The Fire Watch That Was Posted and Did Not Work

If you suspect a concealed smolder right now, stop here

Do not open the cavity. A wall, ceiling, or roof space that has been smoldering is oxygen-starved and fuel-rich, and admitting air to it can produce flaming combustion in seconds, in your face, at the moment your hands are inside it. Call the fire department first and tell them you suspect a concealed fire from hot work. If you have a charged hose line or a rated extinguisher, position it and a second person before anything is opened, and let the fire department make the opening. Before any cavity is opened for any reason, de-energize the branch circuits serving that wall and prove them dead with the live-dead-live sequence in NFPA 70E-2021, 120.5, under 29 CFR 1910.333(b)(2); and do not cut into a wall finish of unknown age without the asbestos and lead survey, 29 CFR 1910.1001 and 1910.1025 in general industry, 1926.1101 and 1926.62 in construction.

The rest of this card reconstructs one fire from the paperwork it left behind, because the paperwork already contained the answer.

What the file held

A single-story multi-tenant building. Our crew replaced a section of 3 in steel pipe in the mechanical room. Nine hours and twenty-four minutes after the last spark, the building's fire alarm panel reported a smoke detector in tenant suite 104, then a waterflow alarm three minutes later. The fire was in the wall cavity between the mechanical room and suite 104.

Nothing in the file suggests a careless crew. That is what makes it worth reading. The permit was filled out, the watch was posted, the periods were kept, and the final check was signed.

Two more facts from the job photos: the cut was made 14 in from the wall face, on the same pipe that passes through the sleeve, and the sleeve was packed with fiberglass against old wood blocking.

Reading the record backwards

The watch field has one line, and the writer filled it truthfully. The form asked where the watch would be posted, and the answer was the room where the work was. Nobody lied and nobody was lazy. The form allowed a single answer to a question that had at least three: the room with the cut, the cavity the pipe passes through, and the tenant space on the far side.

The heat did not travel as sparks, it travelled as conduction. Oxy-fuel cutting brings the kerf to the melting range of steel, and the pipe is a continuous metal path from the kerf into the sleeve. The fiberglass packing around it conducts poorly; the pipe conducts well. The mechanism therefore predicts the highest temperatures in the wall at the point where the pipe passes through, and the fire department found the charring centered exactly there. Mechanism and symptom point the same way.

Check the other end of the range, because a direction you cannot invert is a direction you have asserted rather than established. Move the cut far enough down the pipe, a distance set by diameter, wall thickness, material and how long the torch is on it rather than by a remembered number, and the same torch puts almost nothing into the sleeve. Neither 1910.252(a) nor the adopted NFPA 51B edition publishes a conduction distance, which is why the control is opening and checking the far side rather than picking a clearance, because the pipe sheds heat to the room air along the way. Cut the same pipe with a mechanical saw and there is no thermal input to conduct. Make the pipe plastic and the far side never sees the heat because the pipe fails at the cut long before. So the variable that made this fire is the combination of a conductive member, a short distance, and a combustible in contact on the far side, and the record shows all three.

The watch period was kept and it could not have covered this. Last spark 12:10. Fire watch to 13:10, one hour, which meets the 60 minutes current NFPA 51B editions require in the edition your authority having jurisdiction has adopted and amended, and exceeds the 30 minute general industry floor at 29 CFR 1910.252(a). Monitoring to 16:10. First detection at 21:34 is 9 hours 24 minutes after the last spark, and 5 hours 24 minutes after the monitoring period closed.

That is not an argument that the periods are too short. It is the point that a watch period is a control against a fire that becomes visible in the space being watched. A member ignited by conduction can smolder inside a cavity for many hours before it produces anything a person standing in the next room can see or smell. The control for that failure is upstream: you either do not put the heat into the member, or you open and inspect the far side before you close the job. Both of those duties exist in the requirement to protect combustibles that cannot be moved and to check adjacent areas, and neither is a watch duty.

