How to Tell a Cold Joint From a Starved One

Why this matters

Cold and starved are the two ways a capillary joint fails, they get the same word in the field ("bad joint"), and they need opposite corrections. A cold joint got heat in the wrong place. A starved joint got heat in the right place and not enough filler where the heat was. Call a starved joint cold and you will add heat to the next one, overheat the flux, burn it off ahead of the filler, and starve that joint too. The two look almost nothing alike if you read the right thing, and almost identical if you read the wrong thing.

Before applying heat to any joint: confirm the line is open to atmosphere at a point you can see. A closed section holding liquid becomes a pressure vessel the moment you heat it. Isolate and drain, then crack a fitting or open a valve downstream and leave it open, because 29 CFR 1910.147 requires stored energy relieved or restrained before service and trapped liquid under a torch is exactly that. Never heat a line that has held refrigerant or a flammable fuel until it has been recovered and purged: refrigerant decomposes over a flame into hydrogen fluoride and other acute irritants, which is an inhalation route, so recover and purge rather than relying on a glove or a face shield. Hot work needs its fire prevention and its watch under 29 CFR 1910.252(a) in general industry or 1926.352 on a construction site, together with NFPA 51B in the edition your authority having jurisdiction has adopted or your insurer requires, which is how that standard reaches you rather than on its own.

The gate: does the fillet feather, all the way around

There is exactly one read that separates the two, and it is at the mouth of the joint, on the outside, on the full circumference.

Look at where the filler meets the base metal at the fillet. A joint that reached wetting temperature shows a filler surface that thins to a feathered edge and disappears into the base metal, with a continuous transition you cannot catch a fingernail on. A joint that did not shows filler standing proud with a rounded or scalloped edge, often dull and slightly granular, sitting on the base metal the way a drop of water sits on a waxed hood.

That single read splits every failed capillary joint into two families, and the reason it works is worth holding onto, because it is also the reason the read is so often done wrong:

• A feathered fillet proves the mouth reached wetting temperature. It proves nothing about how far the filler went in. The fillet is the visible bead at the entrance to the gap. The joint's strength is the filler inside the gap, and nobody can see that.
• So a feathered fillet is necessary and not sufficient, and a non-feathered fillet is decisive on its own.

Run the gate before you form any opinion about the torch, the tech, or the filler:

1. Rotate your light, do not rotate your head. A raking light across the fillet shows the transition; a light behind your shoulder washes it out and makes every joint look wetted.
2. Read all four quadrants on a horizontal joint and all around on a vertical one. Wetting is a local property, not a joint property, and the failure is almost always one arc.
3. Run a fingernail or a pick across the transition at the worst-looking arc, with the joint cool. A wetted edge has no step. An unwetted one catches.
4. Only then decide which family you are in. Everything after this branches.

Case one: the joint whose 6 o'clock arc never wet

A 1-inch copper-to-brass valve connection on a horizontal line, brazed, wept as soon as the line was filled.

The gate: at 12 and 3 o'clock the filler feathers cleanly. At 6 o'clock it stands proud with a rounded lip and a dull grey cast, and a pick catches on the transition. That arc did not reach wetting temperature. Cold, at the bottom of the joint.

The mechanism follows the read. The torch was worked from above, and the brass valve body is a far larger heat sink than the copper tube on the other side of the joint, so heat leaving the joint into the valve came out of the underside first. The filler was fed at the top, melted, and ran around the outside of a cold arc rather than into it. Nothing about that is a filler problem or a flux problem.

The direction check matters here, because it is what tells you the read is real rather than convenient. If the underside had reached wetting temperature, the filler would have been drawn up into the gap from below, against gravity, because capillary force across a gap of a few thousandths of an inch is much larger than the head of a few millimetres of liquid filler. Filler that refuses to climb is not fighting gravity, it is telling you the surface below it is not wet. That is the whole reason a correctly heated joint can be fed from the bottom of a vertical run and still fill.

The correction is heat placement, not more heat: work the heat on the heavier member so the two sides arrive at temperature together, keep the flame off the filler itself, and feed at the arc opposite the torch so the base metal is what melts the rod. Adding time with the same placement overheats the top arc, burns the flux out at 12 o'clock before the bottom arc is ready, and gives you a joint that is cold at the bottom and oxidised at the top.

Case two: the joint with a perfect fillet that failed eleven days later

A 3/4-inch copper-to-copper socket joint, brazed, fillet bright and feathered on the full circumference, passed the pressure test at fill, wept eleven days later after the system had run up to temperature and back down a few times.

