Machine-as-input specialization with statistical skepticism. Related work landmarks, not marketing.
docs/research/silicera-paper.md — thesis and scopedocs/research/silicon-split.md — multi-machine strategy protocoldocs/research/compiler-feedback-format.md — SCFdocs/research/llvm-integration.md — remarks bridgedocs/benchmarks/single-machine.md — one-host evidence pathdocs/research/hne-concept.md — HNE companion conceptdocs/research/amd-integration.md — hypothetical surfaces (no partnership claim)docs/research/llvm-integration.md — Phase II compiler feedbackdocs/research/hne-concept.md — Hardware-Native Executiondocs/research/amd-integration.md — independent research posturedocs/research/compiler-feedback-format.md — feedback shapedocs/benchmarks/ — portable-vs-native, size-classesRelated work includes PGO, BOLT, ATLAS, FFTW, hardware counters, and -march=native. Silicera’s contribution is the fingerprint-bound profile loop with mismatch → baseline.
Silicon Split (Machine A + B placeholder until a second Zen box), Portable vs Native vs Silicera, Cold Machine, Wrong Machine. CLI: silicon-split, harness, compare.