Fixture: zhiayang-mikoto
An nRF52840 module in a Pro Micro footprint: SoC, 2.4 GHz meander antenna and matching network, USB-C with CC resistors, a lithium charger with battery-management I2C, and regulated rails.
Attribution
Upstream: zhiayang/mikoto at commit 86318f6c, retrieved 2026-07-29.
Authored by zhiayang, licensed Apache-2.0. The upstream license text is preserved verbatim in LICENSE. Note that upstream declares Apache-2.0 in its README but states no explicit copyright line and leaves the license appendix’s Copyright [yyyy] [name of copyright owner] unfilled, so the attribution recorded here is to the repository author rather than to a stated copyright holder.
Design files are copied unmodified from the upstream repository root, together with the project sym-lib-table and fp-lib-table and the two libraries they name: lib/mikoto.kicad_sym and footprints.pretty. empty.kicad_wks is included because the project file names it as the page layout.
Two things were deliberately not copied. footprints.pretty/3d/ holds 388 KB of STEP models, which are rendering assets: they do not participate in ERC, DRC, or parity, and they would be copied into every sandbox for every repeat of every task. The separate flashbed/ board is a different design and would be a different fixture. misc/, pins.txt, CHANGELOG.md, and .kicad_prl were also excluded. No design file was altered.
Why this board
It is the first fixture in copperbench designed by an individual rather than an organization, and that is the point of including it.
Antmicro’s boards are house-style: consistent net naming, vendor libraries, a design review behind them. The Open Lighting Project’s board is community-maintained. This one is one person’s work, and it carries the marks — 193 off-grid wire endpoints, a symbol library with a .bak file next to it upstream, board and schematic metadata that have drifted apart in 84 places. None of that makes it a bad design; it is a real, working, manufactured RF module with a matched antenna. It makes it a representative design, and a benchmark drawn only from organizationally-produced boards would never test an agent against it.
It also sits at a useful scale. At 67 components on one sheet it is five times the microphone board and still comprehensible, which is what the medium tier is for even though it is single-sheet rather than hierarchical.
Baseline
Measured with kicad-cli 10.0.6:
| Check | Errors | Warnings | Detail |
|---|---|---|---|
| ERC | 0 | 295 | 102 lib_symbol_issues, 193 endpoint_off_grid |
| DRC | 1 | 2 | 1 zones_intersect; 2 lib_footprint_mismatch |
| Unconnected | 0 | ||
| Parity | 84 | 65 footprint_symbol_field_mismatch, 19 footprint_symbol_mismatch |
The one DRC error is zones_intersect: two copper zones overlapping without distinct priorities. KiCad 10 rules on this more strictly than the KiCad 8 the board was laid out in. It is enumerated in fixture.json as the DRC allowlist.
The 84 parity issues need a clearer account, because parity is normally the check an edit task most depends on. Every one of them is metadata, not electrical: footprint_symbol_field_mismatch is a Description or LCSC Part field differing between the board and the schematic, and the footprint_symbol_mismatch entries here are attribute flags such as Do not populate. No footprint is missing, none is duplicated, and no net differs. The netlist parity that a bad edit actually breaks is intact, and because drc_no_new_violations compares against this record by type and count, an agent that breaks real parity still fails.
The 193 endpoint_off_grid warnings are worth keeping rather than regretting. They are a fingerprint of the existing geometry, so an agent that reflows or regenerates the sheet instead of making a surgical edit will move that count and be caught by it.
Scale, and what it does to a surgicality bound
The schematic is 19,991 lines; the board is 26,468. A minimal correct edit here — renaming a net, adding a decoupling capacitor — is on the order of ten to thirty lines. A bound near 1 percent is roughly 200 lines: an order of magnitude of headroom for a clumsy-but-correct run, and still nowhere near the cost of regenerating a 67-component sheet. AC-3.7’s nominal 5 percent would permit 1,000 lines and is not a constraint at this scale.
Verifying this fixture
node scripts/hash-fixture.mjs --check fixtures/zhiayang-mikotoRegenerating the baseline reports
Only when the reference kicad-cli version changes, and the new counts belong in fixture.json in the same commit. kicad-cli writes a .kicad_prl next to the project as a side effect; delete it before rehashing, or the tree hash will change:
cd fixtures/zhiayang-mikoto/tree
kicad-cli sch erc --format json -o ../baseline/erc.json mikoto.kicad_sch
kicad-cli pcb drc --format json --schematic-parity -o ../baseline/drc.json mikoto.kicad_pcb
rm -f mikoto.kicad_prl