Half the Testing, by Design

The Document Almost Nobody Reads

A federal document cut fire alarm testing in half, and most of the people working under it have never opened it.

Under NFPA 72, all of a building's initiating devices get tested once a year, every one of them. Nobody picked that number at a conference table. Like most fire code, it came out of old fires, out of buildings where the alarms stayed silent. The military plays by a different book. UFC 3-601-02, dated 7 Oct 21 with Change 1 on 18 Apr 25, tells its users to run its task frequencies "in lieu of the tasks and frequencies in the National Fire Codes." Table 2-1 drops those same devices to a two-year test cycle. Smoke detectors included.

Where the Numbers Came From

Nobody guessed at that either. The frequencies trace back to a reliability study the Air Force ran in the nineties, AFCESA Technical Report 01-10. The study put the odds of a fire at one event in fifty years, and it put the odds a given test gets done right at ninety-nine in a hundred. Push those two numbers through the arithmetic, and a two-year interval still pencils out to 99 percent reliability.

On paper it reads clean. Half the labor for the same protection. I spent twenty years testing this gear in the field, and here is the part that might catch you off guard. The model holds up. The engineers who built it did honest work with the assumptions they picked.

The trouble is in the assumptions.

What the Field Says

I have stood on bases where six in ten fire alarm systems were dead or half dead. The maintenance gaps piled onto reporting gaps, season after season, until the deficiency list became the only real inventory anyone had. The model promised 99 percent, and the field handed back numbers nowhere near it.

Look at what that 99 percent really asks of one tech on one ladder. He has to dose a smoke head with an approved aerosol, then check its drift against a sensitivity limit the code follows across five years. That is thinking work you do with your hands. The record of it is a second job stacked on the first, and it only counts if the tech who comes after can actually use what got written down.

Cutting the test schedule in half raised what it asked of the technician doing the testing, and no budget line ever paid for that.

The criteria take for granted a tech who runs in two modes at once, the old hands-on one and the new knowledge one. They bank on his records the same way they bank on his testing. The trade, meanwhile, kept turning out hands-only workers while the model quietly rewrote the work under them. Six in ten dead is what that gap looks like twenty years on.

The Part That Runs Toward Us

The people who wrote that document aren't the villains here. They ran a model, and every model is a wager. This one got wagered on the trades, spelled out in federal criteria, on the record.

The crews inside those buildings called it years ago. They kept saying the cycles fall apart the minute staffing and turnover touch them, and being right paid them nothing. Field data was the one input the model never gathered. The real service life of this equipment lives with the people who never stopped putting hands on it, in their heads and in their truck notes. The bill for all that deferral is landing now. Someone has to scope the recovery, and only one labor pool can do it.

So the fix is the boring kind. Bring techs up to the knowledge-worker standard the criteria already score them against, and treat the report as something the building owns, because the next tech inherits whatever shape you leave the panel in. The whole optimization rests on one person on a ladder, and that person needs to be trained and paid like it.

Next: the full review of Capitalism in America, the book behind this piece.