Why Does a Part Chatter on Some Passes But Not Others With Identical Parameters?

Because the parameters that stay identical in your program aren't the only variables in the system. Stickout, engagement, and the part's own stiffness all change pass to pass as material comes off, even though feed, speed, and depth of cut stay the same in the code. Chatter depends on the whole structural loop between spindle, tool, part, and fixture, and that loop is not static through a job.

What's actually changing between passes

The most common culprit is that the part gets less stiff as you remove material. A pocket wall that started at 10mm thick and finished at 3mm has a completely different natural frequency by the last finishing pass, even though the toolpath parameters never changed. Early passes on the same wall might be perfectly stable because there was still enough mass to damp the vibration; the last pass chatters because the wall is now thin enough to resonate.

Tool engagement angle changes this too, especially in pocketing or profiling where the path wraps around corners or transitions between full-width and side-milling cuts. A toolpath that holds constant feed and speed but varies radial engagement is not actually holding a constant cutting condition. A corner where the tool suddenly engages more of its diameter can excite chatter that a straight wall never triggers.

Stickout matters if the program uses different tools or the same tool at different reach for different features, but it also matters within one operation if the part itself deflects and the effective distance from a rigid support point changes as thinner sections appear.

Other things worth ruling out

• Fixture looseness building up. Repeated cutting forces can walk a part slightly in a fixture that seemed rigid at setup, especially with vise work or light clamping. Chatter that starts appearing partway through a job and gets progressively worse often points here rather than at tool wear.
• Tool wear accumulating. A tool that started sharp cuts cleanly; the same tool with a slightly worn edge two-thirds through a long roughing operation rubs more and can trigger vibration it didn't have at the start.
• Coolant or chip evacuation differences. A pass where chips pack in a pocket loads the tool differently than a pass where they clear freely, even at identical feed and speed.
• Harmonics with the spindle's own vibration modes at certain rpm bands, which can interact differently depending on how much the part itself is contributing to the total system stiffness at that moment.

What to do about it

Isolate whether the chatter tracks with wall thickness, engagement angle, or time in the job. That tells you which variable is actually driving it. If it's stock removal thinning the part, split the operation so the final light finishing pass on thin sections runs at a different, empirically found speed than the earlier roughing passes, rather than assuming one set of parameters should work for the whole feature. If it's a corner engagement spike, add a linking move or arc that smooths the transition instead of letting the tool suddenly bite more material. Chasing a single "fix" for chatter that only shows up intermittently usually fails until you've identified which part of the cutting loop is actually moving.