I tested this with my H2D previously. Using precision wall+arachne, without hole/contour compensation enabled. My machine has been calibrated with vision encoder. I tested this previously on my H2D to better understand where dimensional accuracy limits in FDM printing actually come from, using precision wall with the Arachne wall generator enabled and with hole and contour compensation completely disabled.  Positional accuracy and repeatability:The machine itself was calibrated using a vision encoder, so positional accuracy and repeatability of the motion system should not be a limiting factor.  Filament:I’m printing in Bambu PLA Basic, and since no hole/contour compensation is enabled, the slicer is not applying any intentional dimensional offset. I intentionally didn't calibrate the filament because in most common cases, it's not the main factor of tolerance issues.Testing proper method:I printed test parts with nominal hole diameters of 2 mm, 3 mm, and 4 mm, and measured them using a Mitutoyo caliper.  The measured diameters were:1.85 mm for the 2 mm hole, 2.88–2.90 mm approx. for the 3 mm hole, 3.93 mm for the 4 mm hole.  Does this make sense? I think it does. A systematic dimensional error on the order of ~0.1 mm for internal circular features is well within expectations for FDM and already below one quarter of the configured extrusion line width. This error is not caused by poor calibration or lack of machine precision, but by the effective geometric resolution of the extrusion process and how slicers approximate curved features using discrete toolpaths. However, this tiny error in most cases shouldn't cause tolerance issues. Testing problematci / classic method:To reproduce the scenario that users most commonly report as “inaccurate holes,” I repeated the same test using the Classic wall generator, again without precision wall and without any compensation enabled. In this case, the measured diameters were:1.6 mm for the 2 mm hole, 2.67 mm for the 3 mm hole, 3.7 mm for the 4 mm hole.  The resulting 0.3–0.4 mm diameter deviation is again expected and corresponds closely to one to two times the half-line width being effectively consumed by inward perimeter placement.  Conclusion: This comparison clearly shows that slicing strategy dominates dimensional accuracy for internal features, while machine precision mainly determines consistency and repeatability. You simply cannot expect a 0.4 mm nozzle to deliver sub-0.1 mm dimensional accuracy or tolerance in an FDM process, because that exceeds the physical resolution of material extrusion. Circular holes merely expose this limitation more clearly than most other geometries, which is why they are often used as a proxy for accuracy even though they are among the most demanding features to print accurately.

This article aims to fill out the gaps that Bambu Wiki left.https://wiki.bambulab.com/en/software/bambu-studio/fill-patterns Start from settings:Infill & Bridge direction: You might want to check this to make sure your infill is giving the top layer proper support. Infill combination:Pros: Much faster printing.Suitable for patterns like cross hatch, gyroid. Cons: less reliable sparse infill for patterns like Rectilinear, Zig Zag, Cross Zag and Locked Zag. These patterns might sag too much with the combined thick lines on top surface.The top layer of sparse infill might not be supporting the top surface well enough. Sparse infill patterns:Instead of throwing you a lot of random infill patterns, I'm going to categorize them into 3 types: Universal infills, Specialised Infills, Use With Caution Infills.Universal infills: you can use these patterns on essentially any model with any layer height. There might be better choices, but these are good starting points.Cubic: Pros: a high infill rate guarantees great surface quality. prints fast. quite strong. normalised layer print time difference.Cons: self crossing might cause nozzle bumps. straight printing line might cause shrinkage layer lines. Specials: creates chambers inside the part. air and water insulation.Gyroid: Pros: Everybody's favourite. Filament efficient for supporting (good top surface quality under low infill rate). Strong. Would be more resistant on shrinkage. Cons: inconsistent layer printing time, causing layer lines sometimes. Prints slow and shakes the printer.3D Honeycomb:Pros: Very similar to gyroid but it's straighter, so less shakey.Cons: Slightly slower than gyroid.Cross Hatch:Pros: Honestly, I don't know. Maybe it's less shakey.Cons: worse support performance compared to Gyroid and 3D Honeycomb. Straight line layers causes significant layer printing time changes. Not recommended. Specialised Infills: you only use it when you know what you're doing. Usually just skip these options for most models.ConcentricHoneycombHilbert CurveArchimedean ChordsOctagram SpiralLightningTrianglesTrihexagon Except from Lightning, all the patterns above are 2D patterns but multiplied with Z height. You only choose them when you need that 2D pattern.  Lightning on the otherhand would be quite useful for paintings or translucent prints. Or very fast drafts. Use with Caution: there're many catches, use with caution.Line:Pros: prints fastCons: everything elseRectilinear:Pros: efficient, good top surface quality with low infill rate. prints fast. Quite strong if printed well.Cons: sensitive to low layer height, filament fluidity and infill combination. You need to test print a sample, with the layer height you're going to print, with the filament you're going to use. If it doesn't bridge successfully, better switch to cubic or gyroid.Zig Zag, Cross Zag:Pros: very efficient, Zig Zag is Rectilinear but it allows line rotation (instead of a grid like structure). Might be useful for some prints. Quite strong if printed correctly.Cross Zag has a very good surface quality, if it prints, then it's one of the most filament efficient sparse infill options for artworks or building models.Cons: same as Rectilinear. You need to test print if you're not printing with default layer height, or enabled infill combination, or using filaments other than basic PLA. Locked Zag: see bambu wiki for pros, it's quite a innovative design. it's cons are the same as above. but it also prints slower due to the shell parts. Adaptive Cubic: Not recommended unless you're printing very basic geometry. The generated sparse infill cubics could be supporting very poorly. Support Cubic: Not recommended either.