Layer Shifting: Sideways Steps and Staircase Prints, Diagnosed in 4 Questions

The print came off the bed looking wrong in a way you can feel with a fingertip: partway up the wall there is a clean step, and everything above it sits offset to one side, as if the print had been sliced and glued back crooked. That is layer shifting: from one layer onward, the printer kept laying plastic — faithfully, neatly, in the wrong place. In the FixMyPrint community dataset it was the sixth most reported problem: 1,138 of more than 20,000 help posts — a count of community help posts in that analysis, not a measurement of how often layer shifting occurs across all printers. Instead of a fix list, this guide asks four triage questions: a shift has a short list of mechanical families behind it, and the questions sort yours by evidence — which axis, how often, at what height — with a test narrowing the cause before anything is tightened, slowed or reconfigured.

Compiled from published sources — Prusa's Knowledge Base, Bambu Lab's Wiki, Simplify3D's troubleshooting guide, the FixMyPrint dataset — plus clearly labelled community advice. Not personally tested on a printer in our workshop. Manufacturer figures are labelled; where no defensible universal number exists, the guide gives a test instead of a fake-precise setting.

Applies to: any FDM printer · Cartesian (bed-slinger) and CoreXY machines (tests flag where they differ) · PrusaSlicer, Bambu Studio, OrcaSlicer or Cura. Scope: the motion system — belts, pulleys, stepper motors, drivers — plus the two impostors that are not motion faults at all: collisions and parts that move on the bed. Enclosed high-speed machines add one suspect of their own (driver heat, Branch E); open-frame machines meet it less often.

In this guide: Step 0: Read the print · The 4-question triage · First: restore the baseline · Branch A: The belt drive slips · Branch B: The motors stall under demand · Branch C: The nozzle hits something · Branch D: Motor, cable and current · Branch E: The electronics drop out · Branch F: The part moved, not the head · What NOT to do · FAQ

Key Takeaways

• A layer shift is lost position, not bad plastic. The printer counts steps; many conventional machines never check where the head actually is — so a lost step is never noticed or corrected.
• Direction and pattern are evidence. Which axis shifted, whether it repeats, and at what height — three facts that sort most causes before a tool comes out.
• Test before you tighten. Each branch opens with a test; tests give evidence, not proof — each fix's recovery check is the confirmation.
• Never "correct" a shift with steps/mm. That rescales every dimension the printer makes and buries the real fault under a wrong calibration.
• One shift and a staircase are different animals. A single step points at an event; repeated small steps point at a drive that slips a little at a time.

Step 0: Read the Print (60 Seconds)

Look at the shape of the wrongness — and at what it is not doing:

What you see	What it most likely is	Where to start
One clean sideways step at a single height; layers above and below it neatly aligned with each other	Layer shifting — position was lost once, in one moment (Prusa's Knowledge Base and Simplify3D describe the same picture)	The 4 questions below
Repeated small steps in the same direction, layer after layer — a staircase	Still shifting, but a drive slipping gradually — Prusa names a loose or misaligned motor pulley as the usual main cause of staircase shifts	Branch A first
The whole part leans progressively, like a tower — no single step anywhere	Not lost steps: a lean accumulates gradually. Usually the part itself creeping or warping as it grows, or an axis out of square	Branch F's outline test; frame squareness is beyond this guide
Regular ripples or bulges that come and go up the wall (banding) — but the top of the part still sits centred over the base	Z-axis or extrusion variation rather than X/Y position loss — banding suggests the head kept its position, though appearance alone cannot prove positioning was accurate	If line thickness also varies, our under-extrusion guide
The part's base no longer sits on its own skirt/brim outline; the part looks rotated, or its bottom overhangs air	The part moved on the bed — the target slipped, which points away from the head; the two faults can coexist, so confirm with Branch F's tests	Branch F, then our first-layer guide
Only the first few layers look smeared or offset; the rest of the print is straight	A first-layer problem — an over-squished or rocking base, not a mid-print shift	Our first-layer guide

Note when it began — right after a speed increase, a belt adjustment, moving the printer, a new enclosure, or out of nowhere? A change you made is a suspect; no change points at wear, loosening fasteners and heat.

The 4-Question Triage

A triage sequence, not a decision tree: causes overlap, and one fault dresses as another (a collision knocks an axis out exactly like a stall does; a slipping pulley only shows itself under acceleration). The answers are priority rules — cheapest discriminating tests first (reading the shift's direction and pattern costs nothing and halves the suspect list) — and test results overrule the question order whenever they disagree. Throughout: same test print, same filament, same profile, one change at a time, each test before its fix.

