FPF Reference

    Bias Audit and Ethical Assurance

    D.5 · Stable · D-family bias-audit and ethical-assurance boundary pattern · Part D - Multi-scale Ethics and Conflict Optimization

    About this pattern

    This is a generated FPF pattern page projected from the published FPF source. It is canonical FPF content for this ID; it is not a FPF Reference product feature page.

    How to use this pattern

    Read the ID, status, type, and normativity first. Use the content for exact wording, the relations for adjacent concepts, and citations to keep active work grounded without pasting the whole specification.

    Type: D-family bias-audit and ethical-assurance boundary pattern Status: Stable Pattern role: This compact pattern owns bias, fairness, impact-audit, causal-fairness audit consumption, and ethical-assurance boundary use; it does not replace D.1 through D.4.

    Use this when. Use this pattern when a model, metric, policy, publication, decision system, recommendation, method, work plan, system, holon, or FPF claim may create bias, unfairness, human or group impact, causal-fairness overclaim, or ethical assurance risk.

    Not this pattern when. If the ethical value frame is missing, use D.1. If the current question is multilevel ethics entry, use D.2. If the current question is to describe the sides and tension of an interlevel ethical conflict, use D.3. If the current question is mediation or decision use of that conflict, use D.4. If the current question is only evidence, causality, assurance, measurement, or architecture residual without bias, fairness, human or group impact, or ethical assurance, use the direct owner.

    What goes wrong if missed. A model, metric, policy, publication, or decision system passes ordinary evidence or assurance checks while representation, proxy, visibility, metric, language, or human-impact bias remains hidden.

    What this buys. Bias, fairness, human-impact, causal-fairness, and ethical-assurance concerns become auditable without replacing D.1 through D.4, evidence, causal, measurement, or architecture owners.

    Bias and fairness failures often survive ordinary verification. A metric may be accurate while hiding subgroup harm. A model may be predictive while reproducing past exclusion. A policy may look neutral while moving cost to people or groups who were not represented in the evidence. A publication may look technically clear while licensing a harmful use.

    Relations

    D.5builds onEthical Value Plurality and FPF Boundary
    D.5builds onEthical Mediation and Decision Use
    D.5builds onCross-Disciplinary Bias Audit
    D.5coordinates withEvidence Graph Referring (C-4)
    D.5coordinates withTrust and Assurance Calculus
    D.5coordinates withMeasurement & Metrics Characterization (MM‑CHR)
    D.5coordinates withCausalUse-CAL: Causal-Use Questions, Identification, and Realizability
    D.5coordinates withPragmatic Utility and Value Alignment
    D.5explicit referenceEthical Value Plurality and FPF Boundary
    D.5explicit referenceEthical Mediation and Decision Use
    D.5explicit referenceMeasurement & Metrics Characterization (MM‑CHR)
    D.5explicit referenceCausalUse-CAL: Causal-Use Questions, Identification, and Realizability
    D.5explicit referenceEvidence Graph Referring (C-4)
    D.5explicit referenceTrust and Assurance Calculus
    D.5explicit referenceMultilevel Ethics For System-Holon Work
    D.5explicit referenceInterlevel Ethical Conflict Structure
    D.5explicit referenceMulti‑View Publication Kit

    Content

    Problem Frame

    Bias and fairness failures often survive ordinary verification. A metric may be accurate while hiding subgroup harm. A model may be predictive while reproducing past exclusion. A policy may look neutral while moving cost to people or groups who were not represented in the evidence. A publication may look technically clear while licensing a harmful use.

    D.5 keeps this audit and assurance question explicit. It does not replace multilevel ethics. It asks whether the current object and its intended use are ethically unsafe because of bias, unfairness, impact, causal fairness without the required C.28 evidence value, or assurance without the required assurance relation.

