obrien_fleming_boundary,
+    required_days_for_mde,
+    required_n_per_arm_for_mde,
+    simulate_two_proportion_power,
+    two_proportion_z_test,
+)
+from .attributed_completions import (
+    AbandonmentSummary,
+    AttributedCompletion,
+    AttributedCompletionAssemblyResult,
+    AttributedCompletionPolicy,
+    AttributedCompletionResult,
+    assemble_attributed_completion_result,
+    assemble_attributed_completions,
+    assemble_attributed_completions_result,
+    build_abandonment_summary,
+)
+from .merge_retention_report import (
+    RETENTION_THRESHOLDS,
+    MergeRetentionReport,
+    MergeRetentionReportResult,
+    RetentionDistribution,
+    RetentionScoreSummary,
+    RetentionThreshold,
+    build_merge_retention_report,
+    generate_merge_retention_report,
+    generate_merge_retention_report_result,
+    merge_retention_to_export_rows,
+)
+
+__all__ = [
+    "AcceptedWorkLifecycleReport",
+    "AcceptedWorkProgression",
+    "EditRetentionPanel",
+    "LifecycleCoverage",
+    "LifecycleProvenance",
+    "LifecycleReportPolicy",
+    "LifecycleStage",
+    "LifecycleStratum",
+    "MergeDurabilityPanel",
+    "ReworkComponent",
+    "SessionAttritionPanel",
+    "SessionExamples",
+    "generate_accepted_work_lifecycle_report",
+    "AbandonmentSummary",
+    "CIGrain",
+    "CIOutcomeSummary",
+    "ConfidenceBreakdown",
+    "ConfidenceStats",
+    "DPOPair",
+    "DPOMetadata",
+    "DPOPolicy",
+    "DPOProvenance",
+    "DPOProjection",
+    "DPORecipeId",
+    "DPORecipeVersion",
+    "DPOLabelSource",
+    "DPOSkipReason",
+    "DPO_RECIPE_VERSION",
+    "DPO_SKIP_REASONS",
+    "DEFAULT_DPO_NEAR_DUPLICATE_THRESHOLD",
+    "DPONearDuplicateBucketExample",
+    "DPONearDuplicateBucketReport",
+    "DPOBucketSparsityReport",
+    "ConfidenceDistributionRow",
+    "ConfidenceFloorExclusionRow",
+    "CrossSplitDuplicateExample",
+    "CrossSplitDuplicateReport",
+    "DatasetDiagnostics",
+    "DEFAULT_SWEEP_GRIDS",
+    "DecisionBranch",
+    "DecisionLatencyBucket",
+    "DecisionLatencyReport",
+    "DerivationPolicy",
+    "DerivationScope",
+    "DerivedBundle",
+    "DecisionSummary",
+    "DiffSFTProjection",
+    "DiffSFTSample",
+    "DiffSizeStats",
+    "DistributionStats",
+    "FateDiagnostic",
+    "derive_model_report_attribution_share",
+    "ExportRow",
+    "EffectDecayCheckpoint",
+    "EffectDecayReport",
+    "DiffSFTMetadata",
+    "DiffSFTSkipReason",
+    "DIFF_SFT_SKIP_REASONS",
+    "BucketInversion",
+    "BundleValidationError",
+    "BundleFileMetadata",
+    "BundleManifest",
+    "BundleRecord",
+    "BootstrapIntervalDiagnostic",
+    "CalibrationPair",
+    "CalibrationRecord",
+    "CalibrationMetrics",
+    "CalibrationPolicyComparison",
+    "CalibrationStratum",
+    "InspectionRow",
+    "MantelHaenszelRiskDifference",
+    "InterRaterAgreement",
+    "Label",
+    "MannKendallTrendTest",
+    "ModelComparison",
+    "MultiModelComparison",
+    "ModelOutcomeReport",
+    "ModelOutcomeReportResult",
+    "RepositoryShrunkRates",
+    "ModelTemporalTrend",
+    "ModelBalanceRow",
+    "NonInferiorityResult",
+    "NemoGymRuntimeSettings",
+    "OpenEnvRuntimeSettings",
+    "OutcomeReportPolicy",
+    "OperationalReportScope",
+    "PowerSimulationResult",
+    "Projection",
+    "ProportionComparison",
+    "RegretAlternative",
+    "RegretReport",
+    "VerifierCommands",
+    "VerifierCommandsSettings",
+    "LabelConfidencePolicy",
+    "LabelConfidencePolicyKnob",
+    "LabelConfidenceSettings",
+    "LabelConfidenceSensitivityRow",
+    "RecoveryYieldReport",
+    "RecoveryDiffSizeDistribution",
+    "RecoveryProjection",
+    "RecoveryRow",
+    "RecoveryRecipeId",
+    "RecoveryRecipeVersion",
+    "RECOVERY_RECIPE_ID",
+    "RECOVERY_RECIPE_VERSION",
+    "RLVREcipeId",
+    "RLVRRecipeVersion",
+    "RLVRRewardSource",
+    "RLVRTarget",
+    "RLVRDecisionRow",
+    "RLVRInferenceMessageRow",
+    "RLVRTurnRow",
+    "ReliabilityBucket",
+    "SequentialBoundary",
+    "SFTPolicy",
+    "SFTRecipeId",
+    "SFTRecipeVersion",
+    "SFTEligibilitySource",
