What happens when a measure becomes a target, and AI can carry the target out?
A door in (a seed): HR sets a target for team fun. Every outing is attended and the survey says 4.8 out of 5. The only real laugh that year came when the lift got stuck with six people in it.
Goodhart’s law, in its common form, says that a measure which becomes a target stops measuring well. This chapter takes it one step further: a metric with consequences is a policy, and AI makes targets executable.
An audit is not neutral either. It reports against a chart of accounts, and someone chose the accounts. So the chapter asks whose chart this is, and puts two habits to the test: measure to learn, not to rank; and give every target an expiry date.
“Signal” in the title marks a Western idea brought across as a bridge, named with respect.
A line it may earn (a seed): Whatever has no line on the form goes unnoticed: the trust that gets spent, the work that stopped a disaster.
Sections planned so far
- A metric is a policy
- Whose chart of accounts?
- Measure to learn, not to rank
- Every target an expiry date
The image it turns on: a team optimising a weak signal until the signal is all that’s left (after Simon Høiberg).
This chapter might instead be called Every Metric Eventually Applies for Management.
A seed, sketched with AI from the author’s notes and the studio’s reading: the question and the direction, not yet the argument.
Leans on AI Is the Audit and TMC. Leads to Sutra of Pañca Ṛṇa: the price forgets, so what remembers?