Question 1: One clean step, or a staircase? A single step is one event — something happened once: a collision, a knock, one stall. Start with Branch C. Repeated small steps in the same direction mean the drive is slipping a little at a time — Branch A first; Prusa names the loose or misaligned motor pulley as the usual main cause of exactly this pattern.

Question 2: Which axis — and is it the same axis every time? Read the step's direction against the machine: on a Cartesian printer, a left–right step is the X drive line and a front–back step is the Y. On a CoreXY, keep two things separate: the direction of the printed offset, and the direction of abnormal movement when you move the head by hand. Prusa's guidance for its CORE One concerns the second: if the head moves inconsistently along a diagonal during a manual check, suspect the motor — or its pulley — that drives that diagonal (inconsistent X or Y movement usually relates to belt tension). That is a manual-diagnostic observation, not a universal rule for reading printed offsets on every CoreXY. Same axis, print after print: work that axis's drive line (Branches A, D, E). The axis changes between prints: the fault is load- or heat-shaped, not one belt — Branches B and E.

Question 3: Same height or same feature every time — or anywhere? A shift landing at the same height, or where the same curled corner or overhang begins, is geometry talking: something at that height catches — Branch C (or the part letting go at that height, Branch F). A shift landing anywhere, at a different height each print, is a capacity or connection problem — Branches A, B, D or E, separated by their tests.

Question 4: Does the part still sit exactly on its first-layer outline? Compare the finished part's footprint with the skirt, brim or purge lines laid down at the start. Moved or rotated: stop — the part is the prime suspect; go to Branch F before touching the belts (a head fault can coexist, but prove the part first). Still square on its outline: the head lost position, and Branches A–E are the field.

First: Restore the Baseline (Before Any Motion Change)

Before touching a belt, a screw or a current value, preserve the failing project and the exact settings that produced it — you will want them back for comparison — then restore a known-good manufacturer profile — stock printer profile plus matching stock filament profile — and verify what "stock" really covers here: print speed and travel speed at the maker's defaults; acceleration and jerk / junction deviation at the maker's defaults (a tuned profile often raises motion limits while leaving speeds looking innocent, and many slicers send their own acceleration values over the firmware's); and, where your printer offers a quiet or stealth mode, the maker's normal mode — Prusa's own layer-shifting guidance recommends normal rather than stealth mode for bigger or complex prints. Set the printer on a stable surface with nothing resting against it, in the same enclosure state as the failed print. Two paper checks before any test print: open the job in your slicer's layer preview at the shift height — an intentional toolpath or geometry change at that layer (a new feature starting, a support transition) can masquerade as a shift, and no belt adjustment fixes a slicer decision; and if the printer reported an error — motor, homing, temperature or motion — write the code down and consult your model's documentation before changing any hardware. Know the baseline's limits, too: a stock profile restores the slicer's numbers only — it does not undo modified firmware limits, motor current or machine calibration, so if those were ever changed, they stay changed until you restore them separately. Reprint the smallest model that has already shifted once — but only if it reproduces the relevant conditions: tall enough to reach the shift height, with comparable mass, duration and travel pattern; a shortcut model that never reaches the failing conditions proves nothing. Clean on stock? Your tuned profile — or its interaction with a marginal machine — is implicated; diff the two. And for an intermittent fault, treat one clean reprint as weak evidence: the branches below demand repeated passes before they call a fix verified. Still shifting? The branches start from honest ground.

Why It Happens (30 Seconds of Physics)

Most FDM printers are open-loop (Simplify3D's term): the firmware counts the steps it commands and assumes the head arrived. Many conventional machines never continuously verify actual X/Y position, so a lost step goes undetected and uncorrected, and every later layer is printed from the new, wrong origin (some machines do carry encoders or stall detection — if yours does, its error reports are evidence; see the baseline step). The branches below group the causes into three broad diagnostic families — a working map, not an exhaustive law: the drive slips (a belt jumps teeth or a pulley turns on its shaft — the motor counted correctly, the motion didn't happen), the motor stalls (load beats torque for a moment — steps were commanded and not taken; a collision usually acts through a stall, which is why Branches B and C overlap), or the target moves (the part shifts on the bed — not a loss of machine position at all, but it prints exactly like one). Electrical faults and homing or command errors can mimic any of the three; the branch tests are what separate them.

Branch A: The Belt Drive Slips — Loose Belt, Loose Pulley

Signs: staircase shifts, or shifts along one axis only; a belt that feels slack; circles printing as ovals and other quiet dimensional errors (Bambu Lab's Wiki ties ovals to a loose belt on its CoreXY machines); trouble that began after belt or pulley work, or crept in over months as fasteners relaxed.

