The last pass is the one nobody watches. The operator has moved on to the next piece, the program is counting down, and the finish looks acceptable under shop lights. Then the slab goes out, the installer sets it under a window, and a soft wave shows up across the front edge that nobody saw in the shop. The cause is usually the same: the tool never removed what the last pass told it to remove, and nothing was done afterward to let the machine catch up with its own instructions.
Precision grinding solved this decades ago with two ideas stone shops borrow without always naming them. One is the spark-out pass, a final pass run with no additional infeed, which lets the elastic deflection stored in the machine, the tool and the part unwind while the abrasive keeps cutting. The other is dwell time, the pause at one position, which is the same physics working against you. Knowing which surface problems each one fixes, and which they cannot touch, separates a shop that hits gloss consistently from one that reworks.
Deflection, and Why the Last Commanded Pass Leaves Stock Behind
Every grinding or polishing setup is a spring. Spindle bearings, backer pad, the resin body of the abrasive, the fixture, the slab and whatever supports it all deflect elastically under load. When the head is pushed into the stone, force builds until the abrasive cuts at the rate the feed demands, and at that instant the whole loop is bent. Commanded depth of cut and achieved depth of cut are two different numbers, and the difference is not a machine fault. It is stored energy.
That gap is small on a rigid setup and large on a flexible one. A hand polisher held against a cantilevered overhang, an undersupported miter, or a thin edge on a sagging rail is a low-stiffness system, and low-stiffness systems store a lot of deflection for a given force. The operator feels the pad pushing back and reads it as cutting, when a fraction of that pressure is going into bending the loop rather than into the stone. When the pass ends, the bent loop is still holding material against the abrasive.
A spark-out pass is the fix, and the definition is precise: continue the same motion with the infeed instruction set to zero. Nothing is commanded deeper. The system relaxes toward its unloaded position, the abrasive removes the stock that the deflection had been protecting, and the surface trues up. Grinding research describes this as releasing elastic stress from both tool and workpiece and eliminating the micro-waviness left by cutting pressure. It is also the stage of a grinding cycle most responsible for the final roughness value.
Two things follow. The improvement decays exponentially, so the first spark-out pass does most of the work and each one after it does less. And roughness keeps improving after the gross deflections have relaxed, because the abrasive is now plowing and burnishing rather than cutting hard. Systems with high cutting stiffness relative to machine stiffness are slow to true up and need longer spark-out, which is the technical version of what every fabricator knows: floppy setups need more finishing time.
Running Spark-Out Passes on Real Work
The mistake is treating spark-out as something you do when a piece looks wrong. It belongs in the program or the written hand procedure, so it happens on every part whether or not anyone is worried. That is the only way it becomes a control rather than a rescue. What changes between methods is how you make the infeed zero and how you know the pass is finished.
By Hand, at the Edge and on the Face
By hand, zero infeed means zero added pressure: a final pass carrying the weight of the machine only, pad floating, same water volume and same speed as the working passes. The stroke overlaps the way it did before and never stops. The feel changes as it works, with the pad going from grabbing to gliding and the sound flattening out. When the tool stops loading, the pass has done its job.
On Inline and CNC Machines
On a CNC, spark-out is a repeat of the last toolpath at the same Z with no step-down. On a line polisher it is a head running at contact pressure with no added down-force, or a second run through the last stations. The advantage over hand work is repeatability: the pass is identical on every part, so the finish variation you measure afterward comes from the material and the pads, not the operator.
Spark-Out and the Abrasive Sequence
A spark-out pass cannot rescue a skipped grit. The standard wet progression runs 50, 100, 200, 400, 800, 1500 and 3000 grit, usually finishing with a buff, and each step exists to erase the scratch pattern of the one before it. Jump from 100 straight to 400 and the finer pad cannot reach the bottom of the coarse furrows; they survive as haze and swirl. Spark-out corrects error on the scale of machine deflection, not a missing stage.
Spark-out also pays off unevenly across the sequence. During shaping, removal rates are high enough that deflection error is swamped by the stock still coming off. By the pre-polish and gloss stages, residual deflection is the same order of magnitude as the scratch depth you are trying to remove, so the zero-infeed pass is the difference between a clean surface and one that reads uneven under a light.
| Stage | What the pass is doing | Value of a zero-infeed pass |
|---|---|---|
| 50 and 100 grit | Bulk removal and geometry | Low; useful mainly as a flatness check before honing |
| 200 and 400 grit | Erasing the coarse scratch pattern | Moderate; evens out heavy and light areas of the previous pass |
| 800 and 1500 grit | Pre-polish, closing the surface | High; residual deflection is now the same size as the defect |
| 3000 grit and buff | Gloss development | Highest; the whole stage should be run at contact pressure only |
Pro Tip: Run the spark-out pass in the opposite stroke direction from the working pass, at tool weight only. Anywhere the machine still loads up or the pad still grabs is a spot that was holding deflection, and you have just found the high area without a straightedge or a gloss meter.
Dwell Time: The Divot, the Burn and the Over-Polished Halo
Dwell is the same physics pointed at a single spot. Zero infeed does not mean zero removal, which is why spark-out works and why a head parked on a turning workpiece keeps eating. The stored deflection relaxes into one patch instead of across the surface. On a soft or resin-rich material you get a shallow divot; on a hard one, a ring of higher gloss around a slightly dished center, the over-polished halo that catches light at the wrong angle and is nearly impossible to blend out.
On a CNC the worst dwell is unintentional. Every move follows a velocity profile that ramps up to the programmed feed and ramps back down, so at a tight corner the tool slows almost to a stop before accelerating away, and contact time there is far higher than along a straight run. Burn marks on any machined surface come from excessive heat, and heat has three usual sources: rubbing at too low a chip load, a dull or glazed tool, and dwell. Corners and lead-in points collect all three.