+    "SFTSkipReason",
+    "SFT_RECIPE_VERSION",
+    "SFT_SKIP_REASONS",
+    "SFT_CURATED_V1_STRONG_RETENTION_THRESHOLD",
+    "SFTMetadata",
+    "SFTProjection",
+    "SFTSample",
+    "SWEBenchMetadata",
+    "SWEBenchTaskRow",
+    "SedimentRolloutRow",
+    "SedimentTaskRow",
+    "SignalFunnelReport",
+    "SourceIds",
+    "StratificationCheck",
+    "StratificationComparison",
+    "StratificationMetric",
+    "StratificationReversal",
+    "StratificationStatus",
+    "StratifiedRate",
+    "TemporalTrendWindow",
+    "TrendMetric",
+    "TrendStatus",
+    "ShrunkStratifiedRate",
+    "AttributedCompletion",
+    "AttributedCompletionAssemblyResult",
+    "AttributedCompletionPolicy",
+    "AttributedCompletionResult",
+    "MergeRetentionReport",
+    "MergeRetentionReportResult",
+    "RetentionDistribution",
+    "RetentionScoreSummary",
+    "RetentionThreshold",
+    "RETENTION_THRESHOLDS",
+    "WriteResult",
+    "assemble_attributed_completions",
+    "assemble_attributed_completion_result",
+    "assemble_attributed_completions_result",
+    "BayesianProportionComparison",
+    "beta_binomial_probability_of_improvement",
+    "auroc",
+    "build_dataset_diagnostics",
+    "build_abandonment_summary",
+    "build_derived_bundle",
+    "build_derived_bundle_context",
+    "build_dpo_bucket_sparsity",
+    "build_decision_latency_report",
+    "build_manifest",
+    "build_merge_retention_report",
+    "brier_score",
+    "bucket_inversions",
+    "build_model_report",
+    "build_model_report_result",
+    "build_recovery_yield_report",
+    "build_label_confidence_inspection",
+    "build_signal_funnel_report",
+    "build_label_confidence_sensitivity",
+    "build_stratification_checks",
+    "bootstrap_two_proportion_interval_diagnostic",
+    "build_temporal_trend_report",
+    "ci_failed",
+    "ci_passed",
+    "cohens_h",
+    "cohens_kappa",
+    "compare_all_models",
+    "compare_calibration_policies",
+    "compare_models",
+    "decision_branch",
+    "DEFAULT_BOOTSTRAP_ITERATIONS",
+    "DEFAULT_BOOTSTRAP_SEED",
+    "diff_sft_to_export_rows",
+    "dpo_to_export_rows",
+    "effect_decay_check",
+    "export_rlvr",
+    "export_rlvr_from_rollouts",
+    "export_rlvr_from_bundle",
+    "expected_regret",
+    "expected_calibration_error",
+    "generate_decision_latency_report",
+    "generate_model_report",
+    "generate_merge_retention_report",
+    "generate_dataset_diagnostics",
+    "generate_model_report_result",
+    "generate_recovery_yield_report",
+    "generate_label_confidence_inspection",
+    "mann_kendall_trend_test",
+    "load_derivation_policy",
+    "read_derived_bundle",
+    "open_derived_bundle",
+    "BundleLimits",
+    "BundleCapacityError",
+    "BundleRecordStore",
+    "validate_derived_bundle",
+    "generate_label_confidence_sensitivity",
+    "minimum_detectable_effect",
+    "percent_agreement",
+    "non_inferiority_test",
+    "obrien_fleming_alpha_spent",
+    "obrien_fleming_boundary",
+    "project_diff_sft",
+    "project_dpo",
+    "project_nemo_gym_rollouts",
+    "project_recovery",
+    "project_sediment_rollouts",
+    "project_sediment_tasks",
+    "project_sft",
+    "project_swe_bench_tasks",
+    "LABEL_CONFIDENCE_POLICY_KNOBS",
+    "recovery_to_export_rows",
+    "reliability_diagram_data",
+    "required_days_for_mde",
+    "required_n_per_arm_for_mde",
+    "resolve_confidence",
+    "resolve_confidence_breakdown",
+    "resolve_ci_resolution",
+    "resolve_ci_resolution_result",
+    "resolve_policy_v3_confidence",
+    "sft_to_export_rows",
+    "simulate_two_proportion_power",
+    "shrink_stratified_rates",
+    "scope_model_report_evidence",
+    "stratified_sample",
+    "stratify_calibration",
+    "two_proportion_z_test",
+    "write_jsonl",
+    "generate_merge_retention_report_result",
+    "merge_retention_to_export_rows",
+    "write_derived_bundle",
+    "write_manifest",
+    "wilson_score_interval",
+    "NemoGymMetadata",
+    "NemoGymResponse",
+    "NemoGymResponsesCreateParams",
+    "NemoGymRolloutRow",
+    "Verification",
+]