Safety gate — read before any belt work: power the printer off before hands go near belts, pulleys or motor shafts. Belts are adjusted, never forced: an over-tightened belt loads bearings and shafts and can itself stall a motor (Branch B). Where your maker publishes a tensioning procedure for your model, follow theirs; it outranks everything below.

The first test, where the machine allows it — the pliers test (Prusa's own): this is a Prusa diagnostic for machines whose motor shaft is safely accessible — the MK series is the model case. With the printer off, hold the suspect axis's motor shaft gently with pliers, gripping on the shaft's flat, and try to move the carriage or bed by hand. Prusa's standard: there should be no slack and no axis movement. Any movement says the belt-and-pulley pair is not transmitting motion faithfully. On machines with enclosed motors or inaccessible shafts — many CoreXY designs among them — skip it and start from the pulley and tension checks below; it is not a universal first test. What the test does not do is separate the two suspects — a loose belt and a slipping pulley feel identical through it — and a pass does not establish that the belt holds correct tension under dynamic printing load. That is why the next checks exist.

The second check — the pulley itself: check the motor pulley's two grub (set) screws: both tight, and one of them tightened against the flat of the motor shaft (Prusa's assembly rule for its machines). A screw tightened onto the round of the shaft can hold for weeks, then let the pulley rotate a few degrees under a hard move — the classic staircase generator. Also check the pulley is aligned with the idler at the other end, so the belt runs straight instead of wandering as the pulley turns, and that the motor itself is tight in its mount. (Community practice, labelled as such: a fine marker line across pulley and shaft end makes a later slip visible at a glance.)

The third check — tension, by your maker's method only: there is no universal belt-tension number — belt length, width and the mass being moved all change what "right" is, so this guide states no figure as a standard. Where your maker provides an instrument, use it — but read it as a diagnostic indicator, not a measurement of belt tension: Prusa's MK3/S/+ Belt Status readout (LCD → Support → Belt Status) carries Prusa's own guidance of around 250 on X and 275 on Y, within about ±15 — an arbitrary unit inversely proportional to motor load, for that model line only; Prusa has owners loosen a belt reading near or under 240 and tighten one near or over 290. Prusa itself notes the readings shift with the printer's assembly condition, lubrication and bearing state, so before you interpret an out-of-range number, clear the confounders first: pulley secured (check 2), axes moving without stiff or grating spots, rods and bearings lubricated per the manual. Only then does the number say anything about the belt. Prusa also documents a belt-tuner method — tension judged by the belt's vibration frequency — for supported models; your model's current documentation decides which method applies. Prusa models without the readout get Prusa's field description instead: a belt sounding like a low bass string when plucked, its two sides pinching together with a little resistance. Bambu Lab X1/P1 machines set tension with a spring, not a number — fix 2 below gives that procedure. Everywhere else: the pliers test plus your manual.

Fixes, in order:

1. Secure the pulley first. It takes a minute and is Prusa's named main suspect: both grub screws tight, one squarely on the shaft flat, pulley aligned so the belt runs straight. Check: pliers test — no movement; a hand-run of the axis through full travel feels even, with no tight or grating spot.
2. Tension the belt by your maker's method, in small steps. On screw-tensioner machines, a turn or two at a time, re-checking as you go and again after locking (locking can move the setting); on CoreXY machines adjust both sides evenly and keep the gantry square — Prusa warns against losing CoreXY alignment while adjusting. For Bambu Lab X1/P1 (the maker's own procedure): power off; loosen the four belt-tensioner screws without removing them; move the toolhead back and forth several times along its X and Y travel paths, then to the rear, so both belts even out; retighten the four screws — the spring sets the tension. Bambu's own warnings: do not over-torque those screws (it can cause damage and will not improve the result), and run the printer's calibration afterwards — the vibration-compensation (resonance) calibration in the model's menu, because tensioning changes the toolhead's resonance. Check: your maker's readout or field test sits in range, and the test print passes its previous shift height with steps/mm untouched.
3. Inspect the belt itself. Tension cannot repair damage: frayed edges (a belt rubbing a pulley flange), cracked or missing teeth, or a belt that will not hold a setting across two prints is a replacement — same tooth profile and width as the original, from your maker's parts list — after fixing whatever made it fray. Check: the new belt holds its tension and alignment across two consecutive test prints.

If the pliers test is clean, the pulley is secured on the flat, and tension sits at your maker's mark — yet the axis still shifts — the slip is probably not in this belt at all. Move on; do not keep tightening (see What NOT to Do).

