TidalMoon
A market forecaster built with strict rules about what it may know, and when.
- Status
- Research prototype
- Year
- 2026
- Role
- Solo
- Area
- Markets
In plain English
Many market models look brilliant in testing because they accidentally peek at the future. TidalMoon is built to rule that out, so its test results reflect only what it could have known at the time.
How it works
- Store data point-in-time, so each prediction sees only information that existed at that moment.
- Write a frozen evaluation contract (fixed pass-or-fail rules) before any training starts.
- Train deep-learning forecasting models in PyTorch on that data.
- Promote a model only if it beats the contract.
What I did
I built it solo as a research prototype.
What I learned
Getting the data rails right took more work than the model, and it is what makes the results trustworthy.