    Problem

    Bias, fairness, human-impact, causal-fairness, and ethical-assurance concerns can remain invisible after ordinary technical verification. The failure is to let the model, metric, policy, publication, method, work plan, system, or holon be treated as admissible for use while the audited EntityOfConcern, intended use, affected people or groups, evidence, mitigation, and residuals are not explicit.

    Forces

    ForceTension
    Ordinary verification vs. subgroup harmEvidence or accuracy can look strong while representation, proxy, metric, visibility, language, or impact bias remains current.
    Lightweight scan vs. consequential useLocal reversible use may need a small register, while release-bearing or repeated use needs a fuller audit report.
    Fairness wording vs. causal claimMetric disparity, associative fairness, interventional fairness, and counterfactual fairness are different claims.
    Assurance relation vs. ethical permissionAssurance can record examined evidence and residuals, but cannot turn unresolved bias or harm into moral authorization.
    Audit frame vs. neighboring ownersD.5 must keep bias and ethical assurance visible without replacing evidence, causality, measurement, architecture, or D.1 through D.4.

    Solution

    Open a BiasAuditAssuranceFrame@Context:

    BiasAuditAssuranceFrame@Context:
  auditedEntityOfConcernRef
  intendedUseRef
  claimScopeRef?: U.ClaimScope
  qualificationWindowRef?
  affectedPopulationRefs?
  affectedSystemRefs?
  affectedHolonRefs?
  metricOrModelRefs?
  policyOrPublicationRefs?
  biasConcernRefs
  ethicalClaimRefs?
  fairnessClaimRef?
  impactClaimRef?
  causalFairnessUseRef?
  causalUseSupportResultRef?: CausalUseSupportResultRef
  evidenceRefs
  assuranceClaimRefs?
  assuranceUseRef?
  mitigationOrConstraintRefs?
  acceptedResidualRefs?
  admissibleUse
  inadmissibleOverread
  strongerSourceReturnCondition

    The frame is not a universal ethics owner. It is the local audit object used when bias, fairness, impact, or ethical assurance is current.

    Bias and Fairness Recognition

    Current claimWhat D.5 requiresNeighboring owner
    "This metric shows the system is fair."Distinguish metric disparity, proxy choice, subgroup impact, and intended use.C.16 for metric construction
    "This intervention makes outcomes fair."Declare the causal fairness use, C.28 support components and causal-use support result.C.28
    "The model is unbiased."Name represented and missing groups, data-generation limits, model-use limits, and evidence.A.10, C.16, D.5
    "The release is ethically assured."Separate audit findings, mitigations, accepted residuals, and the assurance or evidence relation.B.3, D.5
    "The policy is acceptable because it helps the whole."Check whether a multilevel conflict is live.D.2, D.3, D.4

    Optional Audit Records And Depth

    D.5 may use a compact BiasRegister@Context when the live need is to keep concerns visible during ordinary work:

    BiasRegister@Context:
  auditedEntityOfConcernRef
  intendedUseRef
  claimScopeRef?: U.ClaimScope
  qualificationWindowRef?
  affectedPopulationRefs?
  affectedSystemRefs?
  biasConcernCode
  evidenceRefs
  mitigationOrConstraintRef?
  acceptedResidualRef?
  nextReviewTrigger?

    Use a fuller BiasAuditReport@Context when the object is being released, relied on by other work, exposed to affected people or groups, or used for assurance. Also use it after a material change to source currentness, affected population, ClaimScope or qualification window, model, metric, or policy. The report is a Description episteme or publication-use object; it does not make the audited object fair by existing.

    Lightweight scan is enough when the intended use is local, reversible, low-impact, and the scan finds no affected group, proxy, metric, representation, causal-use, or publication-use concern. Use deeper review when the use is consequential, repeated, automated, externally published, safety-relevant, or regulatorily or deontically constrained. Deeper review is also required when the use crosses a ClaimScope, qualification window, population, or publication boundary, or when an affected group, missing group, proxy variable, threshold, causal fairness claim, accepted residual, or assurance claim is current.

