Think twice / 01
The Rulemaker
You make the rules. They earn the points. Can you get this neighbourhood’s deliveries done?
How to play
You run a neighbourhood delivery service with three couriers: Mira, Dev and Noor. You want deliveries completed quickly, but you don't decide which jobs they take. You decide what earns them points. They choose whichever available job offers the best reward for their time.
- Leave Per delivery selected and press Run shift. Follow a 60-minute afternoon, or use Skip to results to complete it immediately.
- Pause to inspect a courier's choice, or open Follow every request. Look for a request that waits while other jobs are completed.
- Run the same afternoon with Per block, Waiting matters or your own mix. Complete at least two shifts. In What changed?, find a rule that improves something you care about, then check its effect on outstanding requests and the longest wait.
What to notice
A call centre can shorten every call and still leave customers' problems unsolved. The numbers improve, but the service doesn't. Economists have long studied this problem: people can act rationally within a system and still produce outcomes nobody intended.
In the 1970s, economist Charles Goodhart observed how economic indicators could lose their usefulness when policymakers began targeting them. This became known as Goodhart's Law, later memorably expressed by Marilyn Strathern:
When a measure becomes a target, it ceases to be a good measure.
The principle extends far beyond economics. Businesses design KPIs, institutions measure performance, and AI systems optimise reward functions. But does improving the measurement actually improve what it represents?
Your couriers demonstrate the dilemma. Reward distance and longer trips become attractive. Reward waiting time and neglected customers receive attention. Even average waiting time can be misleading when unfinished deliveries aren't counted.
Nobody is cheating; they're following the incentives you designed. Did you improve the service, or just change what your workers became good at?
Read more: Charles Goodhart, Problems of Monetary Management: The UK Experience.
The afternoon shift
Minute 0 / 60Numbered rings = outstanding deliveries. M, D, N = couriers. One street block takes one simulated minute.
Run the default Per delivery shift first. Then change the reward and compare the same afternoon.
Select a courier to follow their decisions.
What changed?
| Reward rule | Delivered | Outstanding | Avg. completed wait | Longest wait |
|---|
Follow every request
| Request | Destination | Arrived | Courier | Status | Wait |
|---|
Rules & model
The job you give the couriers
Get the neighbourhood’s requests delivered during a 60-minute shift. Every replay has the same 38 requests, arrival times and starting positions. You choose what earns points: a completed delivery, distance from the depot, time already spent waiting, or a mix.
How a courier chooses
A free courier takes the available request with the most points per minute needed to finish it. That time includes returning to the depot, one minute collecting, and travel to the destination. Each street block takes one simulated minute. A chosen job is finished before another is considered.
Mira chooses first, then Dev, then Noor. Equal rates go to the lower request number. Waiting points use the wait so far when a job is accepted; couriers do not predict new arrivals. Routes go horizontally, then vertically.
Reading the result
Average completed wait counts arrival-to-delivery time for finished requests only. Longest wait includes unfinished requests, measured up to the current minute. At minute 60, unfinished trips stay outstanding. Points are your incentive, not a performance score comparable across different rewards.
These are invented decision rules. The couriers do not learn, coordinate, get tired or have opinions about fairness. The experiment makes the effect of one incentive inspectable; it cannot tell you how every real delivery service would behave.