If they can snare people into driving for them, there are material switching costs for workers, and 'underpaying' on some level is actually something they can get away with.
At the 'low end of labour' workers generally have zero power, much less during covid, so it's not a 'marketplace' in the more broad sense. If it was, we wouldn't even need minimum wage laws, or Medicare for employed people etc..
If the technical challenge is 'insurmountable' then they need to figure something out: minimum guarnatees etc. , even over the aggregate, as you say.
This could also be an "unintentional-intentional" bug, meaning its presence was just an accident -- but that the company has chosen to prioritize its engineering resources addressing issues that improve its revenue, rather than decreasing it.
There's way more money to be made in having accurate measurements than having inaccurate measurements that optimization engineers can't rely on. The more accurate the measurements, the more confidence you have that your marketplace optimization algorithms are producing the desired marketplace dynamics.
2. I could see a company with various ethical lapses like Uber not prioritizing an unintentional bug for a speedy fix.
3. In the intentional scenario, it'd be a case of https://en.wikipedia.org/wiki/Salami_slicing (popularized by Superman 3 / Office Space). Easier to get away with than a public rate cut, with the downside of bad PR if you get caught.
The interview seems to indicate it's an aggregate underpayment across all trips. The tool is available for examination if you are truly skeptical.
edit: also, you seem to have made your account just to comment on this thread. what's up?
For example, this makes no sense:
"I had this one delivery that was an hour-and-a-half long and I got paid $16 and I thought, 'There is no way that's right,'" Armin Samii, an unemployed computer scientist who has been working Uber Eats on the side, told Salon. "I looked into it and found out that Uber paid me for a one mile delivery instead of a four mile delivery. Of course it's all made worse because I'm on a bike and they don't account for that, but that's a separate issue."
A 11 mile hill climb of Mount Hamilton with 2700 of elevation gain takes me about 1.5 to 2 hours to do. A straight hill climb which you would only find in cities like Hong Kong or Rio de Janeiro is like the worst case scenario. A 4 mile trip in pretty much any major city should be maybe 20-25 minutes with stop signs and traffic lights. Not saying there wasn't an issue with that trip, but this makes no sense.
I have worked a lot with this stuff in my career, so I'm inclined to want to see the data myself so I can try and reverse engineer what may have happened.
Furthermore Armin Samii appears to have worked for a direct competitor of the company he's criticizing while developing this.
Under or overpaying on trips is more of an interest to those engineers trying to optimize marketplace efficiency. Meal delivery businesses have as much of an interest in not underpaying as doing so reduces supply availability and therefore reliability and they have also an interest in not overpaying as that reduces profitability. The ultimate goal is optimal pricing in a three sided market to maximize liquidity as the business model does best when you optimize for volume of transactions. Maximizing liquidity amortizes fixed costs over more transactions. This applies as much to the meal delivery companies as to the restaurants involved.
At the end of the day, distance driven is very imperfect as you have issues like poor GPS signal and in cities you have the dilemma of urban canyons and tunnels. Also, drivers don't always take the suggested route. At vehicle speeds, it's hard problem to solve due to time between GPS pings. Another issue besides GPS inaccuracy is clock inaccuracy, especially with round-trip times between server and client and client and server aren't always symmetrical, especially on cellular networks, and this messes up all sorts of clock synchronization strategies. The peanut gallery on HN is seriously underestimating the complexity of measuring distance using GPS. It's pretty darn good when conditions are ideal, but when they aren't, good luck.
Another question is the magnitude of the underpayment in the 21% of trips. Are we talking about 1-2%, 5-10%, some other deviation? If it was something like 1-2%, we should all be amazed. Gage repeatability and reproducibility matters.