lendwatch-th scoring specification
Rule-based triage scorer from lendwatch-th,
the detection pipeline accompanying Regulating Non-Bank Lenders in the Digital Age (Thanatchon Chongmankhong, 2026).
There are no trained weights. The scorer is a transparent rule score whose weights live in config.py;
scoring.py holds rule_score, evaluation metrics and fit_supervised, which re-estimates weights on reviewer-labelled data.
- Intended use: ordering lending apps and sites for human review. A score is a queue position, never a finding of illegality.
- Not calibrated: weights are a starting point; legal figures in
config.py(rate ceilings, platform rules) must be verified against current official texts. - Evaluation: synthetic only (30 synthetic markets, see the dataset
thanatchon36/non-bank-lending-research). Real-world accuracy is unknown.
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support