The gate: the fillet feathers everywhere. The mouth reached temperature. Not cold. Starved.

Starved means the filler did not travel the full depth of the socket around the full circumference, and the reason is almost always something that stopped the far side of the cup from getting hot, or something that broke capillary flow:

• A local heat sink you introduced. In this case a wet rag had been wrapped around a valve two inches away to protect it. That rag held the far end of the cup below wetting temperature the entire time the mouth was above it, which is why the mouth wet perfectly and the depth did not.
• Gap outside the range the filler needs. Capillary draw depends on the gap: too tight and flow is choked, too wide and there is no capillary rise at all, only gravity. An out-of-round tube end, an over-reamed cup, or a tube not fully bottomed all put part of the circumference outside the range. The workable range is filler-specific and geometry-specific, so take it from the filler manufacturer's datasheet for that alloy and that base metal pair rather than from a remembered number.
• Flux burned off ahead of the filler. Flux has a working range. Held above it while you wait for a heavy member, it is spent before the rod touches the joint, and the oxide it was there to strip comes straight back.
• Fed at one point on a large joint. The filler follows the heat and fills toward it. On a large-diameter socket, one feed point with an uneven heat pattern fills one side.

What made this one fail on a delay rather than immediately is the part worth learning. A shallow but continuous ring of filler at the mouth does seal. It seals against static pressure, which is why it passed the test. It carries a small fraction of the bond area the full socket would have, so the differential movement of a system going hot and cold works that small area at a stress the full joint would never see, and it cracks after a number of cycles rather than at the first one. A joint that passes the test and fails weeks later on a thermally cycling system is a fill-depth question until proven otherwise.

Confirming it when you get to cut one out

You only get certainty by sectioning, and it is worth doing once per crew rather than never.

Before cutting, drain and vent the section and confirm it is at zero on a gauge you can see, not on an assumption about a closed valve. Cut with a tubing cutter rather than an abrasive wheel where the tube allows it: an abrasive wheel on a brazed joint throws hot particles and puts filler alloy into the air, and if the filler is a cadmium-bearing silver alloy that fume is cadmium oxide, which is regulated under 29 CFR 1910.1027 and needs local exhaust or respiratory protection under a written program meeting 29 CFR 1910.134. Cutting galvanized steel in the same way releases zinc oxide fume. Both are inhalation routes and neither is addressed by gloves or a face shield.

Split the fitting lengthwise and read two things:

• Depth of penetration around the circumference. Full depth on the fed side and a shallow band elsewhere is the starved signature.
• The interface itself. On a cold arc the filler lifts off the base metal cleanly and the metal underneath is bright and untouched, because there was never a metallurgical bond, only mechanical contact. On a wetted arc the filler cannot be lifted without tearing.

Photograph it before it goes in the scrap barrel. A sectioned joint is the only artifact that ends the argument about whether a tech runs cold or runs starved, and the corrections are different enough that guessing costs you a second callback.

Sorting your own callbacks

Pull the last several brazed or soldered callbacks and put each one in a column using the fillet read from the photographs or from the removed fitting. The distribution tells you what to change:

• Mostly cold, on heavy members and dissimilar-mass joints: the crew is heating the joint rather than the heat sink. This is a torch-placement and tip-size correction.
• Mostly cold, spread over all joint types: the torch or the fuel is undersized for the diameters being run, and the tech is compensating with time, which spends the flux.
• Mostly starved, with clean fillets: something is stopping fill. Look at fit-up first (out-of-round ends, insertion depth, reaming practice), then at heat sinks the crew is adding on purpose, then at the feed point.
• A mixture with no pattern: stop reading joints and read the material path. Tube stored uncapped, cut with a dull wheel, or reamed with a burr left in place produces both defects at random, because the gap is random.

References

• 29 CFR 1910.147 for relieving trapped liquid and stored energy before applying heat; 29 CFR 1910.252(a) (general industry) and 29 CFR 1926.352 (construction) for hot work fire prevention and the fire watch
• 29 CFR 1910.1027, cadmium, for fume from cadmium-bearing filler alloys, and 29 CFR 1910.134 for the respiratory protection program that supports it
• NFPA 51B, fire prevention during welding, cutting and other hot work, in the edition adopted by your authority having jurisdiction or required by your insurance carrier
• Filler metal and flux manufacturer datasheets for joint clearance range, flux working temperature range, and base metal compatibility
• See related: Common Soldering + Brazing Mistakes Reference; Why a Joint Fails at the Interface and Not in the Metal