C-DAG | evidence dashboard

benchmark / evidence dashboard

WHAT VERIFIED.

One evidence view for what C-DAG actually ran, what verified, what failed, and what remains bounded.

validation coverage

COVERAGE

100k+ public financial rows processed
6 validation lanes
117 file corpus inspected
102 usable structured candidates
  • mortgage performance
  • mortgage applications
  • consumer complaints
  • CRT
  • CAS
  • public loss-exposure records

validation lanes

WHAT RAN

Lane Rows Result Replay Audit-chain
Freddie + Fannie + HMDA 30,000 APPROVE 9,584 / REVIEW 4,368 / DECLINE 16,048 100% replay success on sampled validation audit records verified

verification

WHAT VERIFIED

Replay verification

Replay verification: 100% success on sampled validation audit records across reported validation runs.

Audit-chain integrity

Audit-chain integrity: verified across reported validation runs; tamper behavior covered by tests.

Holdout baseline

Rows: train 58,579 / test 41,421

Test positives: 201

Metrics: AUC 0.573062 / PR-AUC 0.006059

Distribution: APPROVE 36,336 / REVIEW 5,085 / DECLINE 0

Holdout baseline

Holdout baseline: real public outcome validation. Signal remains limited and should be treated as baseline governance evidence, not production model performance.

Surrogate governance

Rows: train 165,297 / test 70,842

Metrics: AUC 0.646938 / PR-AUC 0.005352 / Brier 0.002646

Surrogate: high fidelity; counterfactual consistency unstable

Boundary: validated surrogate governance evidence, not automatic causal truth.

estimated exposure mapping

LOSS EXPOSURE MAPPING

Parsed public records map failure types to replayable evidence and audit-ready artifacts. This does not claim prevention or savings.

Records parsed: 5. Sources present: CFPB, FINRA, SEC, OCC, AI operational-loss research. Not yet parsed: FFIEC.

CFPB / Wells Fargo $3.7B order

Exposure: $3.7B

Failure type: Consumer-harm and servicing-control breakdown.

Missing artifact: Cross-workflow, replayable decision trace.

C-DAG fit: trace, counterfactual, replay, hash-chain, evidence pack, risk-exposure mapping