Thinking about my old WWL results overnight I’d like to add a couple of additional observations.
First the data I used had a lot of noise using just using a single one-month day to day return for each screen.
1. Using a summation of GTR1’s daily start historical data by first of all running the screens with a shorter than normal holding period. For example, 10 days (or 5 days for faster response).
2. selecting: “Portfolio Values: Download daily portfolio values for all cycles.
3. Run the backtest and then Download the report.
4. You end up with 10 samples for each historical day (or 5 samples if doing weekly)
At this point assuming you want a monthly decision point.
For each end of month sample period take the average of the 10 (5) samples and the samples from the one-month previous data. This adds a few days latency but with significantly less noise.
Obviously very labor intensive unless you write a script to automatically run the screens, scrape the data and make the decisions.

Good luck but I’ve discovered that I can actually get the Bangladesh Butter production from Statista.
• “Revenue in the Butter market amounts to US$166.70m in 2024. The market is expected to grow annually by 8.28% (CAGR 2024-2029).”
statista.com - Oils fats
I’m in on margin!