Branch B: The Motors Stall Under Demand — Speed, Acceleration, Jerk

Signs: shifts that began right after a speed increase or a switch to a faster profile or mode; shifts on the fastest moves (long travel moves, dense infill direction changes); sometimes a click or clunk at the moment of the shift — Simplify3D describes the motor audibly failing to reach its commanded position when asked to move faster than it can. The belts pass Branch A's tests.

The first test — the half-speed test (Simplify3D's own): on the same model and profile, reduce both the default printing speed and the X/Y travel (movement) speed by 50%, and reprint. Travel matters as much as printing: the fastest moves a printer makes extrude nothing, and a step lost on a travel move offsets everything printed afterwards. If the shift disappears at half speed, demand was at least part of the cause — evidence, not proof: slowing down also softens collisions (Branch C) and reduces vibration, so treat a clean half-speed print as a direction, then separate the suspects with the next test.

The second test — the acceleration step-down: restore full speeds, and instead reduce acceleration (and jerk, or junction deviation, where your firmware uses it) by roughly a quarter to a third — that step size is this guide's own diagnostic suggestion, not a manufacturer threshold — in the slicer profile and with the firmware's own caps in mind, since many slicers send their own acceleration values during the print. Torque demand peaks while changing speed and direction, not while cruising; if shifts vanish at full speed with gentler acceleration, direction-change torque was at least part of the limit — lowering acceleration reduced the demand, and speed and acceleration interact, so speed is not thereby cleared for every geometry.

Fixes, in order:

1. Set permanent limits where the tests pass — with headroom. Raise speed (or acceleration) back in steps from the passing values until the shift returns, then back off one step: that backed-off setting, not the failing edge, is your machine's honest limit for this class of model. Check: the test print completes twice at the final settings with no offset at the previous shift height.
2. Keep the printer in its full-power mode for demanding prints. Where a maker offers normal and stealth/quiet modes, Prusa's layer-shifting guidance is explicit for its machines: run normal mode rather than stealth for bigger or complex prints (on Prusa's MK series, crash detection is also unavailable in stealth mode). Check: the previously failing large print completes in normal mode at baseline speeds.
3. Reduce the load, not just the numbers. Clean and lubricate rods and bearings per your maker's instructions (Prusa names scratched rods and dry bearings as movement faults, and specifies its own lubricant or a lithium-based general-purpose grease for its machines); on Bambu Lab machines keep the carbon rods clean per Bambu's maintenance procedure — Bambu pairs rod cleaning with belt tensioning for exactly this reason — and never lubricate those carbon rods with oil or grease; they are cleaned, not oiled. Check: axes moved by hand, motors off, travel their full range with even resistance and no stiff patch.

A stall that survives half speed and a one-third acceleration cut at baseline has not been cleared of demand — it may still involve torque margin, binding, or demand that only appears intermittently — but the balance of suspicion shifts to the motor's supply and its drive line. Branches D and E take it from there; do not simply keep slowing down (see What NOT to Do).

Branch C: The Nozzle Hits Something — Collisions and Obstructions

Signs: one shift (or a few), at a specific height or feature; a curled corner, a lifted edge or an overhang at about that height; trouble that arrived with a new model rather than a settings change; sometimes a clunk heard from the other room. The shift is an event, and events have timestamps — this branch reads them.

Safety gate: investigate collisions cold and powered off wherever hands are involved. A nozzle at printing temperature burns on contact, and reaching into a running machine to clear a catch is how a shift becomes an injury. Watch a reprint from outside the machine, or use its camera if it has one.

The first test — the height match: measure the shift's height on the failed print and ask what exists at exactly that height. Prusa's Knowledge Base names the geometry: overhangs can warp upwards as they print and the nozzle can crash into them, and very low infill percentages can leave surfaces that curl into the nozzle's path. A curled corner, a warped edge or a blob standing proud at the shift height is consistent evidence — not proof on its own, but it tells you where to point the next test.

The second test — the hand-travel sweep: motors off (or disabled from the menu), move the head and bed slowly through their full travel by hand, in both axes and — on CoreXY — along the diagonals, feeling for any catch, notch or tight spot. Prusa's named finds: a piece of filament debris stuck around a pulley (usually the Y pulley) from earlier prints; a cable bundle or zip tie positioned so it meets the frame before the axis reaches its end, which also corrupts the printer's sense of the end position. A catch that repeats at the same axis position is the suspect — clear it and repeat the sweep before printing anything.

The third test — watch the moment: reprint and be present (or filming) at the previously failing height. A nozzle strike you can see or hear at the matching layer separates a collision from a stall more cleanly than any inference.

