harmonic-analyzer (pedropaulovc) errors

23 recurring problems engineers report in harmonic-analyzer (pedropaulovc), ranked by impact.

#13 by errors trackedMCADVerification

Recurring harmonic-analyzer (pedropaulovc) errors, by impact

  1. 1A DFM review of release v48 found docs/machining-dfm.md says PM-30MV CNC for the repeat parts while inventory.toml lists the same mill as manual handwheels on all three axes. The guide gives the rocker arm as about 270 x 21 mm against 292 x 29 mm in the STL, a pivot shaft length that disagrees with the print, and 19 bushings that do not exist in the bundle.MCAD · Agent failure · 1 report74
  2. 2A drawing job crashed SolidWorks. The watchdog caught it and the lifecycle library relaunched it without raising. The new instance came up on a Graphics Issue modal the recovery code does not dismiss. Recovery reported success, the seat was not usable, and every COM task on the machine, part, assembly, verify, export, drawing, is blocked until someone clicks the dialog.MCAD · Agent failure · 1 report74
  3. 3An audit of 263 tight bands found 81 slack, worth about 15 hours per build, and 19 that feed no stack at all. Along the way it found two stacks checked separately against the same 0.254 mm budget that add up to about 0.270, and a spec that says plus/minus .13 mm while the drawing sheet prints plus/minus .05 mm.MCAD · Agent failure · 1 report74
  4. 4Since #2101 their drawing guideline says purchased-part supplier dimensions print as untoleranced REF. The shared supplier helpers still put the catalogue tolerance band on the sheets, so a dowel printed as (Ø.1250 +.0003/+.0001). A blind machinist review flagged it as over-specification. The fix removes the band fields from dowels, washers and shims across the fleet. The policy lived in a doc and the behaviour lived in code, and only a reviewer reading the printed sheet connected them.MCAD · Agent failure · 1 report74
  5. 5The base and top frame put matched diameter 1.004 sockets on an 8.820 in row spacing, while the inventory gives the mill 8.75 in of Y travel and 7.5 in of throat, so each face needs two clampings rotated 180 degrees on parts that must slide together by hand with .004 clearance. A sister issue, #2392, found assembly sheets calling for a 20.86 in reading, a digital angle gauge, a 5 to 6 in micrometer and feeler leaves, none of which are in inventory.MCAD · Agent failure · 1 report74
  6. 6The MATERIAL cell links the generic SolidWorks library material instead of the part's Material Specification. An audit of one build found 58 of 111 drawings, 77 of 147 pages, printing something other than the specification. Two print the wrong alloy outright: the cone gear prints Brass where the spec is C67500 manganese bronze, and the knife mount prints Plain Carbon Steel where the spec is O1 tool steel at 58-60 HRC. Another 18 never print their grade, and 29 purchased parts print a generic material.MCAD · Agent failure · 1 report74
  7. 7A balloon leader pointed at nothing because the anchor ray probe gave 138 hits on one build machine and 98 on another over the same 349 bodies.Verification · Agent failure · 1 report66
  8. 8A balloon-anchor ray probe returned 138 hits on one farm seat and 98 on another, same 349 bodies and 12 rays. A seat that misses occluders anchors a balloon on a covered point, so the sheet then changes from seat to seat with identical inputs.MCAD · Agent failure · 1 report66
  9. 9A CAD export check on harmonic-analyzer was red on main after a PR removed a schema check on purpose, rewrote one copy of the test, and missed the other. The test expected a manifest with units = inches to fail before the native CAD seat opened, and after the change that is no longer true by design. The fix states the contradiction and pins the preflight to the project's own checks.MCAD · Agent failure · 1 report66
  10. 10A SolidWorks drawing audit caught a BOM balloon leader running 4.56 mm through a note after re-keys moved the balloon. The fix reads back every balloon's rendered position and proves 2 mm clearance after the rebuild.MCAD · Agent failure · 1 report66
  11. 11Agent-produced drawings for a printer mechanism went through up to eight review rounds on hole callouts, tolerances and missing pin-hole callouts before a machinist rubric passed them. One review round errored on an API rate limit.MCAD · Agent failure · 1 report66
  12. 12An automated machinist review flagged a drawing forcing AISI 1018, a grade present only because a commit swapped the SolidWorks library name Plain Carbon Steel for a buyable name, and changing one material field re-keys the part, its drawing, its review check, and every parent assembly.Verification · Agent failure · 1 report66
  13. 13Centring one lever plate on its trunnion axis raised anchors and receiver rows by 0.96 mm, broke five stale numeric expectations, and forced a native spring recalibration because the balancing moment changed. A registry audit also flagged an untouched part as orphaned because the checker only reads literal name declarations.MCAD · Agent failure · 1 report66
  14. 14Generated supplier diagrams printed no part geometry, so their REF dimensions pointed at empty space, and one text overlap was an enforced finding that failed the inch build.MCAD · Agent failure · 1 report66
  15. 15In an AI drawing-review setup for SolidWorks sheets, the agent harness added an extra intent field to tool calls. The evidence validator required exactly one key, so all 32 real sheet views and crops did not count as inspection and reviewer evidence got thrown out over a metadata key.MCAD · Agent failure · 1 report66
  16. 16The CAD carried jam nuts that do not actually exist in the real device, with drawings, part numbers and checks built around them. Two part numbers are retired with their builders, specs, drawings, registries, tests and a McMaster catalogue entry, and three assembly checks built around the nuts had to be replaced.MCAD · Agent failure · 1 report66
  17. 17The generated assembly sheet tells the builder to face the guide when the lock gap reads under .002, aiming for .002 to .009, but from the drawings and the model JSON the as-made gap is .011 to .031, and facing can only shrink it the wrong way. The spec script, the drawing generator and the model each hold a different version of the same fit.MCAD · Agent failure · 1 report66
  18. 18The remote cache covers parts, assemblies and drawings but not the gates or the STEP/STL/glTF exports. On a fresh checkout about 100 parts and 5 of 7 assemblies restored in seconds, then the run spent the rest of its time in verify, serialized on the SolidWorks COM seat lock.MCAD · Agent failure · 1 report66
  19. 19The same drawing task fails on one seat and passes on another. The same task key failed on one SolidWorks seat with failed to select fulcrum shaft PMI edge at sheet (0.282, 0.205) and succeeded minutes later on a different seat. The code involved was byte-identical to the previous build.MCAD · Agent failure · 1 report66
  20. 20The verify script documents expected component counts in comments. One said 8 parts while the placed rig had 9, because the lever was missing from the list. The comment is the only documentation of what the number means, and the next editor reconciles against it. The proposed fix is to count placements from the build scripts.MCAD · Agent failure · 1 report66
  21. 21Two farm jobs waited 900 s for the SolidWorks connector, logged that they still were not connected, and built anyway. One held a seat 16.8 minutes for about 2 minutes of work. The other died 9 s in with failed to insert native hole callout, twice, and failed the build. The readiness check is documented as best effort, so failures are logged and swallowed.MCAD · Agent failure · 1 report66
  22. 22When a config change lowers the spring variant count, old SLDPRT files stay in the output folder. Nothing except doit clean deletes them, and the glob feeds the cache, so a stale part gets republished as current.MCAD · Agent failure · 1 report66
  23. 23An automated review run returned ERROR because the reviewer CLI hit its session limit, which had to be separated by hand from a real sheet defect.Verification · Agent failure · 1 report58

How it fails

  1. Agent failure23

Sourced from public community reports. Costs and delays on individual entries are modeled estimates, not measurements.