We Made a Rule After One Bad Job

Why this matters

One bad job produces a memo by Monday. The memo applies to every job, every tech, forever, and nobody ever goes back to ask what it cost or what it caught. This case is a seven-tech plumbing shop that did exactly that, and the only reason they found out the rule was a dead loss is that the job checklist happened to log it. The review that undid it is the part worth copying, because most shops carry two or three rules like this and have no record that would let them see it.

The rule, as it was written

The incident: a Friday afternoon faucet replacement. The tech swapped the faucet, connected new supply lines to the existing angle stops (the small shutoff valves under the sink), turned the water on, looked, saw nothing, and left at 4:40 pm. On Monday the customer found the sink base swollen, the laminate floor lifted along the cabinet toe, and water in the ceiling of the room below. The claim and the rework cost the shop roughly three weeks of that tech's billed margin, and the customer left a one-star review describing the shop as careless.

The owner wrote this the following Tuesday:

MEMO - effective immediately - ALL JOBS
Before leaving any job, every tech:
 1. Runs water / pressurizes system 15 min
 2. Puts a dry paper towel under every joint
    and checks it at the end of 15 min
 3. Photographs each joint, uploads to job
 4. Ticks "LEAK HOLD DONE" on the checklist
No exceptions. Not one more floor.

Read it as a record now, not as a reaction. It names no job types, so it covers drain cleanings, water heater flushes, estimates and toilet augers along with fixture swaps. It names no failure mechanism, so nobody can tell later whether it addressed the one that happened. And it carries no end date or review date, which is how a rule written in an afternoon outlives the person who wrote it.

What the job log said nine months later

In December the office manager was cleaning up the checklist template and asked a plain question: how often does the towel come back wet? The log could answer that, partly.

• Jobs under the rule, April to December: 2,340.
• Minutes added: the checklist had a single yes/no box until July, when the shop added start and stop timestamps to the hold step. From July to December, 1,410 jobs carried timestamps, with a median of 14 minutes from start of hold to tick. The 930 jobs from April to June had only the box, so their minutes are an estimate carried over from the later median, and the review said so.
• Total time: 2,340 jobs at 14 minutes is 32,760 minutes, about 546 tech-hours. At the shop's average of roughly 2.1 hours per job including drive, that is the time of about 260 jobs.
• Wet towels logged: 6.

Six findings out of 2,340 jobs looks like a rule doing a small amount of good. The review did not stop there, and this is the first move worth stealing: a finding is not a catch until you know whether the old practice would have made it anyway.

Reading the six

Each of the six had a tech note attached. The office manager and the lead tech read them one at a time.

• Four were real leaks on joints made under supply pressure (two supply-line connections, one valve swap, one water heater union). In all four the note said, in some form, "drip on first pressurization" or "wet within a minute." The shop's practice before the memo was to pressurize, wipe dry, and look again after a couple of minutes. Every one of those four would have been caught by the two-minute look the techs were already doing.
• Two were not leaks from the work at all. One was condensation on a cold supply line in a humid crawlspace on a drain-cleaning job. One was a pre-existing drip from a corroded tub valve two rooms away from the toilet the tech had augered.

So the count that matters is zero. In 546 tech-hours, the 15-minute hold caught nothing that the old two-minute check did not already catch, and it produced two false alarms, each of which cost an awkward conversation with a customer about water that the shop had not caused.

Reconstructing the March job

The second move is to go back to the incident the rule was written for and ask whether the rule would have prevented it. That needs the original record, and the March job record was thin: the checklist was blank (the shop did not require it on small fixture jobs then), the tech's note said "faucet replaced, tested ok," and the invoice listed the faucet and two supply lines. No mention of the angle stops.

The lead tech pulled the photos the restoration contractor had taken on Monday. Both angle stops were the original chrome compression stops, visibly corroded at the stem. The hot-side compression nut on the stop outlet sat crooked on its thread. Water had come from there, not from anything the new faucet connected.

Now ask what kind of leak that is. A cross-threaded compression nut on a worn stop typically seats well enough to hold dry for minutes and weeps slowly as the joint creeps under constant pressure and thermal cycling. The customer's own account was that the cabinet floor was dry on Friday evening when she put a cleaning bottle away. Over a weekend, a few drops a minute is enough to soak a particleboard sink base.

A dry paper towel checked at minute 15 would very likely have stayed dry on that joint. The rule did not match the failure mechanism. It added a longer version of a check that works for fast leaks to every job, including the 74 percent that never touch a pressurized joint, and it would probably have missed the slow weep it was written to stop.

