Measured at the wide boundary
Who pays the costs that never reach your P&L?
Every measurement implies a boundary. The boundary usually gets drawn where the numbers flatter. We draw it first, publish it, then measure. That includes the effects landing outside the entity and inside the ecosystem.
What the practice requires
- The boundary is drawn and published before data collection. It includes the labor, land, capital and care the program depends on.
- An externality register is maintained. Effects the program causes that its own P&L does not record.
- Second-order effects are tracked for at least one full cycle after handoff.
- Measurement happens in the field. A result available only on the test rig is a specification.
Systems literacy is the prerequisite
- Meadows gives the vocabulary. Stocks, flows, delays, feedback loops.
- Most ecosystem failure is a delay nobody modeled.
- Raworth gives the boundary. Decide what counts as inside before deciding whether you succeeded.
Where it goes wrong
- Optimizing the metric inside the app while the cost lands on a driver, a street, a household.
- Measuring under conditions engineered to produce the result.
- Declaring success at handoff. The interesting failures start at handoff plus ninety days.
Measures
- Boundary statement published before measurement. Yes or no.
- Externality register maintained and reviewed on a stated cadence.
- Second-order effects tracked for N cycles after handoff. N stated per program.
- Share of measurements taken in the field rather than on a rig.
- Deployment performance at handoff plus 90 and plus 365 days.
Three cases
These cases come from the public record: settlements, filings, testimony and the companies' own reporting. Citations are being attached case by case. Good, bad and ugly describe one decision against this one question, not the company as a whole.

How Ecosia planted a forest from ad clicks
A small search engine gives away all of its profit and reports more than 200 million trees planted. It accepted a boundary wider than its own margin, then published the counts against it.

Uber's god-view
Internal tools tracked celebrity riders' trips in real time. The company did not see a problem until a journalist did. Metrics improved inside the app while the cost landed outside it.

Volkswagen's clean diesel
Defeat-device software in 11 million vehicles, lowering emissions only during regulatory testing. The boundary was the laboratory. The harm was on the road.
How this one is marked
Green means we have the evidence and it carries a date. Yellow means someone owns the gap and there is a date to close it. Red stops the work until it clears. Early work sits at yellow for a long while, and we mark it that way.