Document the GPU compute component (0.7.0) New Benchmark-Methodology "GPU" section (the two subtests, auto device selection, runtime-loaded OpenCL, why it's ungraded). Scoring & Grades now names GPU alongside network as left out of the overall. Accuracy Notes' "does not measure GPU" bullet becomes "does not measure GPU graphics / video / ML". The Baseline gains the [components.gpu] rows and an integrated-GPU-class note; the machine-inventory field mentions the GPU. Home intro and the sidebar's "not in the grade" line updated. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SNrEbxEc3Q8FWSU9T5PdCx
Note the live soak screen in the TUI Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SNrEbxEc3Q8FWSU9T5PdCx
Document the soak (sustained-load) test New Benchmark-Methodology section covering the soak kernel, the per-second sampling, and the peak / steady / retained / onset / stability figures. Note it in the Home task table, the Scoring page (ungraded, like the link probe), and the result-file field table on The Baseline. Also update the all-core memory-read row for the 0.4.x barrier + 32 MiB floor change. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SNrEbxEc3Q8FWSU9T5PdCx
Document the AES/SHA and all-core-read subtests
Document single-threaded core pinning
Network is no longer folded into the overall grade
Document the network component and the --net-target link test
Populate the wiki: methodology, scoring, compare, baseline, config, accuracy Home page indexes six deep-dive pages that expand on what the README summarises, cross-linking back to it rather than duplicating it.