Each page says what the work was trying to answer, what it found and what was wrong with it. The ones I stopped are here too, with the reasons.
Sole author · March to May 2026
Three months on a study of information asymmetry in Indian derivatives markets. I audited all 46 references against the primary sources after finding I could not defend the paper. It is paused until I understand its statistics well enough to defend those too.
Team leader · 2026
Our team's entry to the International Mathematical Modelling Challenge, which took an Honourable Mention at the international round. Etosha faces six threats at once across 22,935 square kilometres with 295 staff. We built a protection metric, a deployment plan and a feasibility score, then simulated five years a thousand times.
Concept, then verification · February to May 2026
A crypto trading bot built around my own idea about how institutions move price into the places where retail traders keep their stops. The research, the backtesting and the code were done with AI. Working out whether any of it was true is where I actually learned something.
Concept and architecture · March 2026
A Telegram bot that read my Obsidian notes, wrote questions from them and asked me those questions on a spaced repetition schedule. It worked. I stopped using it for reasons that had nothing to do with the code.
Independent, with guidance from Dr. Balbir · 2024
I got a solution marked wrong for dividing by zero, got angry about it, and spent two weeks throwing everything I could think of at the problem. I got nowhere. It is the least rigorous thing on this site and I have kept it for what it taught me.
Design and rules · Notes since 2023, rebuilt in 2026
For years this was just notes in Obsidian. In 2026 I rebuilt it around Andrej Karpathy's LLM Wiki pattern: 349 notes in two layers, where I write the rules and an agent does the ingesting and the linking.