Sounds like an opportunity! One thing about these 'sadistic casino' domains is that small edges can have outsized impact. Even imperfect data that's swamped in noise can work. As long as the noise is consistent enough to be modeled, you can glean actionable insight.

Outdoor billboards are often priced based on raw traffic count. Imagine using a cheap license plate reader to sample traffic looking for enough identity data to map back to actual consumer behavior. Even if you can only do it for a few days and only a fraction of percent of your samples correlate to partial data, given high enough stakes and noise - just adding that as a correction overlay on your existing shitty model can yield a winning edge. In the land of illusions, any ground truth can be gold.