Fixes, in order:

1. Fix the interference at its own source — the three cases are different faults. Overhang curl: work the cooling side — print-fan speed (Prusa's own first suggestion), nozzle temperature, layer time, supports, and geometry that forces extreme overhangs; a curl is a cooling-and-shape problem, and no extrusion setting flattens it. Nozzle contact with infill: look at the infill pattern itself, over-extrusion piling up at line intersections, and how the head travels across the surface — the raised crossing points are what the nozzle is striking. A lifted base or corner: that is adhesion and warping — our first-layer guide covers keeping a base down; a corner that stays flat cannot be hit. Check: on the reprint, the feature at the old shift height lies flat as the nozzle passes it, and the print finishes unshifted.
2. Clear the path and dress the cables. Remove debris from pulleys and belt runs; re-route the cable bundle and zip ties to your maker's assembly-guide positions so nothing meets the frame early. Check: the hand-travel sweep runs the full range in both axes with no catch, twice.
3. Z-hop as a test, not a cure. A small Z lift on travel moves can step over a marginal strike — but if the print only survives with lift enabled, the curl or obstruction is still there waiting for a taller print. Treat a Z-hop pass as evidence consistent with nozzle interference — not proof of which specific cause in fix 1 or 2 produced it, and not the repair itself. Check: the print passes the old height with lift disabled after the underlying fix.

If the height match finds flat geometry, the sweep finds a clear path, and the watched reprint still steps sideways at that layer — the event is not a collision. Take the same-height evidence to Branch F (a part letting go can also keep an appointment) and the anywhere-evidence to Branches B, D and E.

Branch D: Motor, Cable and Current — the Supply Side of One Axis

Signs: shifts confined to one axis whose belt, pulley and tension all pass Branch A; stalls at gentle speeds that Branch B's tests cannot explain; trouble that began after a board, motor or cable change, or after the printer was moved; on a bed-slinger, a Y axis that misbehaves at one end of its travel, where its cable bends hardest.

Safety gate — read before any electrical check: connectors are reseated only with the printer powered off and unplugged — never plug or unplug a stepper connector under power; it can destroy that axis's driver. Current is adjusted only to your maker's specified value for your board and motor: this guide publishes no universal current figure, because drivers and motors differ, and an over-current "fix" cooks motors and drivers — it manufactures Branch E's fault. Opening electrical enclosures and adjusting board potentiometers (trimmer pots) are outside this guide's scope entirely: if your machine's current can only be set or measured that way, that is maker's-manual or support territory, not a guessing game.

The first test — the full-travel move: from the printer's menu, run the suspect axis slowly through its entire travel, back and forth, several times, watching and listening. A stall, hesitation or roughness that repeats at the same axis position — where a cable flexes most, or where a cable chain pinches — points at the connection, not the settings. Smooth motion over the whole range makes a position-dependent cable fault less likely; it does not clear an intermittent one (connectors can fail only under print vibration) — the checks below work through the connection, the cable and the current in that order.

The second check — inspect and reseat the connection: power off and unplugged, unseat the suspect axis's motor connector at the board and at the motor, and inspect both ends — pins straight and fully seated, no discolouration, no chafed or pinched insulation along the cable's flex path — then reseat both firmly. A connector lifted by a hair or a conductor fatigued inside its insulation presents exactly like this branch. What this guide deliberately does not offer is a connector-swap test: plugging one axis's motor into another axis's output to see whether the fault follows is a service-bench diagnostic, legitimate only where the manufacturer's service documentation explicitly authorizes it for your exact model — identical-looking connectors do not establish that two outputs are electrically interchangeable. Where your maker's documentation provides its own motor or driver diagnostics, use those; otherwise, the remaining checks still separate most cases.

The third check — the current value itself: where your firmware reports or sets driver current (many Marlin and Klipper machines do), compare the suspect axis's run current against your maker's specification — but only where the maker's documentation uses the same measurement your firmware displays: current units, RMS versus peak conventions and driver configuration differ between boards, and numbers measured differently are not comparable. Simplify3D lists insufficient current among the electrical causes of lost position: a motor under-supplied simply has less torque to give. The realistic stories are a value lowered by a "quieting" tweak, or a default that arrived with a board swap — not a number that drifted on its own.

