Tengacion Technologies Limited

# TengaLab — Don't Trust the Patch. Prove It. TengaLab is an experimental AI-powered software reliability laboratory developed by Tengacion Technologies Limited for the Meta Global AI Developer Hackathon 2026. TengaLab addresses a critical challenge in AI-assisted software engineering: how can developers verify that an AI-generated repair genuinely fixes a software defect? Using Meta's Muse Spark 1.3 model, TengaLab investigates software failures, generates candidate source-code repairs, and evaluates them through an independent experimental verification system. Our demonstration reproduces a software reliability defect involving event lag, concurrency, and shutdown handling. The original implementation passes only 3 of 6 behavioral tests and is rejected. The AI-generated repair passes all 6 tests and is accepted within the controlled fixture. TengaLab also rejects an intentionally ineffective modification, demonstrating that its verification system evaluates observed software behavior rather than relying on AI-generated confidence. The project uses restricted rootless Docker execution, adversarial testing, structured evidence, and SHA-256 integrity checks. Its broader automated regression suite contains 69 passing tests. The public Proof Console presents recorded verification evidence, while the GitHub repository provides the source code and instructions for reproducing the experiments locally. Our vision is to develop a new reliability infrastructure for AI-generated software, helping engineering teams progress from plausible fixes to experimentally verified repairs. Don't trust the patch. Prove it. Developed by Tengacion Technologies Limited. GitHub: https://github.com/bakomamman-tech/tengalab Live demonstration: https://tengalab-proof-console.onrender.com/