QUANTUM-HYBRID RESEARCH

AN ADVANCE LABS RESEARCH INITIATIVE

Where quantum meets classical, measured honestly.

Three feasibility studies — quantum-accelerated LLM training, Linux as the control plane for quantum hardware, and a hybrid classical/quantum motherboard — plus QLOS v0.1, a user-space quantum OS runtime backed by a 228-test emulator and toolchain. The conclusions are sober; the code exists to measure, not to sell.

[Proven] mathematically proven[Demonstrated] shown on real hardware[Theoretical] rigorous, unproven[Speculative] extrapolation
LEARN HOW IT WORKS →READ THE RESEARCH ↗THE VERDICTS ↓

|ψ⟩ = 1.00|0⟩

|0⟩
|1⟩

A real statevector — gates are rotations, not collapses.

Fig. 0 — a live single-qubit statevector on the Bloch sphere. H, X, Z, S are the real unitaries.

01

Three questions, three honest answers

I

QML-Accelerator

Can quantum subroutines accelerate LLM training?

Not before the mid-2030s. Readiness 2/10.

The asymptotic speedups are real but narrow; I/O bottlenecks are severe and the quantum clock-speed deficit is ~8–10 orders of magnitude. The verdict is now empirical: a 6-qubit hybrid adapter trains (loss 6.45 → 6.04 over 15 steps) — but its parameter-matched classical control trains better (6.36 → 5.93) while executing zero quantum circuits.

71,040 circuit executions to lose to one forward/backward pass

research →qml-accelerator/

II

QuantumLinux

Can the Linux kernel be ported to quantum hardware?

Impossible in principle — and that's the finding.

A literal port is ruled out by the [Proven] no-cloning theorem and measurement postulate: fork(), copy-on-write, and preemption have no physical realization for quantum state. Linux as the classical control plane, via a narrow QALLOC/QEXEC/QMEASURE/QFREE syscall interface, is the only viable design — and QLOS v0.1 implements it in user space.

228/228 tests passing — emulator, runtime, toolchain, kernel-init

research →quantum-linux/

III

HybridBoard

Can a QPU live on a motherboard next to a CPU and GPU?

The binding constraint is latency, not bandwidth.

The control loop must close well inside qubit coherence times. For superconducting QPUs a "motherboard" is a category error — the system is a rack-and-cryostat installation — and no consumer market exists today. The QCX bus spec, ACPI QDEV firmware model, and hybrid scheduler map exactly where the bottlenecks are.

≤2 µs control loop over the QCX bus, or the qubits decohere

research →hybrid-board/
02

QLOS v0.1 — a normalized dev loop for quantum hardware

No operating system can ever run on a QPU — a corollary of the no-cloning theorem, not an engineering gap. What can exist is the control plane. QLOS v0.1 realizes it in user space: a classical runtime that leases qubits, verifies and schedules circuits, and returns only classical measurement shadows through a four-call discipline.

QALLOCQEXECQMEASUREQFREE

Because quantum state can never be inspected after the fact, assemble-time validation against QISA-v0.1.yaml plays the role the MMU plays classically: reject the program before any state exists. The binding contract is QLOS-DESIGN-v0.1.md.

TEST SUITE

228/228

emulator72

QLOS runtime + scheduler55

toolchain85

kernel-init16

Run on every push by ci.yml

EDIT → CC → EXEC → GDB → OBJDUMP

bell.qs

# edit — write QISA-K assembly

$ $EDITOR examples/bell.qs

# cc — assemble → QOBJ v0.1, validated against the ISA spec

$ python toolchain/qas.py examples/bell.qs -o /tmp/bell.qobj.json

# exec — submit through qalloc / qexec / qmeasure / qfree

$ python examples/qrun.py examples/bell.qs --shots 1024

# gdb — single-step one shot, per-instruction amplitudes

$ python examples/qrun.py examples/bell.qs --trace --seed 42

# objdump — disassemble back to canonical .qs — round-trip safe

$ python toolchain/qdis.py /tmp/bell.qobj.json

03

Research index

Recommended reading order per project: research document → workflow → code. Every document carries per-claim epistemic tags.

№STUDYRESEARCHWORKFLOWCODEVERDICT
01Quantum speedup analysis for LLM training01-qml-accelerator.md01-qml-workflow.mdqml-accelerator/2/10
02Linux-on-quantum feasibility, QISA-K, emulator, QLOS02-quantum-linux.md02-linux-workflow.mdquantum-linux/control plane only
03Classical/quantum motherboard architecture03-hybrid-board.md03-hybridboard-workflow.mdhybrid-board/latency-bound
04Web-based 3D quantum visualization for education04-quantum-viz-education.md—site/ — this page8/10
05Interactive explainer — learn quantum computing in the browser04-quantum-viz-education.md—/learn — try itlive

Committed reference artifacts: classical_baseline.json · hybrid_run_log.json · init-report.json

04

Every claim carries its evidence

Untagged sentences are context, not claims. Tags are never silently promoted when material moves between documents, and citations were web-verified against the June 2026 state of the field.

[Proven]

Mathematically proven. The no-cloning theorem is why no OS will ever run on a QPU — no roadmap changes this.

[Demonstrated]

Experimentally shown on real hardware, or verified live against primary sources at the time of writing.

[Theoretical]

Rigorous but unproven in practice. Asymptotic speedups live here — real on paper, unrealized in any machine.

[Speculative]

Extrapolation or conjecture. In code, it becomes a labeled assumption with a tunable constant — never treated as free.

ENFORCED MECHANICALLY — .github/workflows/ci.yml

  • ✓py_compile sweep over every Python file
  • ✓the 228-test quantum-linux pytest suite
  • ✓strict -Wall -Wextra -Werror C11 build of the scheduler demo
  • ✓YAML parse of the QISA-K instruction-set spec

Why explore it anyway? The hybrid classical/quantum split is the asymptotically correct division of labor, not a temporary compromise. Working out the map now — with every claim tagged — is cheap, and shows exactly where the bottlenecks are.