Source-oriented public review package derived from the original September 14, 2026 archive. This repository contains C++/Python source, build scripts, validation tools, methodology and recorded benchmark outputs. It does not contain RED2.

GOLD1 benchmarks prime-counting approaches against pinned comparator versions using controlled worker budgets, repeated measurements, and robust protocol calibration. See the original project documentation below for precise claims and reproduction steps. Recorded benchmark results are historical artifacts, not independent validation or proof of general superiority.

The original documented Linux / MSYS2 UCRT64 workflow is:

./setup_all.sh

./validate_all.sh

GOLD1_AC_QUALIFY_REPEATS=5 RANK_RUNS=5 ./run_interval_matrix.shImportant: This GitHub-oriented source distribution excludes cached vendor tarballs, compiled build outputs, and generated checksums. Some setup scripts may assume bundled dependency archives. Obtain the versions listed in DEPENDENCIES.lock, review setup scripts, and verify dependency availability before executing. Commands above are from the original documentation and have not been executed against this repackaged distribution.

- results/: historical measurements (environment and comparison constraints matter).

- docs/: setup, history, changes, troubleshooting and handover material.

- tools/: benchmark and harness utilities.

- gold1/: implementation and associated source.

- THIRD_PARTY_NOTICE.md,- gold1/PRIMECOUNT_BSD_2_CLAUSE.txt: third-party notices.

- COPYRIGHT.md: original rights statement; legal-owner placeholder has not been filled in. Confirm rights and replace it before publishing.

Do not imply that third-party source is solely original work. The existing licensing documents should be reviewed before public distribution; this packaging is not a legal clearance.

See GITHUB_PACKAGING_NOTES.md.

The default setup is intentionally dependency-light: fast-prime is optional and never blocks Gold1.

./setup_all.sh

./validate_all.sh

GOLD1_AC_QUALIFY_REPEATS=5 RANK_RUNS=5 ./run_interval_matrix.shDefault ranked comparators: Gold1, pinned primecount 8.7, pinned primesieve 12.15.

The default is 12 workers. To run a fair four-worker matrix:

GOLD1_BENCH_THREADS=4 \

GOLD1_AC_QUALIFY_REPEATS=5 \

RANK_RUNS=5 \

./run_interval_matrix.shEvery ranked implementation now honors the same maximum concurrent budget. For two-endpoint comparators, calibration tests sequential 4+4 (four workers per endpoint, one endpoint at a time) against concurrent 2+2 (two workers per endpoint at the same time), then selects the lower robust median. Result JSON and CSV record the budget and selected modes. Compare timings only when the machine, build, interval matrix, run count, and thread budget match.

v2.3.6 measures each sequential and concurrent endpoint protocol three times by default, alternating which mode runs first. If the initial medians are within 2%, it automatically adds two tie-break measurements per protocol. It then selects the lower final median, without favoring either implementation. Complete samples, medians, relative gap, tie-break activity, selection, and reason are recorded.

Controls:

GOLD1_PROTOCOL_CALIBRATION_REPEATS=3

GOLD1_PROTOCOL_TIE_BREAK_REPEATS=2

GOLD1_PROTOCOL_NEAR_TIE=0.02The validated four-worker matrix placed the Gold1/primesieve crossover near width 1.5e11 at scale 1e18. Measure the boundary directly with:

GOLD1_BENCH_THREADS=4 \

GOLD1_PROTOCOL_CALIBRATION_REPEATS=3 \

GOLD1_PROTOCOL_TIE_BREAK_REPEATS=2 \

GOLD1_PROTOCOL_NEAR_TIE=0.02 \

GOLD1_AC_QUALIFY_REPEATS=5 \

RANK_RUNS=7 \

./run_crossover_bench.shThis runs widths 1.25e11, 1.50e11, 1.75e11, and 2.00e11, saving a separate crossover_matrix JSON/CSV pair.

Only if you want the fourth comparator:

GOLD1_WITH_FASTPRIME=1 ./setup_all.sh

./validate_all.shOn Windows this optional path needs Visual Studio C++ Build Tools and Rust/MSVC. On Linux it uses native Rust.

Exact validation gates ranking. Gold1 AC repeatability qualification/fallback is retained. CPU/RAM/OS/timestamps, source SHA-256, harness SHA-256 and available binary SHA-256 fingerprints are recorded. Missing optional fast-prime is explicitly recorded as optional-skipped; it is never silently treated as a failed competitor.

v2.3.6 preserves the v2.3.4 mathematical implementation, the v2.3.5 portable launcher and fixed thread budget, and every exact-validation gate. Its changes are confined to more robust protocol selection, crossover measurement, reporting, tests, and documentation.

The v2.3.4 optimized common-cutoff Delta-Gourdon path and correctness-first endpoint-specific Gourdon fallback are unchanged. The fallback remains in the same 10-case matrix and is explicitly labelled in result metadata; it is not represented as the paired-D optimized path. The matrix runner still retries one child execution that exits without producing RESULT_JSON.

The normal matrix still contains all ten cases, including the 10^14 and 10^15 fixed-width scale tests. A common-cutoff case uses gold1_execution_mode=common_cutoff_delta. If no common tuple validates, Gold1 switches to gold1_execution_mode=dual_cutoff_endpoint_fallback, reconstructing each endpoint with its own independently selected valid Gourdon tuple before taking the exact difference.

Before redistribution, run ./release_stamp.sh "Exact Legal Name". This replaces [YOUR LEGAL NAME] across release-controlled text and regenerates source provenance plus both SHA-256 manifests.