← back to legend
Ants: routing and our own terms (from Nov 2026)

Research pool / test pool

Our board library split: research boards we try approaches on; test boards never inspected or tuned on, so results do not overfit.

How it looks

Drawing of real data: our library split, not a physical object
PAIRRESEARCH POOLTEST POOLP1pic-programmerwater-level-monitorP2rp2040-minimalender3-loadcellP3rp2350-pico2-bb48ftdi-jtag-programmerP4gmsl-serializercvbs-mipi-bridgeP5si-test-boarddual-ipex-csi-interposerP6scalenode-cm4cm5ioP7stmp157-som-evbimx8mp-som-evbP8emberonetiliqua-motherboardplus 9 reserve boards, fetched and processed but in neither pool
The real split in our board library (library/selection.json): 8 matched pairs, one board of each pair in each pool. Board names are the open-source projects they come from.

How it works

PAIRRESEARCH POOLTEST POOLP1pic-programmerP2rp2040-minimalP3rp2350-pico2-bb48P4gmsl-serializerP5si-test-boardP6scalenode-cm4P7stmp157-som-evbP8emberoneits look-alikeits look-alikeits look-alikeits look-alikeits look-alikeits look-alikeits look-alikeits look-alikewater-level-monitorender3-loadcellftdi-jtag-programmercvbs-mipi-bridgedual-ipex-csi-interposercm5ioimx8mp-som-evbtiliqua-motherboardlocked awaywhile we developfinal score: measured on the test pool only

Where it fits

Scores come from the benchmark referee and are counted as clean boards.

Each benchmark board is used as a copper-stripped puzzle; CORE3588 itself is a separate benchmark alongside the pools.

← back to legend