Sense and decide
The 10 Standard Fire Orders + 18 Watch Out Situations
Danger signals that mean pause
A short list, taught in advance, of the conditions where AI output is most likely to be wrong.
Borrowed from the 10 Standard Firefighting Orders and 18 Watch Out Situations used by wildland firefighters in the United States. Read the original source.
Changes a rule or role · An hour or more to set up · Anyone on the team can start it
Try this first
Agree the short list of situations where AI output tends to be wrong for your team, and make sure everyone knows it.
Use it when
Teams can't recognize the moments AI is likely to fail.
Skip it when
The list has to be short and specific to your failure modes. A long or generic one gets ignored. These are warning signs, not a substitute for judgment.
How to introduce it
Wildland crews carry a fixed set of rules and a list of named danger signs that mean stop and check. Build your version for AI. Keep it short and memorable: unfamiliar domain, high stakes, confident claims nobody can verify. Teach it before anyone needs it, so people recognize a risky situation while they are standing in one.
How to show up
Build the list from the team's own experience, not a generic one. Teach it often enough that people recall it under pressure. Keep it short, because a long checklist will not be there when it is needed.
How long it takes
A session to build the list from experience. Ongoing teaching to keep it recalled.
What makes it hard
Writing a list that is short and useful takes real thought about where your errors happen. And a list only works once it is internalized, so teach it repeatedly rather than posting it on a wall.
What it looks like when it's working
People spot the high-risk conditions and check harder in exactly those, catching errors that would have gone through. A list nobody recalls in the moment is doing nothing.
How long until it sticks
Repeated exposure before the danger signs are internalized.
How you know it stuck
People recognize the danger signs and check harder without being told.
The idea behind it
Errors cluster in conditions you can name. You cannot stay vigilant about everything, and you can learn the handful of situations that deserve a second look.
Where it comes from
Wildland firefighting. A short "watch out" list of the conditions under which AI output is most likely wrong. The known danger signs, taught in advance. A fixed set of always-rules plus a list of named danger signals that should make you pause.
Also fits: Risky or hard-to-undo steps go ahead without a stop; Warning signs get explained away
Home problem: Confident claims go unchecked