Most "try our toolkit" pages are screenshots and a contact form. We wanted something a prospective user could actually run — on their own machine, against the real binaries, in the time it takes to read a README. That is quantum-examples: a public, evaluation-only bundle of the Anzaetek Quantum toolkit, with thirteen demos that each execute against the shipped binaries rather than a recording of them.
What it is
The repository ships time-limited, qubit-capped binaries per platform, plus a demo/ tree of runnable example subfolders and a one-command tour:
tar xzf dist/quantum-dist-<platform>-test-*.tar.gz
export QUANTUM_DIST="$(pwd)/dist-<platform>-test"
./demo/run-all.sh
The bundle is fully self-contained — no toolkit source, no Rust, no Python, no jq. The included quantum-client binary speaks the server's JSON-RPC natively, so the demos and the self-test need nothing on the host but a shell. We verify that end-to-end in a stock debian:bookworm-slim container with no packages installed: it extracts the Linux CPU bundle, pings the server, inspects a Bell circuit, and confirms the QOS O(1/N²) sample-complexity law numerically — all from a bare image.
It is an evaluation product, and we are explicit about that. The binaries expire, the CLI is capped (12 qubits for general circuits, with a separate 64-qubit ceiling for the scalable error-correction and Pauli-propagation paths), and the license is evaluation-only. A commercial license is separate. The bundle is also early and moving fast — we would rather ship something runnable now and label its rough edges than wait for polish.
The thirteen demos
Each demo runs against the binaries, not a transcript:
| # | Folder | Shows | |---|--------|-------| | 01 | 01-bell | inspect a circuit (info / compile / optimize) + Aria→Lean 4 | | 02 | 02-qft | a parameterized QFT model → Lean 4 at n=3/4/5 | | 03 | 03-rust-harness | a standalone Rust app talking to quantum-server over JSON-RPC | | 04 | 04-qos | Quantum Oracle Sketching — the O(1/N²) sample-complexity law | | 05 | 05-optimizer | a real optimizer reduction (naive 7 gates → minimal 3) | | 06 | 06-finance | regime labelling + backtest on a synthetic series | | 07 | 07-qml-qcbm | a Quantum Circuit Born Machine, trained (KL→0) | | 08 | 08-qml-classifier | a quantum classifier trained end-to-end | | 09 | 09-mbqc | a measurement-based computation pattern | | 10 | 10-ubqc | delegated blind computation (BFK) | | 11 | 11-lean4 | Aria spec extraction → a Lean 4 theorem | | 12 | 12-ecc | error-correction encode/decode with MWPM | | 13 | 13-pauliprop | Pauli propagation on the scalable expectation path |
All finance and QML data is synthetic — a rescaled, noised, renamed series, not real market data — and the demo says so where it matters.
Platforms
Evaluation bundles are provided per platform. macOS (Apple Silicon) is the complete tour: Metal plus CPU, including the QML and finance demos. The Linux linux-amd64-gpu CUDA bundle is the other full tour — quantum-finance links libtorch_cuda.so, so --device cuda trains demos 06–08 on Cuda(0). The Linux CPU bundles (x86-64 and arm64) ship the libtorch-free quantum / quantum-server / quantum-client CLIs and therefore skip the ML/finance demos, which need a paired libtorch runtime. RISC-V is still in progress.
Aria — the quantum language, in the open
The demo models are written in Aria, a small readable DSL for quantum circuits with built-in proof obligations. An Aria model is human-readable and parameterized, and it exports to QASM, a Lean 4 theorem, and an MBQC pattern — all through the quantum CLI. The docs/aria-tutorial.md tutorial takes you from a first Bell state to parameterized, formally-specified models, and editor support installs in one command:
./editors/install.sh # VS Code + Neovim + tree-sitter highlighting
The language now has its own dedicated, fully open-source home at github.com/Anzaetek/aria-quantum-language-oss-public — Apache-2.0, pure Rust, with its own aria CLI, runtime, 32 worked examples (31 numerically verified), and Lean 4 proofs. See the companion post for that release. Inside quantum-examples, Aria is the surface you read the thirteen evaluation models in.
Why publish an evaluation bundle at all
Because the claims we make about the toolkit — that a circuit inspects, optimizes, and exports to a machine-checked theorem; that QOS hits its O(1/N²) law; that a QCBM actually trains — are checkable, and the honest way to let someone check them is to hand over a binary that runs the check on their hardware. The bundle is the proof, not the brochure.
Clone it, run demo/run-all.sh, and tell us what breaks. For the algo-trading companion, see flow-master-examples.