Gate component: guarantees the data feeds are pinned before the page proper
renders, so every accessor inside BenchmarksLoaded can expect its feed.
Native builds (tests, SSG prerender) compile the data in and render directly;
wasm fetches the staged /data bundle first, behind a lightweight shell.
The “Serialization — the wire codec” section: the LOGOS wire codec head-to-head against
the industry serializers across wire size, encode, decode and single-field random access.
Every figure is measured by the wirebench harness on the same logical data; the headline
(“smallest wire on N/N workloads”) is provable, and the structured/affine showcase is
fenced off from the fair results so the math-hack is never conflated with general data.
The “Our solver vs the field” section: per-family bar charts (ours vs Z3, Kissat, SaDiCaL)
over the families where structure beats brute force, led by the pigeonhole wall. Bars are
linear per row — the slowest fills the track, ours is its true fraction — so a microsecond
refutation reads as the sliver it is. All numbers are measured by
benchmarks/run-solver-vs-z3.sh; the framing is the honest one the data supports — SaDiCaL
completes (we beat it), Z3 and Kissat hit the resolution wall.