How to Torque a Joint You Cannot Re-Check Later

Why this matters

Most joints get a second chance. You tighten them, the system runs, somebody checks them at the next service, and a joint that relaxed gets brought back up. A concealed joint gets no second chance. Behind drywall, under tile, inside a poured chase, under a roof curb, or buried in a slab, the first time anyone touches it again is after the damage. Every shortcut that is survivable on an accessible joint - a reused fastener, a value applied without noting whether it was dry or lubricated, closing up before the joint has settled - becomes permanent here, and it fails weeks later inside a finished wall.

The work below is organized around the record card you build as you go. The card is not paperwork on top of the job; it is the job. If the card cannot be completed, the joint is not ready to close.

Test with water, not air

The proof test is the part of this procedure that can hurt someone, and the choice of test medium is the control.

• Test hydrostatically wherever the system allows it. Water is nearly incompressible, so a joint that lets go under a water test releases almost no stored energy. It sprays and stops.
• A pneumatic test on the same joint stores an enormous amount of energy because gas is compressible, and a failure is not a spray, it is a release. Where a gas or air test is genuinely unavoidable, keep every person out of the line of the joint and any cap or plug, use a regulator plus a relief device set at or below the test pressure, and never exceed the test pressure specified by the governing code and accepted by the authority having jurisdiction.
• Relieve the test pressure to zero and confirm it on a gauge before touching any fastener, including between passes. 29 CFR 1910.147 requires stored energy to be relieved, disconnected, or restrained before servicing, and a joint held at test pressure is a loaded assembly.

Field one: is the joint actually unre-checkable

Answer this honestly first, because it is the cheapest field on the card and it deletes most of the rest.

A great many "concealed" joints are simply inconvenient. An access panel, a removable ceiling tile, a hatch in a soffit, or moving a joint 18 inches so it lands in a chase with a door instead of behind sheet rock turns an unre-checkable joint into a serviceable one. That change costs an hour at rough-in and saves a demolition later.

Write on the card either "accessible via [the specific access]" or "concealed, no access", and if it is the second one, note whether that was a design constraint or a choice. Where a code or the manufacturer requires a joint to be accessible, that requirement decides this field before you do.

Field two: the fastener, the gasket, and the lubrication state

A concealed joint gets new fasteners from one box, and that is not negotiable. A reused fastener carries an unknown friction history, so the torque value on your wrench maps to an unknown clamp load. Mixed sources means mixed plating means mixed friction across a single joint.

On the card, record the fastener specification including coating, the gasket or sealing element, and - the field most often left blank - whether the threads and bearing faces were assembled dry or lubricated, and with what. A recorded torque value without its lubrication state is not reproducible, because the same number produces substantially different clamp loads dry and lubricated. The published value you use states which condition it assumes; match the assembly to it rather than the other way round.

Also record that the mating faces were checked and clean. A radial scratch across a sealing face is a leak path that no amount of clamp load closes, and it is the one defect that is trivially fixable while the joint is open and impossible afterward.

Field three: the control method

Torque control on a good day delivers clamp load within roughly plus or minus a quarter to a third of target. On an accessible joint you absorb that scatter with a re-torque. Here you cannot, so consider buying the accuracy up front where the manufacturer supports it:

• Angle control, where you snug to a defined light torque and then turn a specified additional angle, bypasses most of the friction scatter. Use the manufacturer's snug value and angle; do not invent them.
• Load-indicating hardware gives you a visible acceptance signal rather than a wrench reading.
• Bolt stretch measurement, where both ends are accessible, measures the quantity you actually want.

If torque is the only method available, say so on the card, because it tells the next person how much confidence the number deserves.

Field four: the settling schedule and the acceptance test

This is the field that separates a concealed joint from an ordinary one. A freshly assembled joint loses preload on its own: surface roughness embeds, the gasket creeps, and stress relaxation bleeds tension off at temperature. On an accessible joint you catch that at the next service. Here, you have to make the relaxation happen before you close the wall.

Run the passes and record every one. A workable default schedule, to tune to your own systems:

1. Pass one in a crossing pattern at roughly a third of the final value.
2. Pass two in the same pattern at roughly two thirds.
3. Pass three at the full value.
4. Pass four at the full value, after the joint has been held at test pressure for at least 30 minutes.
5. Pass five at the full value, after a full thermal cycle up to operating temperature and back to ambient, on any joint that will run hot.

Acceptance, stated so it can be checked: per fastener, on a full pass at the specified value, no fastener takes any additional rotation. Unit of analysis is the individual fastener, not the joint average, and the Boolean is all of them, not most of them. One fastener still turning means the joint is still settling and you run another pass. Do not close on a pass where anything moved, and do not average.

