results/: the published benchmark runs, each in the compact form described
in Measurement methodology, from which the site renders one
chapter per run. bench-runs/, where runs are recorded, is not under version
control.
Tests: known-answer tests against the reference programs’
fixtures, point counts against PARI, and the certificate corpus. Every
recorded fixture opens with how it was produced: the generated
tests/common/*.rs with the Sage command that rewrites them, the corpus
tests/fixtures/cert_corpus.txt with the test that records and checks it,
and the report’s report/tests/fixtures/*.tsv with the run and derivation
they come from, as far as these were recorded.
Nix: the pinned toolchains and the derivations for the package,
checks, benchmark executables, reports, assembly listings, and this site;
nix run .#validate-commits -- RANGE runs the checks at every commit of a
range, as the continuous integration does for each
commit a push or pull request adds.
Documents: the chapters; docs/index.md gives their order.
Reference programs: the known-answer-test generators and order
proofs the Rust fixtures come from (kat*.sage, prove_orders.sage,
agm_vectors.sage, sieve.sage, gf2_*.sage), and the surveys behind the
choice of fields and curves: candidate field sizes (field_sizes.sage), GLS
fields (gls_survey.sage), Weil descent (ghs.sage, ghs122.sage), and
complete Edwards curves over pseudo-Mersenne primes (edwards_search.py).