The correction is to ramp rather than stop. Lead the tool in and out on an arc or tangent, run past the end of the work onto sacrificial material where the part allows it, round program corners instead of squaring them, and lower the programmed feed rather than letting the controller do the slowing. On a manual polisher the rule reduces to one habit: the pad moves before it touches and keeps moving until it leaves.
Inline polishers add a failure mode of their own. If the conveyor stops with heads down and water running, every head marks its slab at once, and the damage is uniform enough to pass a quick visual check and fail later. Interlocking a conveyor stop to a head lift is worth doing. The ends of an oscillating stroke deserve the same attention, since the head reverses there and lingers over the outer bands.
Material changes the tolerance here. Granite, at roughly 6 to 7 on the Mohs scale, is forgiving. Marble sits around 3 to 5 and dishes quickly, so a pause granite shrugs off leaves a visible low spot. Engineered quartz is roughly 90 percent ground quartz bound in polymer resin, and that resin softens, discolors and burns from friction heat long before the mineral is in trouble, so it requires diamond tooling rated for engineered stone, controlled speed and steady water. Sintered slabs test hard, with published figures spanning roughly 7 to 8.5 Mohs and varying by configuration; they wear tooling fast and also demand engineered-stone-rated tooling and light, continuous motion.
Water, Pad Condition and Machine Health Over the Long Run
Water flow must not drop during a spark-out pass, and the temptation to cut it is real, because the pass is quiet, low-force and looks like it is barely doing anything. It is doing something. A dry final pass is the fastest way to ruin a resin-bond pad: the bond scorches, smears resin across the stone, glazes the diamond so it stops cutting, and leaves a haze no further polishing will lift. Wet-rated pads run dry once are often finished, and the stone has to go back several grits.
Flow also carries slurry out of the interface. During spark-out the abrasive works on a very thin layer, so loaded slurry between pad and stone changes what the surface sees, and recirculated water heavy with fines will re-cut a face you just closed. Check clarity at the head, not at the tank, and give the final stations the cleanest water available.
Pad condition changes the deflection curve directly. A worn pad is thinner and stiffer, a partially delaminated one is soft on one side, and either makes the same programmed pass behave differently than it did last week. That is why finish drifts when nothing in the program changed. Retiring pads at a consistent point, rather than running them until they visibly fail, keeps the spark-out pass repeatable. Backer choice matters for the same reason: rigid backers hold flatness, flexible ones follow contour, and they store deflection differently.
Then there is the machine. Worn spindle bearings, loose head mounts, tired rails and out-of-balance rotating assemblies all lower system stiffness, which lengthens the spark-out time you need and raises the odds of chatter. If parts that used to clean up in one zero-infeed pass now need three, that is a measurement, not an annoyance. It says the loop got softer, and chasing the bearing beats buying finishing time on every part.
Measuring the Result, and What Spark-Out Will Not Fix
Gloss is measurable, and guessing at it is a bad habit. Gloss meters follow ASTM D523 and ISO 2813, and 60 degrees is the reference geometry, with 20 and 85 degrees used for very high and very low gloss. Readings are in gloss units on a scale where a calibration standard is defined as 100, so the number is comparative rather than physical. Polished granite commonly reads in the 70 to 90 range, higher for mirror finishes, varying by stone and configuration. Take a grid of readings, not one.
The meter tells you level, not shape. For shape, nothing beats low-angle raking light: put a bare lamp almost parallel to the surface, get your eye near the same plane, and sight along the face. Waviness no gloss reading will catch shows up as bands of light and dark. Do it on every finished face before it leaves, in the same spot, with the same lamp. It costs nothing and it catches what customers notice from across a kitchen.
Now the honest limits. Spark-out fixes residual stock left by deflection, the light micro-waviness that cutting pressure creates, and edges that finished heavier than the field. It will not fix a skipped grit, because the scratch is deeper than the correction. It will not fix orange peel caused by a glazed pad smearing resin, because the pad is the problem and running it longer makes the smear worse. It will not fix a dished profile caused by a dwell divot, because the surrounding material has to come down to meet it.
Chatter is its own category. Regenerative chatter happens when the tool cuts a slightly wavy surface, then tracks and amplifies that waviness on the next revolution, leaving regular repeating marks across the direction of the cutting action. It comes from vibration: imbalance, worn bearings, an unsupported part, or external sources in the building. A zero-infeed pass at reduced force may soften the appearance slightly because cutting force drops, but the pattern is generated by the machine, and only fixing the machine removes it.
The cost argument runs on the back of a work order. One zero-infeed pass at the last two grits adds seconds to a cycle and a marginal amount of pad wear. Rework does not work that way: a haze or wave found later means stepping back down the sequence, refinishing the whole face so the correction blends, and re-inspecting. If the piece is installed, add protection, containment, water management and a crew on site. The ratio is not close.
All of this depends on abrasives being consistent from box to box, because a spark-out pass only proves out on pads that behave predictably. Full wet progressions such as the PUMA wet polish pad set keep every stage from the same family, which removes one variable from finish troubleshooting. For quartz and other engineered surfaces, resin-specific tooling like the X Series white resin hybrid pads is built for materials where heat, not hardness, is the constraint.
Free Tool
Polishing Pad Sequence Builder — Build a complete grit progression for your stone and starting condition, so you know exactly which stages need a zero-infeed finishing pass and which do not.
Build Your Sequence →Finish the Last Pass Properly
Dynamic Stone Tools stocks wet and dry polishing pads, buffs, backers and edge wheels for granite, marble, engineered quartz and sintered slabs, plus the polishers and water feeds that keep them cutting cool.
Shop Polishing Tools →