Six real-world meshes from CAD assemblies, multi-body Bambu user files, AI-generated Hunyuan3D output, and a 147 MB presupported figurine. Each one was audited by PrusaSlicer's --info command (the same data that backs the “Fix through netfabb” UI) and by AdMeshio. PrusaSlicer reports manifold = yes for the first five and undercounts defects 460× on the heavy case. AdMeshio surfaces what's actually wrong.
PrusaSlicer's STL loader runs admesh auto-repair silently, so the --info dump always says “manifold”. The fix is real — gcode generation works — but the user never sees what was repaired or how. When a marketplace, print farm, or pipeline that doesn't run the same auto-fix receives the raw file, the silent issues become production failures. AdMeshio's audit is the explicit version of that conversation: every defect is named, severity is scored, and remediation is one CLI command away.
AI-gen blob with inverted normals + scale collapse · 66,252 facets
PrusaSlicer says
number_of_facets = 66252 manifold = yes number_of_parts = 9 volume = 0.054624
No mention of the inverted normals or that the volume reads 0.054 mm³ — the mesh shipped at 1/100 scale.
AdMeshio says
INVERTED_NORMALSsurface points inward — slicer treats the model as inside-outTOO_SMALLbbox under printable size — Hunyuan default scale needs 100× upliftBELOW_MIN_FEATURE_SIZEfeatures under nozzle resolution at this scaleTHIN_WALLwalls thinner than profile minimumBoolean leak after CAD export — non-manifold edges + self-intersections · 6,012 facets
PrusaSlicer says
number_of_facets = 6012 manifold = yes number_of_parts = 16 volume = 66588.59
admesh's silent auto-fix flips the manifold flag to yes. The 16 parts include components that the assembly export left non-manifold; you'd never know without opening every part.
AdMeshio says
NON_MANIFOLDedges with ≠2 incident faces — boolean leak in CAD exportSELF_INTERSECTIONStriangles cross each other — geometry is still corrupt after admeshPRESUPPORTED_MAIN_BODY_SISI inside the load-bearing body, not just the supportCOMPONENT_BREAKDOWNper-part GRI surfaces which of the 16 are badUNDER_TESSELLATEDlow facet count for the print scale — visible facet edges expected+ 9 morereadiness flags: bridge / support / brim / trapped volumeReal Bambu user 3MF — silent NM + SI inside the plate · 8,210 facets
PrusaSlicer says
number_of_facets = 8210 manifold = yes number_of_parts = 1 volume = 2002.71
3MF unwrapped to a single “part” once admesh merged the bodies. The original plate had multiple intersecting bodies; the user-as-author intent is lost.
AdMeshio says
SLICER_COMPATIBLEplate routes through Bambu/Orca/Prusa correctlyNON_MANIFOLDedges between intersecting bodies that admesh merged but didn't sealSELF_INTERSECTIONSoverlapping bodies create SI even after mergeSTRUCTURAL_NECKnarrow load-bearing cross-section — review before printingELEPHANT_FOOT_RISKfirst-layer over-extrusion will widen the baseSlicer-shipped supports too thin / wrong angle — broken before the print starts · 1,612 facets
PrusaSlicer says
number_of_facets = 1612 manifold = yes number_of_parts = 1 volume = 3991.67
Topology look-ahead can't see whether the bundled supports will actually hold. PrusaSlicer slices them as-is; failures show up at the printer, not in the audit.
AdMeshio says
SUPPORT_BRANCH_TOO_THINsupport branches under min printable thickness — will snap during peelSUPPORT_CONTACT_WEAKcontact points too small to anchor the modelSUPPORT_ORIENTATION_MISMATCHsupport angle wrong for the printer the file was sliced forSUPPORT_SPACING_TOO_WIDEgrid wider than the slicer's recommended for this resin / printerPRESUPPORTED_MAIN_BODY_SISI in the main body, not just the support cones+ 14 moretopology + readiness — full report has the restTopology is fine — but the shape itself won't print well · 666 facets
PrusaSlicer says
number_of_facets = 666 manifold = yes number_of_parts = 1 volume = 57443.50
A topology-only check is content. Nothing in --info says the part has a sealed cavity, an unsupported overhang, or a bridge that won't span. The slicer will happily generate gcode that fails halfway through.
AdMeshio says
TRAPPED_VOLUMEsealed cavity — resin or air gets stuck, print will warp or crackCRITICAL_OVERHANGoverhang past the slicer's bridge limit — needs supports or geometry changeEXCESSIVE_BRIDGElong unsupported span — first layer of bridge will sagSTRUCTURAL_NECKnarrow load path — will snap if the part takes any weightUNDER_TESSELLATED666 facets for this part size — visible facet edges on curved surfacesPresupported figurine where the slicer scratches the surface but misses 99.6% of the actual defects · 2,948,793 facets
PrusaSlicer says
number_of_facets = 2948793 manifold = no open_edges = 23 facets_reversed = 14 backwards_edges = 20
Here PrusaSlicer at least admits the mesh isn't manifold, but it reports only 57 surface issues total. Self-intersections aren't part of --info at all. On a 2.95-million-facet mesh that took 8 seconds just to load, that's the entire structural answer the slicer offers.
AdMeshio says
NON_MANIFOLD10,578 non-manifold edges — 460× what --info reportsSELF_INTERSECTIONS399,608 self-intersection pairs — slicer never checks for thisOPEN_BOUNDARYcrack-class boundary loops, not intentional drain holesPRESUPPORTED_MAIN_BODY_SImost SI sits in the main body, not at support contacts — the relaxed tolerance doesn't applyAudit completed28 seconds on the full 2.95M-facet mesh · no GUI freeze · 21 flags total
PrusaSlicer's --info tells you the file loaded. AdMeshio's audit tells you the file will print. They aren't competitors — they answer different questions. If your pipeline relies on the slicer's silent auto-repair to mask defects, every printer that ingests the raw file becomes a roulette wheel. AdMeshio replaces that wheel with a check list.
Drop your own file in the browser Quick Check — you get the same flag list against your printer profile, in <30 seconds, without uploading the file off your machine.
Methodology: each file was loaded once into PrusaSlicer 2.9.4 with --info (same admesh auto-fix path the “Fix through netfabb” UI runs), and once into AdMeshio's audit command against the bambu_x1c_pla_standard profile. Outputs above are verbatim. Corpus spans three categories: parametric CAD parts (STEP imports), community-shared Bambu 3MFs, and AI-generated meshes (Hunyuan3D output).