Person/Atul Gawande/Book/09/The Checklist Manifesto
About
- Subtitle: How to Get Things Right. Published 2009 by Metropolitan Books. Argues that as knowledge and complexity outpace individual expertise, failures increasingly come not from ignorance but from ineptitude — not applying what we already know. The remedy is the humble checklist: a simple tool that catches the predictable, stupid mistakes and frees experts to handle the genuinely hard problems.
- Draws examples from surgery, aviation, construction, and finance, culminating in Gawande's work on the WHO Surgical Safety Checklist.
What is a check?
- Gawande never pins down a single check as a unit — his focus is the checklist as a design artifact. Read closely, an implicit definition emerges: a check is a bounded verification that a specific, high-consequence condition holds before a process advances.
- It verifies rather than teaches — a checklist is not a comprehensive how-to guide; it confirms that crucial things have happened or are true.
- It covers only critical items — the "killer items" that are both essential and easily missed, not everything a person knows how to do.
- It happens at a pause point — the workflow deliberately stops before proceeding (before incision in surgery, before takeoff in aviation).
- It has an observable, binary outcome — a good check answers a yes/no question: Has the patient been identified? Has antibiotic prophylaxis been given? Has everyone introduced themselves? Is the fuel valve open?
Ideas
- Distinguishes DO-CONFIRM checklists (do the work from memory, then pause to confirm nothing was missed) from READ-DO checklists (read each item and immediately perform it, like a recipe).
- Good checklists are short, precise, and practical — not exhaustive. A checklist that tries to spell out everything becomes unusable.
- Checklists also install communication points — forcing a team to talk to each other (e.g. everyone stating their name and role before surgery) is itself a safeguard.
- Frames modern failure as ineptitude (not applying what we already know) rather than ignorance (not knowing what to do); checks are designed to close the ineptitude gap.
- A Gawande check is much narrower than an AI/Eval/LLM as Judge criterion. His checks are intentionally binary, observable, and actionable — they answer "Can we safely proceed?" not "How good is this overall?"
- Useful test for what deserves to be called a check in a CLI jig or eval harness: is it a decision gate with a yes/no answer, or a graded quality judgment? The former is a check in Gawande's sense; the latter is an evaluation.
Links