Repository navigation
Qiskit Lab v0.4.0 adds a native local CHSH workflow and strengthens result persistence, statistical reporting, receipt validation, and recovery of ambiguous IBM submissions. It also bundles the official Qiskit skills with reproducible optional environments and expanded offline regression checks.
Changes
- Native local CHSH experiments:
qiskit-lab chsh --shots 8192 --seed 42builds all four measurement configurations, samples them with independent configuration seeds, records circuits and provenance, and generates the existing native HTML report. This command is ideal/local only; it has no hardware or noise flag. #8, #14 - Finite-sample CHSH confidence bounds: reports retain approximate standard errors and add conservative two-sided 95% Hoeffding intervals. Observed violations and statistically supported violations are separate fields, and contradictory local-setting angles are rejected when supplied. Bounds assume independent shots and preselected settings/sample sizes; they do not account for systematic errors or establish a loophole-free Bell test. #13
- Durable scientific results: execution and QAOA persist results before PNG rendering; presentation failures preserve the scientific receipt and return safe diagnostics. #11
- Ambiguous submission recovery: durable submission intents and unique per-run tags support
recover-submission RUN_ID. Recovery only reads existing IBM jobs, handles zero/one/multiple matches explicitly, and never submits or retries a job. #12 - Versioned receipts and actionable errors: new receipts use schema version 1 with shared validation; historical unversioned receipts remain supported. Local input, filesystem, rendering, dependency, and IBM failures receive relevant next steps without exposing raw SDK bodies. #16, #17
- Official Qiskit skills: Runtime migration, error mitigation, SQD solver selection, and fermionic simulation guidance are bundled with complete resources, upstream licensing, a pinned revision, and file hashes. Deterministic resource checks, representative behavior-evaluation cases, and a review-only update workflow are included; passing resource checks does not prove model skill-selection behavior. #10, #18
- Optional quantum capabilities: isolated locked SQD/fermionic profiles and offline capability diagnostics keep optional packages out of the base runtime. Distribution metadata checks do not establish native-library or solver correctness. #15
- Regression coverage: backend dashboards, interrupted polling, result-presentation failures, and local Chromium report interactions/downloads have targeted tests. Compatible Python and report-asset dependency updates are included. #5, #6, #19
Installation and compatibility
Requires Python 3.11–3.13. From a checkout or extracted source archive of this release:
uv sync --locked
uv run --locked qiskit-lab doctor
uv run --locked qiskit-lab chsh --helpThe wheel contains the Python package, CLI, and report assets. The source distribution additionally includes the DAM kit, project instructions, bundled skills, optional-environment manifests/lockfile, and deployment scripts. SHA256SUMS covers both attached distributions.
- See receipt compatibility for schema 1 and historical receipts. Unsupported explicit schema versions are rejected; removing the version is not a supported migration.
- See optional capabilities before installing SQD or fermionic dependencies.
- See skill evaluations and updates for resource checks versus actual agent-behavior evaluations.
- Publishing a GitHub release does not update DAM deployments automatically. Use the existing-deployment update procedure, preserving owner files, results, credentials, models, and schedule settings.
Validation
- 709 offline tests passed locally on Python 3.13.7 with a project-local pytest temporary directory. Overall branch-aware coverage is 98%;
job_tracking.pyretains 100% line and branch coverage. - Ruff, mypy, shell syntax, both lockfiles, pinned-skill integrity, and deterministic skill-resource checks passed. Agent-behavior evaluations were not run.
- Wheel/sdist versions agree at 0.4.0. Archive inspection verified the DAM kit, all bundled skill resources, optional environment, and absence of runtime/private directories.
- A fresh installed-wheel environment passed hash-checked locked dependency installation, dependency compatibility, installed-package location/version checks, offline doctor, CHSH CLI parsing, and real QASM2/QASM3 round trips outside the source checkout.
- All nine optional native-library smoke combinations passed on Python 3.11–3.13 for SQD, fermionic, and fermionic-sim profiles using synthetic models.
- The release PR passed all 16 checks. Final main commit
68088730dae433071949d571ab867d4979ede9b5passed Python 3.11/3.12/3.13 and kit CI and container/report-assets/Chromium CI. All nine optional-profile smoke jobs passed on the identical release source tree. - Both distributions and
SHA256SUMSare attached. Uploaded assets were downloaded and their SHA-256 digests verified against the local originals.
Scope
No PyPI or container-registry publication is included. Release validation does not contact IBM, execute QPU jobs, or establish live connectivity. Optional native-library smoke uses synthetic models; it does not validate arbitrary solver workloads or actual model skill-selection behavior. The repository has no project-level license; bundled third-party licenses apply to their respective files.