The gaps in the record were part of the finding

Three gaps turned up, and each told the shop something:

• No timestamps for the first three months. The cost figure for April to June is an estimate. The review used it anyway and labeled it, because the alternative was to report only half the cost, which flatters the rule.
• No job-type field on the hold step. To count how many of the 2,340 jobs involved a pressurized joint, the office manager matched invoices to line items. 610 jobs, about 26 percent, had a pressure-joint line (supply connections, valve work, water heater connections, repipe sections). The rest did not.
• No record of what was done to existing components. The March job's real cause, an old stop reused under a new connection, was invisible in the paperwork because the shop had no habit of noting the condition of what it connected to. That gap is what had let the wrong rule get written in the first place: with nothing in the record about the stops, the owner reasoned from the only thing visible, which was that water came out after the tech left.

The review meeting

The owner, the lead tech and the office manager met for forty minutes. The useful frame was not "was the rule wrong" but four questions asked in order:

1. What was the base rate of the incident before the rule? The insurer's claim history gave two leak claims over the previous four years, on roughly 11,000 jobs. That is about one in 5,500 jobs, both on fixture work. The owner had been carrying it in his head as "this happens," because the March job was vivid and recent. See related: Availability Bias After a Dramatic Callback.
2. What does the rule cost per year at that rate? About 546 tech-hours per nine months, so roughly 730 a year at the same volume.
3. What does the rule catch that the old practice did not? Nothing, on the record.
4. What actually caused the incident, and what would catch that? A reused corroded stop and a slow weep, neither of which a 15-minute hold reliably finds.

The owner's first reaction was to keep the rule "just in case." The lead tech's answer was the one that moved the room: the 546 hours had been taken mostly from the last job of the day, which is where techs were already rushing, and the March job had been a last job of the day. The rule had made the conditions that produced the incident more common, not less. See related: Hindsight Bias in the Callback Review, for how to read the original tech's choices without assuming he should have known.

What replaced it

The blanket hold came off. Three narrower rules went on, each tied to a mechanism:

• Scoped hold. On jobs where a pressurized joint is made or disturbed, pressurize, dry the joint, check at 10 minutes. That covers the 26 percent of jobs where the hazard exists. At the April to December volume, 610 jobs at 10 minutes is about 102 tech-hours against 546, a saving of roughly 444 tech-hours over nine months.
• Existing-stop rule. When a fixture connects to an existing angle stop or valve showing corrosion, a stiff stem or a compression fitting that has been made up before, the tech photographs it and offers replacement as a line item. If the customer declines, the note says so. This targets the actual cause and gives the record the field it was missing.
• Next-morning call. Any pressure-joint job finished after 3 pm gets a two-minute call from the office the next morning: "Any water under the sink?" That is the only check in the set that works on a slow weep, because it waits long enough for one to show.

Every rule went into the checklist with a review date nine months out and a field that logs what it found. That last part is the one most shops skip, and it is the reason this review was possible at all.

How they confirmed the replacement

The shop gave the new set the same treatment the old rule finally got. Over the next three months, January to March:

• 212 pressure-joint jobs ran the 10-minute hold, median 9 minutes logged. Two wet findings, both visible within the first minute.
• 41 existing-stop photographs, 17 replacements sold, 24 declines noted.
• 38 next-morning calls. One customer reported a damp cabinet floor. The tech went back that morning and found a supply line nut on an old stop the customer had declined to replace, snugged slowly by a quarter turn. The note from the original visit showed the decline. Total damage: one wet shelf liner.

One slow weep in three months, caught before it reached a floor, by the check designed for slow weeps. That is not proof the new set works; at one event, it is an anecdote pointing the right way. It is, though, the first finding in a year that any rule could actually claim.

The portable part

The pattern here is not that rules after incidents are bad. It is that a rule written from one vivid case usually gets its scope from the emotion of the case and its mechanism from a guess. The review that catches it needs only four numbers: the base rate before, the cost per year, what the rule found that the old practice would not have, and whether the rule could have stopped the incident it was named for. A rule that cannot answer the fourth one was written about the feeling, not the failure. See related: Writing Decision Rules So the Office Stops Asking You, for how to write the replacement so it carries its own scope and review date.

References

• Tversky, A. and Kahneman, D., "Availability: A Heuristic for Judging Frequency and Probability," Cognitive Psychology, 1973 (the basis for overweighting a recent vivid event)
• See related: Availability Bias After a Dramatic Callback
• See related: Hindsight Bias in the Callback Review
• See related: Base Rates: The Number You Should Start From
• See related: Writing Decision Rules So the Office Stops Asking You