Downloads · 30 days
0
Mohamed-Yossri/egyptian-informal-institutions-eval
egyptian-informal-institutions-eval is a machine learning model from Mohamed-Yossri. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Living in Alexandria, Egypt, the formal banking and legal systems do not reflect how most people actually conduct business or handle financial emergencies. Over half the economy is informal. Families and small busines…
Downloads · 30 days
0
Access
Public
Updated Sep 17, 2026
Repo size
—
Likes
0
Public
Click a slice to open those files.
.json5.6 KB · 48%
From the Hugging Face model README
Living in Alexandria, Egypt, the formal banking and legal systems do not reflect how most people actually conduct business or handle financial emergencies. Over half the economy is informal. Families and small businesses rely on two foundational mechanisms:
Standard alignment datasets (RLHF/DPO) and benchmarks (like LegalBench or MMLU) view finance and contracts through a Western common-law or formal civil lens. When someone breaches a contract in that world, it is a civil dispute resulting in a credit penalty or civil litigation.
In Egypt, a Wasl Amana is not a civil debt acknowledgment; under Article 341 of the Egyptian Penal Code, it is treated as criminal breach of trust (تبديد / خيانة أمانة). Defaulting or misusing it does not hurt your credit score—it carries mandatory imprisonment of up to 3 years. Furthermore, signing a blank receipt (Wasl Amana ala al-bayad) is a catastrophic legal trap that informal lenders routinely exploit.
Frontier models, especially compact ones (0.6B–6B) widely deployed on edge and local setups, repeatedly misclassify these informal instruments as standard commercial loans or benign promissory notes. In doing so, they give disastrous advice: advising users to "negotiate civil settlements" or treat the receipt as a dispute over loan terms, ignoring the imminent threat of criminal arrest.
Qwen/Qwen2.5-3B-Instruct (Open-weight, 3.09B parameters, running in float16 on free Google Colab T4 GPU).Closing this capability gap requires moving past superficial translation fine-tuning and addressing legal/economic ontology: