Release Notes — Cash Recovery Engine

v1.0.0 — June 25, 2026

First release of the Cash Recovery Engine: an accounts-receivable prioritization decision engine that maximizes recovered cash per collector-hour.

Capabilities

• Uplift (T-learner) model — learns self-cure vs. pay-if-worked

propensities from resolved history and scores every open invoice by uplift.

• Capacity-constrained optimizer — greedy-by-ratio 0/1 knapsack that fills a

collector-hours budget with the highest expected-cash-per-hour invoices.

• Recovery segmentation — Persuadable / Self-Cure / At-Risk / Critical, each

with a recommended action.

• Deliverables — ranked worklist CSV, structured JSON, one-page executive

summary.

• Live web interface — capacity dial, KPI cards, recovery-curve chart vs.

conventional baselines, segment breakdown, and CSV export.

• Verification — 13/13 checks; auto-detecting real-data evaluation path.

Verified benchmark (held-out synthetic ledger)

• Self-cure AUC 0.817; calibration ECE 0.018 (Brier 0.159).
• Engine recovers +127% cash vs. the strongest simple baseline; 2.94× FIFO at

equal hours; captures 48.7% of the movable-cash ceiling within budget.

• Trains in ~0.1 s on 4,000 historical invoices; fully deterministic.

Status & disclosures

• Status PROOF_INCOMPLETE: verified on synthetic data; no real ledger present,

so no company-specific recovery figure is claimed. The term CERTIFIED is not used. See certification-report.md and proof/LIMITATIONS.md.

Known limitations

• Individual uplift is identifiable here only because the synthetic world exposes

both potential outcomes; on real data uplift is an estimate validated by a holdout/A-B test.

• Effort-hours and days-to-pay are modelled parameters.
• The engine is decision-support: it does not move money or contact customers.