It’s like how a normal person who’s worked with a couple of solo independent contractors (say, for some work on their house) and has a sense of how they both price things from past bids, might offer a job they need done by a certain time to both of them—exact same job—at different rates, all else being equal (the homeowner rates their work as comparable), just because they know one of them historically bids lower. They’d prefer the cheaper one, but are willing to pay either rate, and need the job done by time X, so offer it to both, first to accept gets it.

Now throw in a bunch more factors than just prior bid history, thousands of workers in the dataset instead of two, and such high volume and pace of work that you can afford to periodically experiment by setting, say, a random 10% of each set of offer-receivers lower than your formula would usually suggest, to see if anything’s changed and maybe they’re more-desperate for some reason (you don’t even need to know why… though, imagine if you could spot reasons some workers might be more desperate! Hm…)

Now (maybe) extrapolate to similar, slower-paced efforts in less-marginal areas of work. Interesting (yikes) possibilities.