44-EFT.WP.Data.ModelCards v1.0 | Chapter 14 Fairness, Ethics & Usage Restrictions

Home ／ Docs-Technical WhitePaper ／ 44-EFT.WP.Data.ModelCards v1.0

Chapter 14 Fairness, Ethics & Usage Restrictions

I. Chapter Purpose & Scope

II. Terminology & Dependencies

• Dependencies: EFT.WP.Core.DataSpec v1.0 (contracts/exports); EFT.WP.Core.Metrology v1.0 (units & dimensional checks); EFT.WP.Data.DatasetCards v1.0 (privacy, ethics & compliance; coverage/splits).
• Math & symbols: Fairness disparities use gap_metric (e.g., abs_diff or ratio); inline symbols like Δ_gap must be wrapped in backticks; any division/integral/composite operator must use parentheses; no Chinese in formulas/symbols/definitions.

III. Fields & Structure (Normative)

fairness:

axes: ["class","region","device","language"]

gap_metric: "abs_diff|ratio"

threshold: 0.05

stratification: ["val","test"]

mitigation:

enabled: true

methods: ["reweight","resample","calibration","post-hoc-threshold"]

reeval_required: true

reporting:

include_ci: true

table_axes: ["axis","bucket","metric"]

significance: {test:"bootstrap", alpha:0.05}

ethics:

intended_use: ["academic","benchmark","safety-research"]

prohibited_use: ["surveillance","biometric_identification","unlawful_discrimination"]

disclosures:

sensitive_attributes: ["N/A"]

human_in_the_loop: true

risk_notes: "Model outputs must be reviewed in high-stakes contexts."

governance:

review_process: ["internal-ethics-board"]

update_policy: "on-drift|on-incident|quarterly"

usage:

regional_compliance: ["EU-GDPR","US-CCPA"]

access_control:

roles: ["owner","maintainer","reader"]

enforcement: ["signed-url","token","ip-allowlist"]

rate_limits:

qps_max: 1000

burst: 200

monitoring:

online_checks: ["fairness_gap","drift_kl","error_rate"]

alert_rules:

- {name:"fairness_gap_breach", rule:"Δ_gap>0.05 for 60m", severity:"high"}

IV. Fairness Evaluation & Thresholds

1. Axes & buckets: axes declare evaluated dimensions (class/region/device/language, etc.) aligned with Dataset Card coverage; report per-bucket sample counts and 95% CIs.
2. Disparity metrics:
   • Absolute gap: Δ_gap = ( metric_ref - metric_grp );
   • Ratio gap: Δ_gap = ( metric_grp / metric_ref );
     Exceeding threshold is blocking or requires mitigation plus re-evaluation.
3. Mitigation & re-eval: Record mitigation.methods (reweight/resample/calibration/post-hoc threshold) with reeval_required=true; re-evaluate per Chapter 11 (seeds/repeats/ci/significance).

V. Ethical Disclosures & Human-in-the-Loop

• Use boundaries: intended_use/prohibited_use make scope explicit; set human_in_the_loop=true and describe review stages for high-stakes contexts.
• Sensitive attributes: If applicable, disclose collection/processing/de-identification; prohibit inference or decision-making on non-consented sensitive attributes.
• Governance: Record internal ethics review via governance.review_process; define update_policy and incident triggers.

VI. Usage Limits & Online Monitoring

• Regional compliance: regional_compliance mirrors the Dataset Card; if cross-border transfer applies, include mechanism/template IDs in the export manifest.
• Access & rate: Specify access_control and rate_limits; production monitoring uses monitoring.online_checks and alert_rules; threshold breaches must trigger degradation or block.

VII. Metrology & Units (for performance/latency/power)

VIII. Machine-Readable Fragment (Drop-in)

fairness:

axes: ["class","region","device"]

gap_metric: "abs_diff"

threshold: 0.05

stratification: ["test"]

mitigation: {enabled:true, methods:["reweight","calibration"], reeval_required:true}

reporting: {include_ci:true, table_axes:["axis","bucket","metric"], significance:{test:"bootstrap", alpha:0.05}}

ethics:

intended_use: ["academic","benchmark"]

prohibited_use: ["surveillance","biometric_identification"]

disclosures: {sensitive_attributes:["N/A"], human_in_the_loop:true, risk_notes:"Human review required in high-stakes."}

governance: {review_process:["internal-ethics-board"], update_policy:"on-drift"}

usage:

regional_compliance: ["EU-GDPR"]

access_control: {roles:["owner","maintainer","reader"], enforcement:["signed-url","token"]}

rate_limits: {qps_max: 1000, burst: 200}

monitoring:

online_checks: ["fairness_gap","drift_kl"]

alert_rules: [{name:"fairness_gap_breach", rule:"Δ_gap>0.05 for 60m", severity:"high"}]

IX. Consistency with Evaluation, Training Data & Deployment

• Fairness evaluation uses frozen splits and stratification mapped to Dataset Card coverage.
• If mitigation is applied (reweighting/post-hoc thresholds), mirror it in optimization/hyperparams and evaluation, and run significance tests.
• Online monitoring and deployment endpoints are specified via Chapter 16 API bindings (OpenAPI snippets).

X. Export Manifest & Audit Trail

export_manifest:

artifacts:

- {path:"fairness/by_axis_metrics.csv", sha256:"..."}

- {path:"fairness/mitigation_report.md", sha256:"..."}

- {path:"usage/alert_rules.yaml", sha256:"..."}

references:

- "EFT.WP.Core.DataSpec v1.0:EXPORT"

- "EFT.WP.Core.Metrology v1.0:check_dim"

- "EFT.WP.Data.DatasetCards v1.0:Ch.13"

XI. Chapter Compliance Checklist

• axes/gap_metric/threshold are explicit and aligned with Dataset Card stratification; per-bucket counts and 95% CIs reported with significance tests.
• When thresholds trigger, mitigation is applied and re-evaluated; results meet targets or residual risks and usage limits are disclosed.
• Ethical disclosures and human-in-the-loop process recorded; prohibited uses are clear and enforced at deployment.
• Regional compliance, access control, and online monitoring rules complete; performance/energy units pass check_dim.
• Export manifest lists tables/reports/alert configs with sha256; references carry “Volume vX.Y:Anchor.”