Targeted Change

Asha asked her assistant to "show cooking time on the cards". The cooking time appeared, along with a new font, a rewritten server file and a renamed field in the database. A small request with no fence around it invites a large answer.

The rule

Name the place, the change, and what must not change.

A targeted change alters one behaviour in one known place: a label, a bug, a rule, a layout detail. Most refinement is targeted change, and each one should produce a small diff that takes a minute to review.

A targeted request has three parts:

• Place — the file, component or screen: "the dinner card component".
• Change — the exact new behaviour: "under the title, show the minutes field as '25 min'".
• Fence — what must stay as it is: "do not change styles, other files or the server".

Then check the diff against the request. Are the files touched the ones you named? Is the size about what the change deserves? A one-line label change should not come back as a 200-line diff.

Bug fixes are the most common targeted change, and they need evidence as well as a place. "Fix the bug" gives the assistant nothing to aim at. "With an empty pantry, tapping Suggest shows a blank screen; show 'Add a few items first' instead, and change only the suggestion screen" gives it the trigger, the expected behaviour and the fence in one sentence.

The shortcut

Predict the diff before you ask. Guess which file changes and roughly how many lines. If the real diff is bigger, or somewhere else, stop and ask why before accepting it.

When the prediction is badly wrong, one of two things is true. Either the assistant wandered, so reject the change and ask again with a firmer fence. Or the change was never really targeted — it needs new data or touches several parts — and belongs to the next lesson.

Drills

Answer fast — trust the rule.

Remember

• Name the place, the change, and what must not change.
• Predict the diff before you ask; compare, then check the result in the app.
• Trap: a small request with no fence, accepted without a look at a large diff.