    Compact Bias Concern Taxonomy

    CodeConcernTypical question
    REPRepresentation, coverage, sampling, proxy choice, missing group, or shifted population.Who or what is missing, over-weighted, proxied, or moved out of scope?
    ALGAlgorithmic, modeling, objective, ranking, optimization, or threshold behavior.Which model or optimization choice changes outcomes for whom?
    VISVisibility, interface, dashboard, presentation, or publication framing.What becomes easy to see, hard to see, or too authoritative by display?
    METMetric, measurement, scale, comparator, normalization, or threshold.What does the metric count, hide, compare, or turn into a pass or fail claim?
    LNGLanguage, naming, category, definition, group label, or claim wording.Which words change what can be asserted, counted, blamed, or done?

    The codes are only concern locators. They do not replace the governed object, affected people or groups, intended use, evidence, mitigation, or accepted residual.

    Causal Fairness Boundary

    A fairness claim may be associative, interventional, or counterfactual. C.28 supplies the causal-use question, rung, estimand, separate support components, common-threat result, and CausalUseSupportResultRef. D.5 keeps the bias/fairness audit and its conclusion.

    When counterfactual fairness is consequential, reusable, published, or used for assurance, the cited C.28 components expose the additional counterfactual-identifiability assumptions required for that question. If the audit relies on an estimated fairness result, they also expose the estimate and its estimation-consistency result. Missing identification or consistency lowers the C.28 result to bounded or unsupported; more of the same data does not repair either gap.

    Cite the C.28 result from the existing BiasAuditReport@Context. Do not open a separate C.28 fairness card; D.5 defines no such output. Metric-only fallback remains cheaper: when only metric disparity is claimed, record the metric or evaluation result and stop. An interventional proxy may support a bounded interventional fairness statement, but it does not establish counterfactual fairness without the required estimand and support components.

    The C.28 result is one evidence basis. It does not certify fairness, approve a release, or supply ethical assurance; D.5 and any downstream decision or assurance pattern make those separate conclusions.

    Ethical Assurance Boundary

    Ethical assurance is not a stamp of moral permission. It is an assurance claim that bias, fairness, impact, and accepted residuals have been examined for the current use.

    Use B.3 for the assurance relation. Use A.10 for evidence provenance and source currentness. Use D.3 to describe an interlevel ethical conflict and D.4 to mediate or use it in a decision. Use C.30.ILC when the issue is an architecture residual rather than a bias or fairness audit.

    Archetypal Grounding (Worked Slice)

    A hiring-screening model has high aggregate accuracy and an internal note says it is "fair." D.5 first asks what fairness claim is being made. If the claim is only a metric disparity comparison, the audit records the metric, affected groups, intended use, missing evidence, and admissible use. If the team claims counterfactual fairness, C.28 must expose the causal estimand, additional counterfactual-identifiability assumptions, and an estimate with its consistency result when that estimate is relied on. Missing conditions lower the C.28 support result before D.5 decides its audit use. If the audit exposes a conflict between company efficiency and applicant harm across declared scopes, D.3 describes that conflict and D.4 guides its decision use.

    Bias-Annotation

    Bias riskFailureMitigation
    Audit as document ritualA register or report exists but does not change intended use, residuals, or constraints.Tie each concern to audited EntityOfConcern, intended use, evidence, mitigation, and accepted residual.
    Metric fairness overclaimA metric comparison is published as causal or counterfactual fairness.Recover the fairness claim kind. For counterfactual fairness, require C.28 identification assumptions and estimation consistency when an estimate is used before D.5 consumes the support result.
    Assurance as authorizationEthical assurance is treated as permission to proceed.Record assurance as assurance or evidence relation and keep D.4 and D.5 use separate.
    Bias category replaces objectREP, ALG, VIS, MET, or LNG code is treated as the governed object.Use codes only as concern locators; keep audited EntityOfConcern and intended use explicit.

