Data products you can open and inspect.

ResultConnects documents, people, events, and citations in one inspectable investigation workflow.

ResultTurns public legal documents into a searchable corpus with evidence-linked answers and an explicit refusal path.

ResultTurns generated SQL into a reviewable proposal and prevents direct model-to-database execution.

ResultA credit-policy analyst can test where to draw a first-EMI risk line, see the confidence interval around the answer and who it declines, and take a recommendation or a refusal to governance. On the published assumptions the honest output is a refusal: no evaluated band clears zero.

ResultThe selected model clears prevalence/random, repayment-delay, and logistic references on repeated paired development evidence, but ties calibrated Extra Trees because the PR-AUC advantage does not clear the prespecified practical margin. At 10% holdout capacity, 600 historical rows contain 431 observed defaults with 71.8% precision and 3.25× lift, each reported with uncertainty.

ResultA fraud strategy analyst can compare screening approaches at a fixed review capacity, see what each buys and costs, and take a recommendation or a refusal to governance. On the pre-agreed checks the honest output is a refusal: the proposed model catches 472 more fraud attempts while holding up 472 fewer good customers at identical cost, and is still not promoted, because its calibration and population stability fail checks written before the result was known.

ResultOn the fixed synthetic holdout, the calibrated Extra Trees challenger improves AUROC from 0.5000 to 0.9796 and Brier score from 0.0475 to 0.0228. The dashboard turns those model results into an inspectable threshold tradeoff without presenting them as clinical evidence.

ResultDelivered a source-traceable 7, 14, and 28-day planning surface for 104 sub-ICBs, backed by 32.9 million validated source rows, rolling-origin evaluation, immutable releases, and a live edge API.

ResultSupport leaders can frame one staffing-meeting question from two independent historical baselines, inspect the latest observed demand window, and make missing live inputs visible before discussing coverage.

ResultDelivered bounded public investigation across 20,439,666 masked loan-period rows, exact portfolio rate-and-mix attribution, private object storage, authenticated data retrieval, and a zero-incremental-cost production release.

ResultRetain the ordinary logistic baseline for the measured 1% review queue. It reaches 0.160 PR-AUC and 51.3% recall while the class-weighted challenger performs worse on ranking, recall, and probability error.

ResultA subscription lifecycle analyst can locate material renewal and engagement changes, compare cohorts and segments on governed definitions, and test a finite contact program without confusing historical risk with causal treatment response.

ResultA fraud operations lead can compare threshold and staffing choices, see captured and missed fraud, inspect anonymized cases, and identify where extra review capacity stops paying back.

ResultDelivered a public trust-to-investigation workflow over 693,640,933 physical disclosure records, 264,922,553 loan-period facts, 9,240,038 security-period facts, and all 167 approved source and derived release artifacts.