The record's silences are the finding. There is no line for the far side of the wall. No line saying the cavity was opened or that anyone entered suite 104. No temperature check on the pipe on the far side. No entry under alternatives, which is where a grooved coupling or a saw cut would have been considered and would have removed the mechanism entirely. The "all clear, no smoke or odor" at 16:10 is accurate and it describes the mechanical room.

The check that would have found it, and how it reads wrong

The measurement is the pipe's surface temperature on the far side of the wall, taken at the end of cutting and again at the end of the watch, compared against the same pipe well away from the work. Rising or materially elevated on the far side means heat crossed.

That measurement fails in one specific, flattering way. An infrared thermometer aimed at bare bright steel reads LOW, because bright metal has low emissivity and the instrument's default setting assumes a high-emissivity surface. The error is a systematic offset for that surface, not a random spread, and it runs in the direction that tells you the pipe is cooler than it is. The fix is a contact probe on the bare metal, or an infrared reading taken on a target of known emissivity such as a strip of matte tape applied to the pipe and given time to reach the pipe's temperature. Reading the pipe with a contact probe is a burn exposure, so it is done with an insulated-handle probe and contact-heat-rated gloves, not with a bare hand as a first check.

The comparison also needs a reference on the same pipe away from the work, because a systematic offset from ONE instrument cancels in the difference between two readings taken with it, leaving only that percentage of the difference. An absolute reading against a fixed threshold carries the whole offset; a difference does not.

What changes on the next permit

Three field changes, all of them small, all of them things the form prevented rather than the crew.

A volume list, not a location. Every adjacent space gets named on its own line: the room across each boundary, the cavity inside each boundary, the space below the slab, the ceiling plenum. Each line carries its own inspected-at time and initials. The hot work permit SOP makes this a required field; this incident is why.

A hold point where a conductive member crosses an assembly. If the cut is on metal that passes through a wall, floor, or roof, the far side is opened and inspected before the cut, and either the contact combustible is removed or the cut moves far enough away that conduction is not a factor. This is a decision made before the torch lights, because after it lights the only remaining option is watching.

A thermal check at close-out, with its basis stated. Contact probe, on the far side, differenced against the same pipe away from the work.

What it would have cost. Opening the sleeve from the suite 104 side, confirming the blocking, and either pulling the blocking back or moving the cut is on the order of half an hour of one person's time. Suite 104 was closed for six business days. Six business days of closure against half an hour of prevention is roughly 100 to 1 counted in eight-hour working days, and roughly 290 to 1 counted in calendar hours, so the basis is stated rather than left to the reader, and that arithmetic is the same on every job of this shape whether or not a fire ever happens.

Sibling-rule check. The perimeter reasoning here is the "behind" direction the hot work card names, and the sparks-and-openings analysis is not re-derived. The watch and monitoring durations are quoted separately as the federal floor and the adopted NFPA 51B period, with the longer governing, matching both sibling articles. The detection bypass in this record carries a start time and a restoration time, which the records card requires; the volumes field does not carry per-volume times, and that absence is the finding rather than an oversight in the retelling. No instruction in this card sends anyone into a cavity without the electrical, survey, and fire-department steps stated in the opening block.

References

• 29 CFR 1910.252(a) for fire prevention and protection during welding, cutting, and brazing in general industry, including the 30 minute fire watch floor; 29 CFR 1926.352 and 29 CFR 1926 Subpart F for construction work
• NFPA 51B, Standard for Fire Prevention During Welding, Cutting, and Other Hot Work, in the edition adopted and amended by your authority having jurisdiction, for the fire watch and monitoring periods and adjacent area checks
• NFPA 70E-2021, 120.5 for the live-dead-live proving sequence, applied under 29 CFR 1910.333(b)(2)
• 29 CFR 1910.1001 and 1910.1025 in general industry, 29 CFR 1926.1101 and 1926.62 in construction, for asbestos and lead before disturbing unknown wall finishes
• See related: What Hot Work Actually Requires Before It Starts; Running a Hot Work Permit That Would Hold Up