Fixes, in order:

1. Reseat, then replace, the suspect cable. A connector lifted by a hair or a conductor fatigued inside its insulation both present exactly like this. Reseat firmly first; if the full-travel stall survives reseating, the cable is a wear part — replace it with the maker's part. Check: ten full-travel passes with no stall at the old position, then the test print unshifted.
2. Restore the maker's current — to spec, not above it. If the value was below specification, return it to the specified figure exactly, and save it the way your firmware requires. Check: the test print completes at baseline speeds, and an hour into printing the firmware reports no temperature warnings and the electronics fan runs as it should. Judge motor and driver heat only that way — never by touch: a healthy stepper can run uncomfortably hot without being faulty, and driver heatsinks may be electrically live. If you need an actual number, use an appropriate instrument — an infrared thermometer read from outside the enclosure — without opening an energized enclosure or contacting moving parts; sustained heat questions belong to Branch E.
3. If the motor itself is indicted — the cable replaced or proven, the current at spec, the full-travel stall unchanged, and any maker-authorized diagnostic pointing at the motor — replacement is a parts job done by your maker's manual. Check: full-travel moves and the test print both clean on the new motor, with Branch A's pulley checks repeated on its shaft.

If the connection, cable and current all come back clean and one axis still loses position at gentle demand, the remaining suspect is its driver — Branch E's heat story when the fault is time-shaped, and your maker's support (with any service-documentation diagnostics it authorizes) when it is not.

Branch E: The Electronics Drop Out — Driver Heat and Cooling

Signs: shifts that arrive deep into long prints and never in the first hour; the same file succeeding as a short print, or failing only with the enclosure closed or the room warm; shift heights that wander between prints (time-shaped, not geometry-shaped). Simplify3D describes the mechanism plainly: motor-driver electronics that overheat can stop the motors temporarily, until the electronics cool down — and a motor stopped for a moment mid-print is a layer shift with no mechanical culprit.

The first test — the timing comparison: line up your failed prints. If the shifts happen at different heights but similar elapsed times — and short prints of the same file's opening hours never shift — the pattern is consistent with a heat-dependent fault. It is evidence, not proof: long prints also contain more travel moves and more chances for every other branch, which is why this branch's second test changes the heat and nothing else.

The second test — the cool-down A/B: reprint the previously failing job with electronics cooling improved and nothing else changed: verify the electronics fan actually runs when the drivers work, clear its intake and the board's vents of dust and filament scraps, and — where your maker allows it — print with the enclosure door open so the board breathes room air. Late shifts that vanish under better cooling support heat as the cause; late shifts that persist at the same elapsed time send you back to Branch D with the cooling ruled down, not out.

The third check — the heat evidence, gathered without touching anything: never judge temperature by laying a hand on a driver heatsink or a motor. A healthy stepper can run hot enough to be uncomfortable without being faulty, and driver heatsinks may be electrically live, so touch proves nothing and risks a shock or a burn. Judge heat from the outside instead: an electronics fan silent when it should be spinning, vents exhausting noticeably hot air, and above all the records — firmware that monitors its drivers may have logged an over-temperature warning, and maker-approved diagnostic readings (or an infrared thermometer used from outside the enclosure, never by opening an energized one) give you numbers without contact. Their job is to corroborate the timing evidence.

Fixes, in order:

1. Restore the cooling path. A working electronics fan, clear vents, and a board not buried under the printer's own clutter; a fan that does not run is replaced by your maker's procedure. Check: the previously failing long print completes, with the fan audibly running during it.
2. Undo any over-current. If Branch D's check found current above your maker's specification — a common inheritance from a previous owner's "more torque" fix — return it to spec. More current can significantly raise resistive heating in drivers and motors, though how large the effect is depends on the operating conditions. Check: the long print completes with no firmware temperature warnings logged and the electronics fan running throughout — and, where you took instrument readings, they are no worse at the end than at the one-hour mark.
3. Manage chamber heat for long prints. For materials that do not need a hot chamber, a closed enclosure in a warm room cooks the electronics along with the part; follow your maker's guidance for door and vent use on long prints. Check: two consecutive long prints complete unshifted in the configuration you will actually use.

If cooling is verified, current is at spec, and drop-outs still arrive on a timer — the driver or board remains the prime suspect, though not a confirmed diagnosis, and that is a support conversation. Bring the timing evidence: model, axis, elapsed times, and what the cooling A/B showed.

Branch F: The Part Moved, Not the Head — the Adhesion Impostor

Signs: a "shift" whose lower layers are perfect and whose part, on inspection, no longer sits where it started — rotated a few degrees, slid a few millimetres, or rocking on a curled base; tall, thin or top-heavy parts are the usual victims; sometimes the part detaches completely and the printer finishes the job printing into air, which is this same fault with the volume turned up.

The first test — the outline test (Question 4, up close): before removing the failed part, compare its footprint with the outline the printer laid down at the start — but know which reference you actually have. A brim is printed attached to the part's own footprint, so it documents where the part began. A skirt sits at a configurable gap from the part, so it only helps if you know the offset your slicer set. A purge line may be placed anywhere on the plate and proves nothing about the part's position. Against a known original footprint — a brim, or the first-layer position your slicer preview documents — a part sitting offset or rotated moved: that supports the adhesion diagnosis and points away from the head (the two faults can coexist, but a moved part is established first, in ten seconds, with no tools). Every belt you tighten for a moved part makes a healthy machine worse.

