v0.5.1 — Site, benchmark, paper, and discovery release

v0.5.1 — Site, benchmark, paper, and discovery release

This release packages C-DAG for public evaluation and discovery.

Added

• Public site support for https://cdag.quest
• Benchmark / Evidence Dashboard
• Clean benchmark URL support: /benchmark
• docs/paper.md paper source
• robots.txt, sitemap.xml, llms.txt, and llms-full.txt
• Social preview / feature image metadata

Updated

• README with public site, benchmark, DOI, and paper links
• Replay wording changed from 1.0 to:
  100% success on sampled validation audit records across reported validation runs
• Public benchmark language cleaned to avoid missing-measurement placeholders

Verification

• 59 passed, 1 warning
• CLI smoke passed
• Default reference risk remains 0.849375
• Default decision remains DECLINE
• No core inference math changed
• No Python/package code changed for this release

Boundaries

C-DAG remains a public reference implementation and workflow evidence benchmark. It is not a production lending model, live lending-decision system, legal/regulatory certification tool, or guaranteed financial-outcome system.

v0.5.0 - CRT and CAS validation expansion

v0.5.0

This release expands C-DAG with real CRT and CAS validation paths while preserving the original reference model behavior.

Added

• Public CRT validation path
• Freddie/STACR-style loan-level validation support
• Fannie CAS April 2026 loan-level validation support
• CAS external header-file parsing
• Streaming CAS ingestion for large 3GB+ files
• CAS-specific calibrated policy configuration
• Real CRT validation report updates
• Public mortgage holdout validation baseline
• C-DAG public-facing naming update while preserving causal-credit-risk-engine as the formal implementation name

Validation examples

Freddie/STACR-style CRT validation:

• Rows processed: 10,000
• Accepted rows: 10,000
• Rejected rows: 0
• Decision distribution: APPROVE 9,299 / REVIEW 0 / DECLINE 701
• Replay success rate: 1.0
• Audit-chain verification: true
• Evidence-pack mode: sampled, 1,000 rows

Fannie CAS April 2026 validation:

• Rows processed: 10,000
• Accepted rows: 10,000
• Rejected rows: 0
• Base policy distribution: APPROVE 7,722 / REVIEW 2,278 / DECLINE 0
• Calibrated CAS policy distribution: APPROVE 7,948 / REVIEW 686 / DECLINE 1,366
• Replay success rate: 1.0
• Audit-chain verification: true

Mortgage holdout validation baseline:

• Train rows: 58,579
• Test rows: 41,421
• Test positives: 201
• AUC improved from 0.538963 to 0.573062
• PR-AUC improved from 0.005601 to 0.006059
• Decision distribution after calibration: APPROVE 36,336 / REVIEW 5,085 / DECLINE 0
• Replay success rate: 1.0
• Audit-chain verification: true

Unchanged

• Core inference math unchanged
• Reference demo model unchanged
• Default example risk remains 0.849375
• BUSL-1.1 license posture unchanged
• No raw data or validation outputs are tracked

Important limitation

This is public financial dataset validation, not production validation. It does not use customer data, does not make credit eligibility decisions, and does not prove regulatory compliance.

v0.4.0 - public financial validation expansion

v0.4.0

This release expands causal-credit-risk-engine with public financial dataset validation paths across mortgage performance, mortgage application, and consumer complaint governance workflows.

Added

• Public institutional mortgage validation path
• Freddie Mac single-family loan-level validation support
• Fannie Mae historical performance validation support
• HMDA / CFPB modified LAR validation support
• Public mortgage model and policy configuration
• CFPB Consumer Complaint validation path
• Complaint escalation / response-risk governance model
• Sampled evidence-pack support for large validation runs
• Progress output for long validation workflows
• Additional validation tests

Validation examples

Mortgage validation example:

• Rows processed: 30,000
• Accepted rows: 30,000
• Rejected rows: 0
• Datasets used: Freddie Mac, Fannie Mae, HMDA/CFPB
• Decision distribution: APPROVE 9,584 / REVIEW 4,368 / DECLINE 16,048
• Replay success rate: 1.0
• Audit-chain verification: true
• Evidence-pack mode: sampled, 1,000 rows

CFPB complaint validation example:

• 10,000 real CFPB complaint rows processed
• Decision distribution: APPROVE 0 / REVIEW 9,837 / DECLINE 163
• Replay success rate: 1.0
• Audit-chain verification: true

Unchanged

• Reference demo model remains unchanged
• Default example risk remains 0.849375
• Core inference math unchanged
• Existing BUSL-1.1 license posture unchanged

Important limitation

This is public financial dataset validation, not production validation. It does not use customer data, does not make credit eligibility decisions, and does not prove regulatory compliance.

v0.3.0 - pilot-ready causal explainability engine

v0.3.0

Initial public release of causal-credit-risk-engine.

This release provides a production-shaped, source-available reference implementation for causal explainability, counterfactual reasoning, model-risk review, and audit trace generation.

What is included

• Config-driven causal DAG for credit-risk explainability
• Versioned model and policy configuration
• Exact discrete inference
• Counterfactual decision traces
• Structured JSON audit records
• Deterministic replay from saved audit artifacts
• Tamper-evident audit hash-chain support
• Fairness diagnostics for subgroup review
• CSV batch decision mode
• FastAPI service surface
• OpenAPI export
• CLI demo workflow
• Evidence pack export
• Local enterprise seams for auth, tenancy, audit storage, model registry, and policy registry
• Governance documentation for pilot evaluation, procurement review, security posture, and integration boundaries

Validation

• 47 tests passing
• Default example risk output preserved: 0.849375
• Replay verification passes
• Audit chain verification passes
• End-to-end demo workflow passes

Important limitation

This is not a production credit scoring model, not a lending approval system, and not proof of EU AI Act compliance.

It is a reference implementation for causal explainability architecture and governance workflows.

License

Source-available under BUSL-1.1.

Commercial production use requires written permission.

Related technical brief

Zenodo DOI: https://doi.org/10.5281/zenodo.19779499