SDK guide / 09

The high-level classes and functions, organized by the work they enable.

This is a navigation reference, not a generated signature dump. Start with a workflow, then open the maintained source for constructor fields, validation rules, and serialized schemas.

Import model

The root package is intentionally small.

The root exports the client and run context. Scientific plans, reports, enums, and evaluation functions live in the module that owns their evidence contract.

imports.pyPython

from iso_obs import ReliabilityClient, RunContext

# Analysis functions live with the evidence contract they evaluate.
from iso_obs.divergence import compare_traces
from iso_obs.execution import evaluate_simulation_campaign
from iso_obs.dataset_ingestion import ingest_dataset_source
from iso_obs.assurance import evaluate_release_assurance

READING THE API

A Plan declares the question and design. An Observation or Result carries evidence. A Report preserves the conclusion, uncertainty, scope, limitations, and provenance. The module function performs the assessment.

Capture

Runs, resources, environments, and trace extraction.

Relevant SDK classes and functions
Class or function	Import from	Use it to
`ReliabilityClient`	`iso_obs`	Top-level authenticated client for Studio resources.
`RunContext`	`iso_obs`	Run lifecycle, ordered event buffering, failure handling, and artifacts.
`ProjectsResource`	`iso_obs.client`	Create and list projects.
`SystemsResource`	`iso_obs.client`	Register system versions and artifact provenance.
`RunsResource`	`iso_obs.client`	Create, stream, complete, fail, and retrieve runs.
`ReliabilityEnvironment`	`iso_obs.environments`	Abstract interface for seeded reset, stepping, and environment identity.
`metric`	`iso_obs.metrics`	Declare a project-specific metric for suite tooling.
`EventSignalSpec`	`iso_obs.trace`	Map normalized events to numeric analysis signals.
`extract_trace_steps`	`iso_obs.trace`	Build analysis-ready steps and explicit extraction diagnostics.

Evaluate

Comparison, replication, multiplicity, and perturbations.

Relevant SDK classes and functions
Class or function	Import from	Use it to
`PairingAssessment`	`iso_obs.divergence`	Assess exact replay, seed matching, configuration matching, or unpaired evidence.
`SignalSpec`	`iso_obs.divergence`	Declare path, category, tolerances, and persistence.
`DivergenceReport`	`iso_obs.divergence`	Report the first difference and first sustained meaningful divergence.
`compare_traces`	`iso_obs.divergence`	Compare aligned baseline and candidate traces.
`PairedObservation`	`iso_obs.replication`	Represent one matched scalar baseline/candidate outcome.
`PairedAnalysisConfig`	`iso_obs.replication`	Configure deterministic bootstrap and randomization analysis.
`PairedEffectReport`	`iso_obs.replication`	Carry effect summaries, confidence interval, p-value, and limitations.
`analyze_paired_effect`	`iso_obs.replication`	Estimate a prespecified paired effect across independent pairs.
`MultiplicityPlan`	`iso_obs.multiplicity`	Declare analysis mode, error criterion, method, and dataset separation.
`MultiplicityReport`	`iso_obs.multiplicity`	Carry raw and adjusted results with bounded evidence labels.
`adjust_hypothesis_family`	`iso_obs.multiplicity`	Apply Holm, Bonferroni, BH, or BY adjustment.
`CausalPerturbationPlan`	`iso_obs.causal_perturbations`	Declare controlled contrasts, estimand, margins, and evidence requirements.
`CausalPerturbationReport`	`iso_obs.causal_perturbations`	Report contrast effects and the worst-member campaign disposition.
`assess_causal_perturbations`	`iso_obs.causal_perturbations`	Evaluate paired perturbation-response evidence.

Discover

Failure artifacts, phenotypes, boundaries, surfaces, and counterfactuals.

