46-EFT.WP.Data.Benchmarks v1.0 | Chapter 14 Privacy, Security & Compliance (Benchmark-side)

Home ／ Docs-Technical WhitePaper ／ 46-EFT.WP.Data.Benchmarks v1.0

Chapter 14 Privacy, Security & Compliance (Benchmark-side)

I. Chapter Purpose & Scope

II. Terminology & Dependencies

1. Terms: data_minimization, deidentification, k_anon, ε_dp, lawful_basis, data_residency, DLP, KMS, RBAC/ABAC, mTLS, SSE-KMS, BYOK, DPAs, SCCs, incident/IRP.
2. Dependencies: privacy & compliance (Pipeline v1.0, Ch.14), evaluation protocol (ModelCards v1.0, Ch.11), metrics & units (this volume, Ch.6), scoring gates (this volume, Ch.8), units & dimensions (Core.Metrology v1.0:check_dim).
3. Math & symbols: wrap inline symbols; any division/integral/composite operator must use parentheses; for path quantity T_arr use
   • T_arr = ( 1 / c_ref ) * ( ∫ n_eff d ell ), or
   • T_arr = ( ∫ ( n_eff / c_ref ) d ell ),
     declaring gamma(ell) and d ell. No Chinese in formulas/symbols/definitions.

III. Fields & Structure (Normative)

benchmark_compliance:

privacy:

policy: "no-PII|limited-PII|special-category"

lawful_basis: ["consent","contract","legitimate_interest","research"]

data_minimization: true

pii_inventory: ["<fieldA>","<fieldB>"]

deidentification:

methods: ["hash-id","mask","truncate","generalize","noise"]

k_anon: 10

l_diversity: 2

ε_dp: null

retention:

policy: "min-necessary"

delete_after_days: 365

data_residency: ["EU","US"]

dlp:

enabled: true

rules: ["creditcard","ssn","email"]

security:

encryption:

at_rest: "SSE-KMS|AES-256"

in_transit: "TLS1.2+"

kms: {provider:"cloud-kms|hsm", byok:true}

access_control:

model: "RBAC|ABAC"

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

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

audit_log: true

network:

segmentation: ["private-subnet","sg-allowlist"]

egress_policy: "deny-by-default"

secrets:

manager: "vault|cloud-secrets"

rotation_days: 90

hardening:

container: ["non-root","readonly-rootfs","seccomp","no-new-privs"]

artifact_signing: true

submissions:

payload:

required_artifacts: ["reports/*.jsonl","env.lock","protocol.yaml","metrics.yaml"]

checksum: "sha256"

max_retention_days: 365

handling:

quarantine_on_pii: true

reviewer_roles: ["maintainer","reviewer"]

redaction_policy: "hash-or-drop"

compliance:

regions: ["EU-GDPR","US-CCPA","CN-DSL"]

data_transfer:

mechanisms: ["SCCs","intra-region-only"]

third_parties:

processors: ["<vendorA>@v1.0"]

dpas_signed: true

incident_response:

contact: "security@org.example"

sla_hours: 72

runbook_ref: "security/irp.md"

audits:

schedule: "annual|quarterly"

artifacts: ["privacy/pii-scan.txt","security/pen-test.md","compliance/dpia.md"]

IV. De-identification & Data Minimization

• Minimization: collect only fields necessary for benchmark goals and reproducibility; reconcile pii_inventory with task contracts and review periodically.
• De-identification: combine hash-id|mask|truncate|generalize|noise as needed; if DP is used, record ε_dp, scope, and risk notes; include k_anon/l_diversity parameters and validation evidence in exports.
• DLP & compliance scanning: enable DLP on submissions; hits go to quarantine and are processed per redaction_policy before evaluation.

V. Licensing, Residency & Cross-Border

• License mapping: licensing aligns with distribution/mirroring; propagate restrictive clauses to task policies.
• Residency: data_residency limits storage/processing regions; record anchors for SCCs or equivalents for transfers.
• Third-party processing: register processors and DPA status; make data flows traceable in lineage.

VI. Access Control & Submission Handling

• Access: RBAC/ABAC + mTLS/allowlist; submission channels use short-lived credentials and least privilege; sensitive ops (retrieve/delete/export) must be audited.
• Artifact handling: verify all payloads with sha256; auto-delete at end of retention; PII detection triggers isolation/remediation with full evidence chain.

VII. Incident Response & Governance

• Severity: info/minor/major; policy violations or leaks are major.
• SLA: respond and contain within sla_hours; IRP must record confirm/isolate/root cause/remediate/re-test.
• Appeal & arbitration: provide appeal windows and review procedures; publish changes on the benchmark portal.

VIII. Coupling with Scoring/Gates/Leaderboard

• Fairness/harm/compliance blocking takes precedence over performance promotion.
• Scoring/gates (Ch.8) apply after compliance passes; remediation requires re-evaluation and public updates.