The second test — the rock test: with the part cooled and still on the bed, press a top corner gently, side to side. A part that rocks or twists on its base is being held by a fraction of its footprint — free to be pushed by the nozzle's passing, layer by layer. Rocking demonstrates poor support and adhesion; on its own it does not prove when the part moved — that timing comes from the outline test above. Combine it with the base inspection: corners lifted off the plate (warping) mean the grip was failing before the shift height arrived.

Fixes, in order:

1. Restore the grip. Clean plate, correct first layer, corners that stay down — that is a whole discipline of its own, and our first-layer guide teaches it end to end. Check: the reprint's corners are flat against the plate at the old shift height, not just at layer three.
2. Give tall and top-heavy parts a wider stance. A brim sized to the part's leverage, an orientation that puts the broad face down, or supports that brace the part against travel-direction pushes — slicer-side answers to a mechanical problem. Check: the finished part still sits exactly on its original footprint — brim or documented first-layer outline — unrotated, when the print ends.
3. Slow the forces that push. If a well-gripped part still creeps on fast direction changes high up, Branch B's acceleration step-down reduces the push without touching adhesion — the two branches share this one fix honestly. Check: the rock test on the reprint finds a part that does not move.

If the part never moved — outline test clean, rock test solid — this branch is closed and the shift is real: return to the triage answers and work the head-side branches with the part's innocence established.

About the Community Numbers in This Guide

The FixMyPrint dataset behind this guide is one analyst's scrape of 20,000+ help posts and roughly 60,000 fix suggestions from one of Reddit's largest 3D-printing help communities, collected over one year and classified by problem type, printer model, and whether the asker confirmed a fix. Layer shifting accounts for 1,138 of those posts (#6 overall). Read that number for exactly what it is: a count of community help posts in that analysis — not a measurement of how often layer shifting occurs across all printers. Our research notes for this guide record that post count but not a confirmed-fix count for this category, so — exactly as with the cold-pull figure in our under-extrusion guide — this guide quotes no confirmation percentage for layer shifting rather than printing a number it cannot stand behind. What the count itself says is enough: in that dataset this is one of the most reported ways a print fails, and it fails at the hands of a short, testable list of causes.

What NOT to Do

• Do not change X/Y steps/mm to "correct" a shift. Steps/mm calibrates distance on a healthy machine; a shift is position lost mid-print. Rescaling makes every dimension the printer produces wrong and buries the slipping or stalling part under a false calibration.
• Do not crank belts as tight as they will go. Overtight belts load bearings and motor shafts, add friction that can itself stall a motor, and — on Bambu Lab's spring tensioners — over-torqued screws can cause damage and, in Bambu's own words, will not provide better results. The maker's mark is the target, not the maximum.
• Do not raise motor current "for more torque" beyond your maker's specification. Over-current overheats motors and drivers and manufactures Branch E's drop-outs. Current is restored to spec, never improvised upward.
• Do not slow the whole printer to a crawl and call it fixed. The half-speed test is a diagnostic. A permanent crawl hides a mechanical fault that keeps wearing — and returns on the tall print, at hour nine, when it costs the most.
• Do not touch, hold or bump the machine mid-print. On an open-loop printer, a knock is a layer shift — Simplify3D's own example is bumping the printer and the head simply continuing from its new position, unaware. And never reach into a moving machine to clear a catch (Branch C's gate).
• Do not change belt tension, speed and current in the same print. If it improves, you own three unverified changes; if it worsens, you cannot retreat. One change, one test, one verdict.
• Do not keep reprinting the full model to test a fix. Print the smallest piece that has already shifted once, at the settings that shifted it — a fix is only verified against the failure it claims to cure.

When to Stop and Ask for Help

If the pliers test is clean, the pulleys are secured on their shaft flats, tension sits at your maker's mark, the baseline profile at reduced demand still shifts, the cooling A/B changed nothing, and the outline test shows the part never moved — stop adjusting, and gather the evidence you now own: which axis, one step or staircase, same height or same elapsed time, mode and enclosure state, and what each branch's test showed. A support ticket carrying that record gets a different conversation than "it layer shifts". What lies beyond this guide's reach — a failing driver or board, a bent frame or racked gantry, a motor with an internal fault — belongs to the maker's manual and support, and your branch notes are exactly what they will ask for first.

