Assemble a release argument without manufacturing a confidence score.

Release evidence

A chain of qualified questions—not one score.

1. 01
   Promote

   Verify that a reproduced failure is coherent enough to become a durable regression asset.

2. 02
   Execute

   Run the promoted pack against the proposed system version and complete every declared gate.

3. 03
   Validate transfer

   Compare simulation with real anchors inside predeclared regions and tolerances.

4. 04
   Challenge

   Bring independent evidence designed to find what the known regressions may have missed.

5. 05
   Assemble

   Bind regressions, challenges, scope, policy, and open defeaters into one dossier.

6. 06
   Monitor

   Tie post-deployment signals back to the exact scoped assurance report.

Regression promotion

Qualify the asset before using it as a gate.

Discovery regressions require coherent provenance. Release-gate candidates additionally require a meaningful replicated intervention association, one-minimal counterexample, simulation validation, a real-world anchor, and the correct canonical case.

from iso_obs.promotion import (
    PromotionIntent,
    RegressionPromotionRequest,
    review_regression_promotion,
)

request = RegressionPromotionRequest(
    request_id="promote-blind-intersection",
    request_version="1",
    intent=PromotionIntent.RELEASE_GATE_CANDIDATE,
    source_failure_content_digest=failure.content_digest(),
    counterfactual_design_digest=design.content_digest(),
    counterfactual_report_digest=counterfactual_report.content_digest(),
    minimal_counterexample_report_digest=minimal.content_digest(),
    regression_pack_digest=regression_pack.content_digest(),
    condition_bindings=condition_bindings,
    limitations=("Requires execution against the proposed version.",),
)

review = review_regression_promotion(
    request,
    source_failure=failure,
    counterfactual_design=design,
    counterfactual_report=counterfactual_report,
    minimal_counterexample=minimal,
    regression_pack=regression_pack,
)

External validity

State where simulation evidence transfers—and where it does not.

Use SimToRealValidityPlan for region-specific feature and failure-rate gap tolerances. Use TransportValidationPlan when the question is whether simulator discrepancy remains bounded within a declared anchor envelope.

Release assurance

Let the weakest blocking evidence govern.

Every required failure fingerprint must link to a qualified promotion and a release-eligible campaign. Independent passing challenge evidence is required for support within scope.

from iso_obs.assurance import (
    ReleaseAssuranceDossier,
    evaluate_release_assurance,
)

dossier = ReleaseAssuranceDossier(
    dossier_id="policy-v18-warehouse-release",
    dossier_version="1",
    release_scope=release_scope,
    policy=assurance_policy,
    regression_evidence=known_failure_evidence_links,
    challenge_evidence=independent_challenge_evidence,
    defeaters=declared_defeaters,
    limitations=("Human release authority remains accountable.",),
)

report = evaluate_release_assurance(
    dossier,
    promotion_reviews=promotion_reviews,
    campaign_reports=completed_campaign_reports,
)

Post-deployment

Escalate when new evidence challenges the scoped claim.

Sequential monitoring uses preregistered e-processes whose thresholds may be checked after every observation. Historical threshold crossings remain visible even if the current evidence value later falls.

from iso_obs.surveillance import (
    SequentialMonitoringPlan,
    evaluate_sequential_surveillance,
)

report = evaluate_sequential_surveillance(
    monitoring_plan,
    assurance_report,
    ordered_observations,
)

Relevant API

Classes and capabilities in the release chain.