Skip to content
← All projects
Recursive Reflective Solver (RRS v13.0) / Prototype

When the obvious answer fails, try a better question.

A solver project investigating how structured search, constraints, and feedback can help tackle difficult logic problems.

Explore how it works ↓
SYSTEM / CONCEPT
InputsInformation to work withIndependent stepsKeep responsibilities separateReview the resultCheck · learn · refine
Architecture sketch / not a product screenshot

Explore the project

Open the design and demonstrations, then follow the references for more detail.

Explore the concept

Interactive 40-Point Benchmark Explorer comparing RRS runtimes and proof generation against Z3, Kissat, PyMatching 2, and LLVM across 10 constraint domains.

The design, in more detail

Definition

A standalone Rust computational reasoning engine prototype executing sub-millisecond constraint solving and machine-checkable proof generation across 10 formal verification and constraint domains.

The problem

Formal verification, post-quantum cryptography, and hardware ASIC logic equivalence checking currently rely on disparate, memory-heavy solvers (Z3, Kissat, NuSMV, LLVM-MCA) that suffer from execution bloat, non-deterministic runtimes, and opaque verification certificates.

The approach

RRS v13.0 operates under principles of Bare-Metal SIMD GF(2) bit-parallelism, physical multi-megabyte disk dataset ingestion, and zero-information state arbitration. By operating directly on raw bit vectors with zero dynamic runtime allocation in its hot path, RRS achieves 38× to 1,473× speedups against reference academic and industrial tools.

How it works

  1. SIMD GF(2) Acceleration: Vectorized bit-parallel matrix transformations executing polynomial arithmetic over Galois fields at hardware memory bandwidth.
  2. Physical Dataset Ingestion: Real-world disk benchmarking ingesting up to 39.1 MB raw datasets (NIST McEliece KATs, EPFL AIG benchmarks, Artix-7 bitstreams).
  3. Machine-Checkable Proof Ledger: Emits cryptographic DRAT and Büchi proof certificates enabling third-party verification without re-running the solver.
  4. Side-by-Side Reference Baseline Probing: Executes live subprocess benchmarks against Kissat, Z3, PyMatching 2, Yosys, NuSMV, and LLVM-MCA.
  5. Sub-Millisecond Determinism: Zero heap allocation in constraint inner loops guarantees predictable worst-case execution times for mission-critical systems.

Project notes

Speedup vs Kissat (ASIC)
48.0× (0.30 ms)
Speedup vs Z3 (SMT)
474.0× (0.39 ms)
Speedup vs PyMatching (QEC)
859.0× (0.15 ms)
Hot-Path Allocation
0 Bytes Heap

These are the project’s documented design notes. Consult the linked implementation and its version before relying on a specific capability.

Development history & next steps
2024

Algorithmic Inception

Derived Galois field polynomial bit-parallel reduction algorithms in Rust.

2025

Telemetry Gauntlet Expansion

Engineered 40-point verification gauntlet ingesting physical multi-megabyte disk datasets.

2026

v13.0 Benchmarking Release

Integrated machine-checkable proof output and sub-millisecond edge AI BitNet GEMM accelerator.

What comes next

Porting the SIMD Galois field kernels to custom RISC-V co-processors and evaluating space-hardened FPGA synthesis.

Source material & related links

Follow the documentation, repositories, and related sites behind this project.

RRS Core Engine & 40-Point Benchmark Gauntlet

Standalone compiled Rust engine (native/rrs_core/) and automated test gauntlet.

Empirical Benchmark Results & Proof Certificates

Comprehensive timing data verifying 48× to 1,473× speedups across 10 constraint tracks.

Explore 40-Point Benchmark MatrixRRS Core Repository

Topics: Formal Verification · Rust · SIMD · Galois Fields · Cryptography · Bare-Metal · Algorithms