SCPN-Quantum-Control Reproducibility Manifest

1. Source Code & Licence

Repository

github.com/anulum/scpn-quantum-control — currently in private access mode. Source access is granted to commercial licence holders, academic researchers, contributors under CLA, and peer-review reviewers. Request reviewer access →

Licence: AGPL-3.0-or-later (commercial licence available). SPDX header on every source file. ORCID 0009-0009-3560-0851.

2. Pinned Versions

Component	Version / Constraint	Source
scpn-quantum-control	0.9.5	`pyproject.toml`
scpn-quantum-engine (Rust crate)	0.1.0, PyO3 0.25	`scpn_quantum_engine/Cargo.toml`
Python	≥ 3.10 (tested 3.10–3.13)	`pyproject.toml`
Rust edition	2021	`Cargo.toml`
qiskit	≥ 1.0.0, < 2.0 (2.2 has GH #15476)	`pyproject.toml`
qiskit-aer	≥ 0.14.0, < 1.0	`pyproject.toml`
qiskit-ibm-runtime	≥ 0.20.0, < 1.0	`[ibm]` extra
numpy	≥ 1.24, < 3.0	`pyproject.toml`
scipy	≥ 1.10, < 2.0	`pyproject.toml`
mitiq	≥ 0.30, < 2.0	`[mitigation]` extra
ripser	≥ 0.6, < 1.0	`[topology]` extra
ruff	0.15.9	agentic CI matrix
mypy	≥ 1.20.0, < 2.0	agentic CI matrix

3. Phase 1 Campaign — DLA Parity on ibm_kingston

Pinned state

Git commit: 1b60f7b2aa9b05d6b346d7dae645b9d4cd02b4af

Branch: main, pushed to origin/main 2026-04-11

Backend: ibm_kingston (IBM Heron r2, 156 qubits)

Circuits: 342 at $n=4$, 2048 shots each

Trotter step: $t_\text{step} = 0.3$

Trotter depths swept: $\{2, 4, 6, 8, 10, 14, 20, 30\}$

Reps per point: 12 (depths 2, 20, 30) or 21 (depths 4, 6, 8, 10, 14)

Coupling matrix: $K_{ij} = 0.45\, e^{-0.3|i-j|}$

Data files (342 circuits across 4 JSON sub-phases)

data/phase1_dla_parity/phase1_bench_2026-04-10T183728Z.json # 42 circuits, wall 44.1 s
data/phase1_dla_parity/phase1_5_reinforce_2026-04-10T184909Z.json # 72 circuits, wall 56.7 s
data/phase1_dla_parity/phase2_exhaust_2026-04-10T185634Z.json # 138 circuits, wall 97.5 s
data/phase1_dla_parity/phase2_5_final_burn_2026-04-10T190136Z.json # 90 circuits, wall 65.1 s

Each JSON stores: experiment (sub-phase name), timestamp_utc, backend, job_ids, wall_time_s, n_circuits, t_step, circuits (per-circuit metadata including depth, sector, initial state, Qiskit counts), and aggregated_* (cumulative statistics across all preceding sub-phases).

IBM Quantum job IDs

d7ck79m5nvhs73a4nr10 # phase1_bench_2026-04-10T183728Z
d7ck7hb0g7hs73dqvbg0 # phase1_bench_2026-04-10T183728Z
d7ckcrh5a5qc73dosbmg # phase1_5_reinforce_2026-04-10T184909Z
d7ckft95a5qc73doseu0 # phase2_exhaust_2026-04-10T185634Z
d7ckide5nvhs73a4o780 # phase2_5_final_burn_2026-04-10T190136Z

All jobs are retained in the IBM Quantum Platform job log. Total QPU wall time across all five jobs: 263.3 s ~ 4.4 minutes.

Analysis output

figures/phase1/phase1_dla_parity_summary.json # all per-depth statistics
figures/phase1/leakage_vs_depth.png # Figure 1
figures/phase1/asymmetry_vs_depth.png # Figure 2

The summary JSON contains the exact per-depth Welch t-statistics, p-values, means, SEMs, 95 % confidence intervals, and the readout baseline used on the Phase 1 results page.

