I Found Two Bugs in a Hackathon's Judging Tool. Neither Explained Why I Lost.

A developer recently shared their experience participating in the Africa Deep Tech Challenge 2026, where their project, StacksNG, failed to advance to the semifinals. The developer initially investigated technical issues, discovering two significant bugs in the hackathon’s evaluation tooling: a misconfigured base URL in the profiler and a compatibility mismatch between the evaluation harness and the local model server. After patching these, the developer calculated a competitive score, yet they were still rejected. The final feedback revealed that the project was disqualified during an 'originality gate' because it utilized a stock model without fine-tuning. The author reflects on the importance of understanding hidden judging criteria, noting that while their RAG-based architecture was technically sound, it failed to meet the implicit requirement of demonstrating model modification. The post serves as a cautionary tale for hackathon participants to look beyond published scoring formulas.