Fix one part and the symptom moves

1.2aThe effect lives in the connectionssrc 02

What a product does to people comes from how its parts connect. That is the starting claim of systems thinking · tư duy hệ thống. Three features of the connections make a fix go wrong:

• Loops amplify. When an effect comes back to feed its cause, it is a feedback loop · vòng lặp phản hồi. It grows whatever runs through it — the benefit and the harm.
• Delays hide the cause. An effect that arrives years after a decision is not traced back to it, so nobody corrects it in time.
• Incentives keep it running. A part that pays someone — a metric, an ad model — resists any fix that slows the loop.

The harm needs no bad part: !each part can be reasonable on its own.Surprising: I expected a harmful outcome to trace back to one bad decision. Each part can pass its own review, so checking parts one at a time finds nothing.

1.2bWho benefits now, who pays latersrc 01

A loop splits benefit from cost. The benefit is immediate and lands on the dashboard; the cost is delayed and lands on someone else, or on the same person years later. Don Norman's humanity-centered design widens the frame for this reason: from the user in front of you to the people and the years around the product.

That split is why a loop can look healthy for a long time. ?Nothing the team measuresConfused: If the cost lands outside every dashboard, who on a product team is supposed to notice it? sits where the cost lands.

1.2cFour moves before you fix a partsrc 02

1. Draw the parts around the thing you are changing. Who uses it, who pays for it, what measures it. A loop stays invisible until its parts share one page.
2. Find the loop. Follow the arrows ★that come back to where they started.Actionable: Use it on the metric my team celebrates this quarter, at the next planning review. That loop will amplify whatever you put in it, so it is where a change matters most.
3. Mark the delays. Write beside each effect whether it arrives in weeks or years. The long ones are the ones nobody will blame on you — or notice until it is late.
4. Ask who pays. Name who benefits now and who pays later. When they differ, no one inside the loop has a reason to stop it.

1.2dAt work: four situations

• A fix shipped; the complaint came back somewhere else. The loop routed the symptom to a new place. I draw the parts around both complaints and look for the arrow they share.
• A metric goes up and something human goes down. →The dashboard sees one side of the loop.Connection: Same shape as 1.2b: the split between who gains and who pays. I add one measure from outside it, and note when it would arrive.
• A feature that rewards coming back. Streaks, feeds, notifications — each return feeds the next. I ask what the loop amplifies after a year, and for whom.
• "Add a take-a-break reminder." A fix inside the product, while the model that drives the loop stays. I name the part that pays for the loop and ask whether the fix touches it.

1.2eInfinite scroll, 2006src 03

In 2006 Aza Raskin designed infinite scroll to spare people a click. Business models built on time online turned it into a loop. Studies the BBC cites link social media overuse to depression and loneliness, and Raskin said he felt guilty.

1.2fAntibiotics for a coldsrc 04

A doctor prescribes antibiotics for a cold that would pass anyway. Repeated across a population, such misuse drives resistance, by WHO's account. No single prescription caused it.