IX. Metrology & Units (SI)

• Mandatory: metrology:{units:"SI", check_dim:true}; report performance/energy/volume/network in QPS(1/s), T_inf(ms), ρ(—), size_bytes, net_mbps; normalize units first before composition.
• Path quantities: if privacy/security flows involve T_arr (e.g., arrival-time checks/rewrites), register delta_form/path/measure and validate using either equivalence.

X. Machine-Readable Fragment (Drop-in)

benchmark_compliance:

privacy:

policy: "limited-PII"

lawful_basis: ["consent","research"]

data_minimization: true

pii_inventory: ["user_id_hash","email_hash"]

deidentification: {methods:["hash-id","mask"], k_anon:20, l_diversity:2, ε_dp:null}

retention: {policy:"min-necessary", delete_after_days:180}

data_residency: ["EU"]

dlp: {enabled:true, rules:["email","creditcard"]}

security:

encryption: {at_rest:"SSE-KMS", in_transit:"TLS1.2+", kms:{provider:"cloud-kms", byok:true}}

access_control: {model:"RBAC", roles:["owner","maintainer","reviewer","reader"], enforcement:["token","ip-allowlist","mTLS"], audit_log:true}

network: {segmentation:["private-subnet"], egress_policy:"deny-by-default"}

secrets: {manager:"vault", rotation_days:90}

hardening: {container:["non-root","readonly-rootfs","seccomp","no-new-privs"], artifact_signing:true}

submissions:

payload:

required_artifacts: ["reports/summary.json","env.lock","protocol.yaml","metrics.yaml"]

checksum: "sha256"

max_retention_days: 365

handling: {quarantine_on_pii:true, reviewer_roles:["maintainer","reviewer"], redaction_policy:"hash-or-drop"}

compliance:

regions: ["EU-GDPR"]

data_transfer: {mechanisms:["SCCs"]}

third_parties: {processors:["processorA@v1.0"], dpas_signed:true}

incident_response: {contact:"security@org.example", sla_hours:72, runbook_ref:"security/irp.md"}

metrology: {units:"SI", check_dim:true}

XI. Lint Rules (Excerpt, Normative)

lint_rules:

- id: PRIV.POLICY_ALLOWED

when: "$.benchmark_compliance.privacy.policy"

assert: "value in ['no-PII','limited-PII','special-category']"

level: error

- id: PRIV.MINIMIZATION_ON

when: "$.benchmark_compliance.privacy.data_minimization"

assert: "value == true"

level: error

- id: PRIV.DPI_PARAMS

when: "$.benchmark_compliance.privacy.deidentification"

assert: "has_key('methods') and (has_key('k_anon') or has_key('ε_dp'))"

level: error

- id: SEC.ENCRYPTION_REQUIRED

when: "$.benchmark_compliance.security.encryption"

assert: "value.at_rest in ['SSE-KMS','AES-256'] and value.in_transit >= 'TLS1.2+'"

level: error

- id: SUBM.ARTIFACTS_REQUIRED

when: "$.benchmark_compliance.submissions.payload.required_artifacts"

assert: "len(value) >= 1"

level: error

- id: COMP.REGIONS_ALLOWED

when: "$.benchmark_compliance.compliance.regions[*]"

assert: "value in ['EU-GDPR','US-CCPA','CN-DSL']"

level: error

- id: IR.SLA_DEFINED

when: "$.benchmark_compliance.compliance.incident_response.sla_hours"

assert: "is_number(value) and value > 0"

level: error

- id: METROLOGY.SI_AND_CHECKDIM

when: "$.metrology"

assert: "units == 'SI' and check_dim == true"

level: error

XII. Cross-Reference Anchors

• Evaluation protocol & runtime environment: EFT.WP.Data.ModelCards v1.0, Ch.11; this volume, Ch.10.
• Fairness, ethics & safety stress: this volume, Ch.13.
• Pipeline-side privacy & compliance: EFT.WP.Data.Pipeline v1.0, Ch.14.
• Units & dimensions: EFT.WP.Core.Metrology v1.0:check_dim.

XIII. Chapter Compliance Checklist

• De-identification & minimization strategies, evidence (k_anon/ε_dp), and pii_inventory complete and verifiable.
• Encryption, access control, network isolation, and key management (with rotation) active; sensitive ops auditable.
• Residency/cross-border & third-party processing registered; DPAs/SCCs anchors present in exports.
• Submission workflow & artifact handling compliant: required artifacts, checksums, quarantine & retention policies explicit.
• SI metrology with check_dim=true; if T_arr appears, delta_form/path/measure registered and validated.
• Machine-readable fragment is drop-in and lint-clean; export_manifest.references[] use “Volume vX.Y:Anchor.”