A solder bridge is paste that has nowhere to go. Either the deposit on the two pads holds more solder than the gap can pull apart when it melts, or the stencil did not seal and paste bled under the aperture edge onto the mask between the pads. On a 0.5 mm pitch connector the copper gap is 0.25 mm, the mask dam takes 0.09 mm of that, and the print has to land inside 0.08 mm on each side of the dam. No reflow profile repairs a print that is that far out.
So the first measurement is the deposit, not the oven. A single web between two pins with the rest of the row clean points at one aperture or one patch of stencil that is not sealing. A bridge that moves along the row from panel to panel points at board support and gasketing. A bridge that starts on every unit after a paste lot change points at the paste. Each has a different fix, and only one of them is in the printer recipe.
What the pad geometry actually leaves you
Take a 0.5 mm pitch header: 0.25 mm pad, 0.25 mm space. Green mask dam is held at 3.5 mil, which is 0.089 mm, and other colours at 4 to 5 mil. Subtract the dam and the copper clearance either side of it is about 0.08 mm. Print the aperture at 100 percent of pad width and the paste edge sits exactly on the pad edge, with the mask shoulder right there to catch anything that slumps. Print at 90 percent and you gain 0.012 mm of margin per side.
The mask itself is only 20 to 25 um thick at the shoulder. Solder does not wet cured mask, but liquid metal sitting on a 25 um ledge does not need to wet anything to form a web: it needs volume and a place to sit. That is why widening the dam is the wrong lever. A wider dam takes copper away from the pad, the fillet loses its footprint, and a mask sliver at 0.5 mm pitch breaks and lands on the next pad, which is a bridge that no paste change will fix.
Why release and gasketing decide it before the oven does
Paste transfers from an aperture on area ratio, the open area divided by the wall area. Below about 0.66 the deposit becomes inconsistent from aperture to aperture. At 0.4 mm pitch, a 0.19 mm wide, 1.0 mm long aperture in a 0.12 mm stencil gives 0.19 square millimetres of area against 0.286 square millimetres of wall, which is 0.66 exactly, so the aperture is already at the limit. Put the same aperture in a 0.15 mm stencil and the ratio drops to 0.53, release falls to 60 or 70 percent on some apertures, and the press operator does the natural thing: more squeegee pressure and a slower stroke.
Pressure that is high enough to lift the blade edge, or high enough to press a slightly bowed panel flat, lets paste escape sideways under the stencil when it lifts after the stroke. The deposit grows a tail onto the mask, and the tail joins two pads at reflow. The usable rule is the lowest pressure that clears the stencil in one stroke, with board support and panel flatness fixed first. Warp at 0.75 percent is 1.9 mm across a 250 mm diagonal, enough to open a gap under the foil.
Where the extra paste comes from on a mixed panel
Pin in paste is the usual culprit on a board that also carries a fine pitch connector. A 1.0 mm hole in a 1.6 mm board is 1.26 cubic millimetres of barrel. A 0.6 mm pin takes 0.45 of that, so the joint needs about 0.8 cubic millimetres of metal. Paste at 50 percent metal by volume means about 1.6 cubic millimetres of paste, roughly three times what a 1.2 by 0.25 mm aperture holds at 0.12 mm. That volume has to come from wide apertures cut in the same foil that prints the 0.5 mm pitch row, and the honest answer is a step stencil at 0.15 and 0.12 mm, not a global thickness increase that overfills every passive on the panel.
Two more sources. Paste left on the stencil between wipes dries at the aperture wall, release drops, and the operator compensates with pressure. And a mask skip over a trace that runs between two pads turns the trace into a wick: the solder follows the exposed copper and the bridge forms under the part, where AOI cannot see it and functional test finds it as a short.
Three ways it gets misdiagnosed
Slowing the conveyor to let the solder pull back. By the time the joint is at 217 C the bridge is already solid metal. Surface tension separates a marginal bridge only when the volume is close and the flux is still active. A longer soak burns the flux out and leaves the bridge, plus a duller joint.
Blaming the paste lot. Paste slump matters, and a type 3 powder with particles up to 75 um is the wrong choice under 0.5 mm pitch where the deposit is 0.12 mm thick, but a lot change that fixes one panel and not the next is not a paste problem. Measure the deposit with SPI, per aperture. A print where 30 percent of apertures come in below half volume is a stencil support problem wearing a paste costume.
Fixing it in rework. A bridge that is reworked with a hot air pencil and wick loses mask at the shoulder, and the next print bridges in the same place because the mask ledge that was holding the solder is gone. Every rework on a 0.5 mm pitch row makes the next unit harder, which is why the fix has to be in the print.
What SUPERB controls at the printer
Every board passes 3D SPI before reflow, with volume, height and area logged per aperture. Stencil thickness is chosen on the component mix at 0.10 to 0.125 mm, with a step stencil where a through-hole connector and 01005 passives share a panel. Squeegee pressure is set to the lowest value that clears the foil in one stroke, paste type 4 is used at 0.5 mm pitch and below, reflow is twelve zones with SAC305 and nitrogen under 500 ppm, and inspection is 3D AOI plus DIP AOI with X-ray on hidden joints. The bare board side is held at 3.5 mil mask dam on green and 4 to 5 mil on other colours, 50 um on HDI, minimum pitch 0.45 mm standard and 0.30 mm on HDI, BGA pad 10 mil on HASL and 8 mil on ENIG, and boards ship with a section report.
The honest boundary: the aperture belongs to the customer's pad. Where the artwork carries a 1:1 aperture on a 0.5 mm pitch row, we say so at the pre-production check, because one file changed before the first print costs nothing. Assembly limits are on turnkey PCBA assembly, inspection on AOI compared with X-ray, and the bare board window on PCB capability.