Pre [2026-03-16]: Started with roadmap, lasso, and idea finished. Settled on lasso with volume and liquidity filters to remove high multicollinearity. Granger causality test to determine lead-lag relationships. Perplexity Claude Sonnet 4.6 used for prompt building, api configuration, and optimizing roadmap and plan. GPT-4 API used for phase two to pre-filter market questions. Below; zip of the regression engine; PDF explainer of running the engine and expected results. [Currently Untested]

Post [2026-03-16]: Not a lot of debugging. Very impressed with perplexity computer, I mostly had to change the inner bounds for most of the polymarket mapping (how many markets to pull, how many to test at a time). Figuring out the run was slow as well considering I am not a native python user. Final zip is below with another explainer on how to run. Tweaks as I use it but as of right now the outputs are what I am expecting with maybe a disapointing quantity but I can always relax the AI to allow more.

[2026-03-18]: While the program itself is very strong, the causality regression needs fixing (current outputs are skewed with high multicollinearity). Will keep the polymarket API configuration from phase 1 and 2 with tweaks to the filtering and caching system. Phase 3 can stay as is while 4-5 need complete changing. The Granger causality is eliminating most markets and then the Lasso is filtering the rest out till the only significant markets are typically those with all bets on either poles (1% and 99%). The hardest part was the API configuration so having all those setup means this project is essentially a test on my econometric skills.

[2026-03-26]: With the announcement of CIRO approval for Wealthsimples binary trading platform I am going to shift to mapping the eligibility of current markets for use in Canada. The causality testing I will leave untouched but instead pull the working Gamma and CLOB API phases for us in a pre-screening filtering engine. I figure this might be more beneficial for short term, then with more time pending, I can work on fixing the causality testing to be more robust.