finassure-fragility-index

FinAssure Fragility Index

The FinAssure Fragility Index (FFI) is a regularized-logistic meta-model selected by grouped internal Brier-score comparison among regularized logistic regression, random forest, and XGBoost. It estimates a pre-registered benchmark assurance-policy breach probability from permitted aggregate audit telemetry and declared scenario context. It is not a novel random-forest method and it is not an individual credit-risk model.

On the 360-run frozen grouped holdout, FFI achieved AUROC 0.8878 (95% CI 0.8534–0.9197), PR-AUC 0.8110 (0.7574–0.8630), Brier 0.1327 (0.1157–0.1497), and ECE 0.0596 (0.0548–0.1027). FFI is identical by construction to the selected unweighted full-telemetry logistic comparator. Its AUROC differences against class-balanced logistic, uncalibrated random forest, and XGBoost are not uniformly conclusive; the artifact should therefore be read as evidence for the representation and protocol, not algorithmic supremacy.

The output is a policy- and scenario-conditioned research prioritization signal. It does not evaluate individuals, approve deployment, certify compliance, determine legal fairness, or replace independent validation and accountable human review.

See the model card, ESWA results ledger, anonymous manuscript, audit-telemetry Dataset, and live Space.