"I Open-Sourced 135 Tests That Shouldn't Exist in the Same Repo"

#distributedsystems #rust #softwareengineering #riemannzetafunction

Most research code is abandoned after publication.

We built the opposite: a system where every architectural claim is backed by an executable test, every limitation is documented inline, and every commit is sealed with zero diagnostic noise.

HARMonis Prime SET-11 — Sealed.

What we built:
• 135/135 invariant tests passing
• Neumaier-compensated θ(t) oracle (ε ≤ 1e-14)
• Backlund-type truncation bound with monotonicity verification
• Criterion 30-run benchmark baseline
• SHA-256 integrity placeholder for Odlyzko dataset

What we did NOT build (honest limitations):
• True MPFR ζ(½+it) — f64 fallback only
• Odlyzko dataset validation — manual download pending
• FPGA acceleration — software simulation only
• AVX-512 kernel — experimental, may not compile

The proof is one command:
git clone https://github.com/Ayub19123/Harmonis-Prime.git
cd Harmonis-Prime
git checkout 476fd34
cargo test --lib -- --nocapture

Whitepaper (permanently archived):
Zenodo: doi.org/10.5281/zenodo.20777632
Figshare: doi.org/10.6084/m9.figshare.32732766

Phase 2 begins: MPFR Z(t) oracle, Odlyzko automation, AVX-512 kernel.

Every claim has a failing test first. The precision is eternal.