Practise reliable habits: spreadsheet formula

• Task: What exact job is the system allowed to do?
• Assessment: What representative evidence would show acceptable performance?
• Steward: Who monitors and corrects it?
• Kill switch: What observation pauses use?

For a chatbot answering opening-hours questions, Mira records:

• Task: answer only from the published hours page.
• Assessment: 40 questions covering branches, holidays, ambiguity, and missing information; all dates and times must match the page.
• Steward: digital services lead checks weekly samples and corrections.
• Kill switch: any invented closure or repeated failure to say “I cannot find that.”

This does not guarantee perfection. It makes the claim testable and gives errors a route to action.

Apply TASK to a spreadsheet formula that totals room-booking fees. The assessment can use hand-calculated cases. Notice that the same discipline helps whether or not the tool is labelled AI.

Create a TASK card for one system in your life. Use a threshold you could actually observe; “works well” is not a threshold. Ask another person to identify one missing edge case.