Relevant SDK classes and functions
Class or function	Import from	Use it to
`SimulationEvidenceManifest`	`iso_obs.evidence`	Keep simulator verification, calibration, validation, anchors, and omissions separate.
`ReproductionManifest`	`iso_obs.evidence`	Bind scenario, seed, environment, system, parameters, and artifacts for reproduction.
`FailureEvidenceBundle`	`iso_obs.evidence`	Package a failure as a deterministic, content-addressed artifact.
`FailurePhenotypePlan`	`iso_obs.failure_phenotypes`	Declare taxonomy, features, splits, and conformal calibration.
`FailurePhenotypeReport`	`iso_obs.failure_phenotypes`	Return assigned, ambiguous, or novel-candidate prediction sets.
`fit_failure_phenotype_model`	`iso_obs.failure_phenotypes`	Fit robust class prototypes using fit-only evidence.
`classify_failure_phenotype`	`iso_obs.failure_phenotypes`	Classify one query using held-out class-conditional calibration.
`FailureBoundaryPlan`	`iso_obs.failure_boundaries`	Declare dimensions, anchors, cells, thresholds, and design requirements.
`FailureBoundaryMap`	`iso_obs.failure_boundaries`	Carry reliable, unreliable, and unresolved cells with intervals.
`fit_failure_boundary_model`	`iso_obs.failure_boundaries`	Fit a bounded failure model from anchor observations.
`map_failure_boundary`	`iso_obs.failure_boundaries`	Certify cells only where model and evidence allow.
`FailureSurfacePlan`	`iso_obs.failure_surface`	Declare disaggregated strata and acceptable failure-rate limits.
`FailureSurfaceReport`	`iso_obs.failure_surface`	Carry per-stratum intervals and a noncompensatory surface conclusion.
`estimate_failure_surface`	`iso_obs.failure_surface`	Estimate stratified failure rates after design completeness.
`ExperimentSelectionPlan`	`iso_obs.experiment_selection`	Declare exploratory regions, costs, severity, and budget.
`ExperimentSelectionReport`	`iso_obs.experiment_selection`	Carry the next-region recommendation and acquisition decomposition.
`recommend_next_experiment`	`iso_obs.experiment_selection`	Recommend the next region with a complete acquisition decomposition.
`CounterfactualDesign`	`iso_obs.counterfactual`	Bind replay controls, intervention, estimand, and paired trials.
`CounterfactualEvidenceReport`	`iso_obs.counterfactual`	Carry bounded in-simulation intervention evidence and gating checks.
`MinimalCounterexampleReport`	`iso_obs.counterfactual`	Report local one-minimality and every untested single removal.
`analyze_counterfactual_design`	`iso_obs.counterfactual`	Assess a replicated intervention association within simulation scope.
`assess_minimal_counterexample`	`iso_obs.counterfactual`	Assess local one-minimality over executed ablations.

Execute

Regression packs, simulator adapters, work plans, and campaign results.

Relevant SDK classes and functions
Class or function	Import from	Use it to
`RegressionCase`	`iso_obs.regression`	Represent one canonical, neighborhood, boundary, or control case.
`SeedPanel`	`iso_obs.regression`	Declare seeds and their exact randomization strategy.
`GateCriterion`	`iso_obs.regression`	Declare exact or confidence-bound gate requirements.
`RegressionPack`	`iso_obs.regression`	Bind cases, seeds, gates, source failure, and simulator evidence.
`RegressionGateReport`	`iso_obs.regression`	Carry per-criterion and overall regression evaluation.
`evaluate_regression_pack`	`iso_obs.regression`	Evaluate a complete case-by-seed observation matrix.
`SimulationAdapterManifest`	`iso_obs.simulation`	Declare backend identity, capabilities, parameters, artifacts, and limitations.
`CompatibilityReport`	`iso_obs.simulation`	Explain whether an adapter can execute the requested pack and intent.
`SimulationExecutionPlan`	`iso_obs.simulation`	Carry identity-bound work items and execution batches.
`compile_simulation_plan`	`iso_obs.simulation`	Preflight and expand a pack into deterministic work.
`SimulationRunResult`	`iso_obs.execution`	Return status, trace, diagnostics, and artifacts for one work item.
`SimulationCampaignReport`	`iso_obs.execution`	Carry completeness, issues, evaluations, and any eligible gate result.
`evaluate_simulation_campaign`	`iso_obs.execution`	Verify results and evaluate a complete confirmatory campaign.

Recorded data

Ingestion, synchronization, audit, splitting, and replay.

