Add Fraud Decision Intelligence model-governance ledger contracts

[mdheller]

Purpose

Add model-governance-ledger support for the Fraud Decision Intelligence capability family defined in ProCybernetica.

Upstream doctrine anchor:

• SocioProphet/ProCybernetica:docs/capabilities/fraud-decision-intelligence-control-plane.md

Required records

Introduce or extend ledger entries for:

• fraud model card;
• fraud benchmark record;
• feature-health record;
• label-provenance record;
• residual-fraud lift record;
• threshold and review-capacity simulation record;
• drift and adversarial-displacement monitoring record;
• production-readiness non-claim boundary.

Claim boundaries

The ledger must not allow:

• model score promoted as proof of fraud;
• Kaggle/public benchmark promoted as production validation;
• weak/delayed label treated as ground truth without provenance;
• human-impacting decision authorized by model artifact alone;
• feature with material missingness drift used without a feature-health record.

Acceptance criteria

• Positive fixture: public-synthetic fraud model card references benchmark, feature-health, label-provenance, and residual-lift records.
• Negative fixture: benchmark metric alone cannot satisfy production readiness.
• Negative fixture: model score alone cannot authorize a denial or enforcement action.
• Negative fixture: label without provenance is rejected.
• Documentation states this repo owns model governance records only, not runtime fraud scoring or policy execution.

[mdheller]

Implementation progress:

• Added schemas/fraud-decision-ledger-record.v0.1.schema.json.
• Added positive fixture examples/fraud-decision-ledger-record.example.json.
• Added invalid benchmark/production fixture examples/fraud-decision-ledger-record.benchmark-production.invalid.json.
• Added invalid score-as-proof fixture examples/fraud-decision-ledger-record.score-as-proof.invalid.json.
• Added stdlib validator tools/validate_fraud_decision_ledger_record.py.
• Added validate-fraud-decision-ledger Makefile lane and wired it into make validate.
• Updated README.md with Fraud Decision Intelligence ledger contracts and invariants.

Commit trail observed from connector:

• Schema: 9f9da9218126bb2246fee3da821000410cdd3769
• Positive fixture: ab603d9ed2b08b81f8d8a0938802b870bf7a9411
• Invalid benchmark/production fixture: 74d3aaf8e4d23529e5d84a2dbaaa63b39d546c7b
• Invalid score-as-proof fixture: a9e45487f128f9b2b85f2613f7a06486f40fb95a
• Validator: d1d7fcc71d283206b781dd1b07bfa6fb7404e050
• Makefile lane: 5f90851e08e9962c1a4b5c2f3924f5e7b60fb171
• README docs: 4606f80d3bf2480ae7c8d2a5032e7667c4508e25

Validation intent:

• Good fixture must pass.
• Benchmark-only production readiness fixture must fail.
• Score-as-proof / approval-without-provenance fixture must fail.

Remaining before closure:

• Observe make validate-fraud-decision-ledger green.
• Observe full make validate green.