    Conformance Checklist

    IDRequirementPurpose
    CC-D5-1The audited EntityOfConcern, intended use, any affected populations or Systems, bias, fairness, impact, or ethical claim, evidence, assurance use when current, repair return, and admissible use are named. ClaimScope and qualification window are explicit when they delimit the audit.Keeps audit scope inspectable.
    CC-D5-2Metric, causal fairness, evidence, assurance, publication, and architecture-residual claims use their direct owners.Prevents D.5 from swallowing neighboring patterns.
    CC-D5-3Ethical assurance is recorded as assurance or evidence relation, not moral permission.Keeps assurance from becoming ethical authorization.
    CC-D5-4If the audit exposes interlevel conflict, use D.3 for the conflict description and D.4 for mediation or decision use.Keeps D.5 connected to the D cluster without replacing it.

    Common Anti-Patterns and How to Avoid Them

    Anti-patternWhat goes wrongRepair
    Ethics ghettoBias or fairness is left in a separate ethics note while the model, metric, release, publication, or work plan keeps operating unchanged.Put the concern on the audited EntityOfConcern and its intended use, then name the mitigation, constraint, or accepted residual.
    Checklist charadeA checklist is completed without naming affected people or groups, evidence, current use, or residuals.Use BiasRegister@Context for a light scan or BiasAuditReport@Context for deeper review; do not treat a blank checklist as assurance.
    Bias whack-a-moleOne disparity is patched while proxy, representation, metric, visibility, or language concerns move elsewhere.Keep REP, ALG, VIS, MET, and LNG concerns visible until the admissible use and accepted residual are explicit.

    Consequences

    This pattern keeps bias, fairness, impact, causal-fairness audit consumption, and ethical assurance from being scattered across technical patterns. It also prevents D.5 from swallowing all ethics. The cost is that teams must say which bias or fairness claim they are making. The gain is that ethical assurance becomes a typed assurance or evidence claim rather than a comforting label.

    Rationale

    D.5 exists because bias, fairness, human-impact, causal-fairness audit consumption, and ethical assurance often survive ordinary technical checks. It keeps those concerns in one audit frame while preserving direct owners: metrics and measurement remain with measurement patterns, causal fairness remains with causal-use patterns, assurance remains an assurance relation, and multilevel ethical conflict remains with D.2 through D.4.

    Audit record depth is selected by use, reliance, exposure, source currentness, and residual risk. A compact register is enough for local low-impact use when no live concern remains; a fuller report is required when release, reliance, affected people or groups, source-currentness change, causal fairness, accepted residual, or assurance use is current.

    SoTA-Echoing

    Source linePractical implication for this pattern
    Fairness and bias audit practiceRepresentation, proxy, metric, visibility, language, and impact concerns must be tied to intended use, affected groups, source-currentness, and accepted residuals.
    Causal fairness and causal inferenceAssociative, interventional, and counterfactual fairness claims need different evidence values and cannot be interchanged by wording.
    Assurance and governance practiceAn assurance record can support bounded reliance, but does not grant moral permission under unresolved residual harm or replace D.3 and D.4 when interlevel conflict is exposed.
    FPF episteme and publication disciplineBias registers and reports are descriptions or publication-use objects; they do not make the audited object fair by existing.

    Relations

    • Builds on D.1 and coordinates with D.2, D.3, and D.4 for value frame, multilevel entry, conflict description, and mediation or decision use.
    • Coordinates with A.10 for evidence and source currentness.
    • Coordinates with B.3 for assurance relation and reliance.
    • Coordinates with C.16 for metric and measurement construction.
    • Coordinates with C.28 for causal fairness, including counterfactual-identification assumptions, estimation consistency when an estimate is used, and the bounded causal-use support result.
    • Coordinates with E.17 when publication or publication-use relation changes admissible use.

    D.5:End

    Last Updated: 2026-08-30 — upstream FPF commit e400eab3 (github.com/ailev/FPF)