Stop-work conditions — stop experimenting at once, whatever branch you are in: exposed wiring, discoloured or heat-marked connectors, a burning smell, visible mechanical damage, or electrical faults that keep repeating. Power the printer off and contact support. In particular, never answer repeated electrical faults by experimenting with motor current.

FAQ

Q: My print leans like a tower but has no step anywhere. Same fault?
A: Almost never. A shift is a single moment of lost position and leaves a visible step; a lean accumulates gradually. Start with Branch F's outline and rock tests — a part creeping on a failing grip leans beautifully — and if the part never moved, suspect an axis out of square, which is frame territory beyond this guide.

Q: It shifted once, eight hours into a ten-hour print, and has never done it again. Chase it?
A: Lightly. One shift is one event: run Branch C's height match (what was at that layer?) and look for obvious collision evidence — a curled feature at that height, debris, a strike mark — glance at the pulley debris check, then log the height and axis and print on. Investigate further only if it repeats. A one-off shift has more possible parents than a collision or a knock: a marginal pulley letting go once, an intermittent cable, a single stall or an obstruction can each do it exactly once.

Q: My circles print as ovals. Is that layer shifting?
A: A cousin, not the same fault: nothing is lost mid-print, but Bambu Lab's Wiki ties ovals and dimensional errors to a loose belt on its machines, and the remedy starts in the same place — Branch A's pliers test and tension check. Treat ovality as the belt drive asking for attention before it graduates to steps.

Q: Would a camera or timelapse actually help?
A: Yes — as evidence, which is this guide's whole method (labelled community practice). A frame from the shift moment separates a nozzle strike from a silent stall, and a timestamp separates a height-shaped fault from a heat-shaped one. It is the cheapest branch test you can run on a fault that only appears when you are not watching.

Q: My printer has crash detection / step-loss recovery. Is this guide still relevant?
A: Crash detection and recovery may save some prints — they cannot guarantee it, and they do not detect every lost-position event — and even a saved print does not name the cause, which is still in the machine. Note the limits, too: on Prusa's MK series, crash detection is not available in stealth mode — one more reason Prusa's guidance puts demanding prints in normal mode. When recovery starts triggering regularly, that frequency is the symptom — work the branches.

Summary: The Six Branches at a Glance

Diagnose before you tighten: read the print (Step 0), run the four triage questions, restore the stock-profile baseline — then work the branch the evidence points to:

Branch	Lead signal	Confirming test	First fix
A — Belt drive slips	Staircase steps; one axis; slack belt; ovals	Pliers test on the motor shaft, where safely accessible (slack = evidence); grub screws on the flat	Secure the pulley; tension by your maker's method
B — Motors stall under demand	Shifts after a speed-up; clicks on fast moves	Half-speed test; acceleration step-down at full speed	Permanent limits with headroom; normal mode for big prints
C — Nozzle hits something	One step at the same height or feature	Height match; hand-travel sweep for catches	Fix the curl; clear the path; cables per assembly guide
D — Motor, cable, current	One axis; belts pass; stalls at gentle speed	Full-travel move; connection inspection and reseat; current vs maker's spec	Reseat/replace the cable; restore specified current
E — Electronics drop out	Shifts late in long prints; warm enclosure	Timing comparison; cool-down A/B with nothing else changed	Restore cooling; undo any over-current
F — The part moved	Part offset or rotated against its own outline	Outline test; rock test on the cooled part	Restore grip; brim/orientation; slower pushes

Every fix ends with its verification — if the named sign did not move, that was not your cause.

Honesty note: Compiled from manufacturer and community documentation — not personally tested on a printer in our workshop, and it does not pretend otherwise. Figures are labelled as such, and dataset figures carry their denominators where the research records them. Where the evidence ran out — a universal belt tension, a universal motor current — the guide says so instead of inventing a number.

Sources: FixMyPrint — dataset analysis (layer shift: 1,138 posts, #6 of 20,000+); Prusa Knowledge Base — Layer shifting; Simplify3D — Print Quality Troubleshooting Guide (Layer Shifting); Bambu Lab Wiki — X1/P1 Series belt tensioning procedure; community advice: Bambu Lab Community Forum — Layer shifting (X Series). All source pages opened and read on 2026-10-30.

Related guides: Stringing after changing PLA filament? Check these 5 things first · How to dry PLA filament: every manufacturer's numbers, compared · First Layer Won't Stick on Your Ender 3? 8 Things to Check Before Touching Settings · Why Is My Top Surface Rough and Puffy? Pillowing, Diagnosed in 3 Questions · 3D Print Blobs and Zits: Seam, Moisture, or Pressure? · Under-Extrusion: Gaps, Thin Walls and Weak Prints, Diagnosed in 4 Questions