4. Rerun Protocol

Minimal rerun (no hardware)

Recomputes all Phase 1 statistics from the raw JSON count dictionaries. Does not touch IBM hardware. Wall time < 10 seconds on a modern laptop.

git clone <access-granted-url> scpn-quantum-control
cd scpn-quantum-control
git checkout 1b60f7b
python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
python scripts/analyse_phase1_dla_parity.py

Expected output: figures/phase1/phase1_dla_parity_summary.json, leakage_vs_depth.png, asymmetry_vs_depth.png, and a console summary matching Table 1 on the Phase 1 results page to the last reported digit.

Full rerun including Rust engine

Rebuilds the PyO3 crate from source and runs the complete test+bench matrix (4,771 tests collected, 9 deselected). Wall time ~15 minutes on a 12-core machine.

rustup toolchain install 1.83.0
pip install maturin==1.12.6
cd scpn_quantum_engine && maturin develop --release && cd ..
pytest tests/ -q --ignore=tests/test_hardware_runner.py

Full rerun including IBM Quantum hardware

Re-submitting the 342-circuit campaign requires an IBM Quantum Platform account with a Heron r2 backend authorised for the ibm-q/open/main instance (or paid Open Plan). Estimated ~5 minutes of QPU time. The current account is blocked on a 180-minute promotional allocation; see the Phase 2 status panel.

pip install -e ".[dev,ibm,rust]"
export QISKIT_IBM_TOKEN=<your-token>
python scripts/run_phase1_dla_parity_campaign.py \
--backend ibm_kingston --t-step 0.3 --qubits 4 \
--depths 2 4 6 8 10 14 20 30 --reps 21

5. Previous Hardware Campaigns

Campaign	Backend	Date	n	Circuits	Data path
Kuramoto-XY preprint	ibm_fez	2026-02	4, 6, 16	~33 jobs	curation pending
Pipe cleaner (diagnostic)	ibm_kingston	2026-04-10	4	retrieved	`.coordination/ibm_runs/pipe_cleaner_retrieved_*.json`
Phase 1 DLA parity	ibm_kingston	2026-04-10	4	342	`data/phase1_dla_parity/` (4 files, see Section 3)
Micro probe (cycle exhausted)	ibm_kingston	2026-04-10	—	—	`.coordination/ibm_runs/micro_probe_*.json`

6. Code Quality Gates

The same gates run locally via pre-commit install, in the preflight hook (tools/preflight.py), and in GitHub Actions CI. A commit is considered reproducible only if all gates pass.

gitleaks 8.21.2

Generic secret scanner. Blocks any commit that looks like it contains credentials.

tools/check_secrets.py

Custom vault-pattern scanner. Reads the local credentials vault, extracts high-entropy tokens, and blocks any commit containing them. See Contributing for the precedent incident.

ruff check + format

Linting and formatting (ruff 0.15.9). Mirrors CI configuration exactly.

mypy

Strict type checking on all public API surfaces (mypy ≥ 1.20).

pytest

4,771 tests collected (9 deselected). 97 %+ coverage gate. Parametrised across Python 3.10–3.13.

check_version_consistency.py

Verifies that the version string matches across pyproject.toml, __init__.py, CHANGELOG.md, and docs/index.md.

7. What This Manifest Deliberately Does Not Claim

Honest limitations

Independent replication on a second Heron r2 device has not yet been performed. Phase 2 will submit the same protocol to ibm_marrakesh; the result will be added to this manifest when available.
Coherence wall. Beyond ~250 two-qubit gates, decoherence dominates and error mitigation reduces but does not eliminate systematic bias. This is an intrinsic NISQ limit, not specific to our protocol.
Test count. 4,771 tests are collected by pytest --collect-only. Earlier documentation cited 4,828+; that figure has not been reproduced on the current commit and is being corrected in a follow-up documentation refresh.
Hardware access is private. The IBM Open Plan Quantum Platform account is personal. Reviewers cannot directly re-execute the campaign without separate access arrangements.
Active development. v0.9.5 is pre-1.0. Internal APIs may change between releases. Pinning to a specific commit hash is mandatory for reproducible results.

8. Contact

For reviewer access, data requests, or methodology questions:

protoscience@anulum.li · ORCID 0009-0009-3560-0851

Miroslav Šotek · ANULUM Research · Marbach SG, Switzerland