ExampleA recorded run, kept as an example of how the agent works.
I can build that from the verified panel plus the follower graphs of the two big delivery apps, then narrow it to people who actually order.
build_population
Florida delivery orderers 4,820 twins
18–44 · Florida metros · 2+ orders / month
4,820 twins after deduplication. Want to ask them something?
Here is the run I would set up.
define_simulation
run_simulation
1,488 of 2,400 twins resolved
read_results
Tipping holds, but it is defensive: most who tip say they are protecting the next order, not rewarding this one.
High confidence
inspect_twins
twin_07724 · 19, Gainesville, student, orders on promotions only. · No tip
I only order when there is a deal. Adding a tip cancels out the deal, so I get it delivered to the lobby and skip it. I would tip cash if I had any.
twin_09931 · 27, Miami, orders weekly, fee-sensitive. · No tip
There is already a service fee, a delivery fee and a small-order fee. I assume some of that reaches the driver. If it does not, that is the app's problem to fix, not mine.
compare_runs
Nine points move to no tip, and almost all of them are the fee-sensitive group rather than the students.
check_validation
91%
alignment
1,240
held-out humans
Measured against held-out humans from the same panel, on questions this population has not seen. High for direction and ranking; treat the exact percentages as approximate.
Everything here is an example of usage. Sign in to ask a population your own question.
Sign in