Relevant SDK classes and functions
Class or function	Import from	Use it to
`DatasetBundle`	`iso_obs.dataset_io`	Content-addressed local dataset artifact with episodes and evidence.
`DatasetBundleInspection`	`iso_obs.dataset_io`	Structural inventory without a quality verdict.
`DatasetIngestionPlan`	`iso_obs.dataset_ingestion`	Declare source mapping, timestamps, channels, scope, and limitations.
`DatasetIngestionReport`	`iso_obs.dataset_ingestion`	Report row, episode, channel, missingness, and typed issues.
`ingest_dataset_source`	`iso_obs.dataset_ingestion`	Ingest JSONL or Parquet using an explicit plan.
`McapDatasetIngestionPlan`	`iso_obs.dataset_mcap`	Declare MCAP topics, encodings, payloads, clocks, and windows.
`ingest_mcap_source`	`iso_obs.dataset_mcap`	Read MCAP in physical order with contract and CRC validation.
`Rosbag2McapIngestionPlan`	`iso_obs.dataset_rosbag2`	Bind split rosbag2 metadata verification to an MCAP plan.
`Rosbag2RecordingEvidence`	`iso_obs.dataset_rosbag2`	Record exact file identity and metadata/recording disagreements.
`ingest_rosbag2_mcap_source`	`iso_obs.dataset_rosbag2`	Ingest a split recording without global sorting or deduplication.
`DatasetSynchronizationPlan`	`iso_obs.dataset_synchronization`	Declare clock, interpolation, gap, and uncertainty requirements.
`DatasetSynchronizationReport`	`iso_obs.dataset_synchronization`	Carry temporal coverage, gaps, uncertainty bounds, and typed issues.
`audit_dataset_synchronization`	`iso_obs.dataset_synchronization`	Assess temporal coverage and propagated uncertainty.
`DatasetReliabilityAuditPlan`	`iso_obs.dataset_reliability`	Declare label, provenance, episode, and evidence requirements.
`DatasetReliabilityReport`	`iso_obs.dataset_reliability`	Carry evidence-qualified label states and review obligations.
`audit_dataset_reliability`	`iso_obs.dataset_reliability`	Audit dataset evidence without rewriting suggestions as truth.
`DatasetSplitPlan`	`iso_obs.dataset_splitting`	Declare assignments, lineage, independence, duplicate groups, and holdouts.
`DatasetSplitReport`	`iso_obs.dataset_splitting`	Carry role counts and blocking, contamination, and review issues.
`audit_dataset_split`	`iso_obs.dataset_splitting`	Detect contamination and unresolved split evidence.
`MixedModeDetectionReport`	`iso_obs.mixed_mode_detection`	Carry uncertain change intervals and before/after prediction sets.
`FailureReplayRequest`	`iso_obs.replay_capsule`	Select one episode interval, assertions, channels, and replay controls.
`FailureReplayCapsule`	`iso_obs.replay_capsule`	Carry reconstruction fidelity, claim qualification, artifacts, and issues.
`compile_failure_replay_capsule`	`iso_obs.replay_capsule`	Join recorded evidence and adapter capabilities into replay readiness.
`build_canonical_regression_case`	`iso_obs.replay_capsule`	Build a canonical case only from qualified exact replay evidence.

Decide and monitor

Promotion, transfer, assurance, and surveillance.

Relevant SDK classes and functions
Class or function	Import from	Use it to
`RegressionPromotionRequest`	`iso_obs.promotion`	Bind the evidence chain proposed for regression promotion.
`RegressionPromotionReview`	`iso_obs.promotion`	Report requirement-by-requirement promotion readiness.
`review_regression_promotion`	`iso_obs.promotion`	Verify artifact identity and readiness requirements.
`SimToRealValidityPlan`	`iso_obs.sim_to_real`	Declare regions, features, anchors, and acceptable gaps.
`SimToRealValidityReport`	`iso_obs.sim_to_real`	Carry regional sim-to-real gaps with simultaneous uncertainty control.
`assess_sim_to_real_validity`	`iso_obs.sim_to_real`	Assess feature and failure-rate gaps with familywise control.
`TransportValidationPlan`	`iso_obs.transportability`	Declare strata, anchor envelope, discrepancy sources, and limits.
`TransportValidationReport`	`iso_obs.transportability`	Carry scope-bound discrepancy evidence within the anchor envelope.
`TransportApplicabilityReport`	`iso_obs.transportability`	Report whether one requested operating point is inside validated scope.
`validate_transportability`	`iso_obs.transportability`	Validate simulator discrepancy within the anchor envelope.
`ReleaseScope`	`iso_obs.assurance`	Bind system, operational domain, evidence envelope, and policy.
`ReleaseAssuranceDossier`	`iso_obs.assurance`	Assemble regressions, independent challenges, and defeaters.
`ReleaseAssuranceReport`	`iso_obs.assurance`	Carry the noncompensatory release disposition, reasons, and open defeaters.
`evaluate_release_assurance`	`iso_obs.assurance`	Return blocked, restricted, or supported within scope.
`SequentialMonitoringPlan`	`iso_obs.surveillance`	Declare anytime-valid production signals and thresholds.
`SequentialSurveillanceReport`	`iso_obs.surveillance`	Carry threshold trajectories and continue-or-escalate disposition.
`evaluate_sequential_surveillance`	`iso_obs.surveillance`	Continue monitoring or escalate review of the scoped claim.

Canonical detail

Continue to exact constructors and validation rules.

SDK guideComplete concepts and worked examples ↗Python sourceExact class fields and function signatures ↗Executable examplesDataset, ROS 2, rosbag2, and replay programs ↗CLI sourceCommands, outputs, and exit behavior ↗