Field five: the proof test and its hold time

The test pressure and the minimum hold duration come from the governing code and the authority having jurisdiction, not from an article. What is yours to decide is how much longer than the minimum you hold a joint you will never see again.

For a concealed joint, hold for at least twice the code minimum duration and, where the schedule allows it, leave it under test overnight. The reason is that the failures this procedure exists to catch are slow: a joint that is marginally under-clamped does not fail in the first ten minutes, it weeps under a slowly relaxing preload. A short test proves the assembly, a long test proves the assembly plus the settling.

Record the test medium, the pressure, the start and end times, the gauge reading at both ends, and the ambient temperature at both readings. That last one matters because a temperature change moves the pressure of a closed system on its own, and a pressure drop that tracks a temperature drop is not necessarily a leak.

Field six: where it is

The last field is the one nobody fills in and everyone needs. Record the joint's location by dimension from two fixed references that will still exist after finishing - a corner and a floor line, not a stud that gets covered. Photograph the open joint with a tape in frame. Note the direction the fitting faces.

A concealed joint that eventually does fail costs a demolition, and the size of the demolition is set entirely by how well its location was recorded.

The card filled in

A hydronic manifold connection in a wall chase, to be closed behind board and tile. Four bolts on a flanged connection, the system runs at 180 degrees F, the shop's plumbing code sets the test pressure and a minimum hold.

• Field one: the chase can take an access door, but the customer's tile layout will not accept one on this wall. Recorded as concealed, no access, customer-driven. That decision is on the card and initialled, which matters later.
• Field two: four new bolts and nuts from one box, coating recorded; new gasket; assembled dry, because the published value is a dry value; faces inspected and clean, one light burr dressed off the flange face before assembly.
• Field three: torque control, because the manufacturer publishes no angle procedure for this connection. Noted on the card so the confidence level is explicit.
• Field four, the passes as they actually ran:
  • Pass one, crossing pattern at about a third of value: all four bolts took rotation, as expected.
  • Pass two at about two thirds: all four took rotation.
  • Pass three at full value: all four took rotation.
  • Pass four at full value, after 30 minutes at test pressure: two of the four took additional rotation, the other two did not move.
  • Pass five at full value, after the system was brought to 180 degrees F and returned to ambient: one bolt took a small additional rotation, the other three did not move.
  • Pass six at full value: no bolt moved. Acceptance met.

Six passes, and the acceptance criterion was met on the sixth, not the fifth. That is the whole point of stating the Boolean as all fasteners rather than most: on pass five, three of the four bolts were done and the joint was 75 percent settled, which is exactly the state that reads as finished to a tech in a hurry. Closing on pass five would have buried one bolt that was still relaxing.

• Field five: hydrostatic, at the code test pressure, held overnight - roughly ten hours against the code minimum. Gauge stable, start and end readings recorded, ambient noted at both, which explained a small overnight change without anyone having to guess.
• Field six: dimensions from the room corner and the finished floor line, photograph with a tape in frame, fitting orientation noted.

The failure mode if the card is skipped: the joint gets one pass at value, a short test, and a wall. It seeps eight weeks later, the customer sees a stain on the tile, and now the shop is removing tile to find a joint whose location nobody wrote down, with no record of what was installed and no way to tell whether the cause was preload, gasket, or a scratched face. The repair is not the joint, it is the wall - and the argument about who pays for the tile is decided by whether field one was recorded and initialled.

What changes the schedule

A joint that never runs hot can drop the thermal-cycle pass, but not the pressure-hold pass. Embedment and gasket creep happen without heat.

A joint with a soft or fibre gasket relaxes more and needs more passes than a hard-faced metal-to-metal joint. Expect the acceptance pass to arrive later and do not read the extra passes as a defect.

A joint under a slab or in a pour raises the stakes on field six far more than on field four, because the failure cost is dominated by locating and reaching it. Photograph more, dimension more.

A manufacturer-specified re-torque interval overrides all of this. If the equipment documentation says re-torque after a stated number of operating hours, the joint is not eligible to be concealed without access. That instruction is the manufacturer telling you the joint will relax on a schedule you cannot pre-run, and burying it makes the installation non-compliant with the instructions the equipment is listed under.

References

• 29 CFR 1910.147, energy control, for relieving or restraining stored energy before servicing, including a joint held at test pressure
• The governing plumbing, mechanical, or fuel gas code and the authority having jurisdiction for test pressures, minimum hold durations, and any accessibility requirement for concealed joints
• Manufacturer assembly documentation for torque values and their stated lubrication condition, angle procedures where published, and any specified re-torque interval
• See related: What Torque Actually Controls and What It Does Not; The Torque Sequence That Matters; Thermal Expansion Mismatch Inside a Joint