NewsBB-0153

100-part test: Claude-generated KiCad symbols show fewer pin errors than the stock library

Dave Vandenbout ran his datasheet-to-symbol-lib Claude skill on 100 parts with 20 or more pins and diffed the results against KiCad's stock symbols. With every difference checked against the datasheet, generated symbols had a 2.77% pin error rate and the library 21.92%. The README notes that the same agent generated the symbols and judged both sides.

Open original source github.com · Dave Vandenbout

Claimed
Why this evidence label?

Published per-part results, generated symbols and diffs; design data only, no hardware involved.

How we check entries
Preview image for 100-part test: Claude-generated KiCad symbols show fewer pin errors than the stock library
The source's own preview image, from github.com.

Briefing

The sample covers 4,986 pins across 77 KiCad libraries, with pin counts from 20 to 400. Each symbol was generated from the datasheet alone and diffed with cmpparts, and every difference was attributed to one side, the other or neither. Per-part verdicts are published as JSON lines.

Most of the library's findings are missing alternate pin functions, which is incomplete metadata rather than a wrong netlist, and naming differences were counted separately. Without both, the generated symbols had 49 findings across 14 parts against 165 across 57 for the library. The README adds that the runs were not independent: conventions set on early parts shaped later ones.

Nothing here was built; it is design data only.

From the GitHub API

Repository facts

Repository facts for devbisme/100_part_test, from the GitHub API
Repositorydevbisme/100_part_test
Stars1
LanguagePython
Last push6 Aug 2026 2 mo ago

Tags

Agent

Report a problem with this entry