qs-dmss 0.1.4

Evidence-first reference package for the QuantumScalar Dark Matter Simulation Suite.

QS-DMSS

QS-DMSS is a deterministic, evidence-first reference build of the QuantumScalar Dark Matter Simulation Suite. This repository now ships the productization spine needed to move from prototype scripts into a reproducible package:

• Installable Python package
• Bundled demo assets for installed-package smoke testing
• Config-driven simulation CLI
• Local-first run cockpit and JSON API
• Parameter sweeps and multi-run comparison in the cockpit
• Experiment registry with saved comparison reports and bundles
• Objective-driven decision profiles with ranked recommendations
• Template-defined decision campaigns across multi-parameter search grids
• Run ledger with stable run IDs and config digests
• Evidence bundle with artifacts, metrics, manifest, and HTML report
• Replay and verification commands for reproducibility checks
• GitHub Actions CI and containerized runtime

What This Build Includes

The current reference implementation focuses on the backbone needed for productization:

• A NumPy-based split-step Schrodinger-Poisson solver
• YAML configuration loading with explicit validation
• Structured run outputs under runs/<run_id>/
• Structured experiment outputs under experiments/<experiment_id>/
• A local cockpit for launch, inspection, verification, replay, and bundle download
• Sweep support for exploring one parameter across multiple deterministic runs
• Decision campaign support for expanding a template into a multi-parameter grid automatically
• Comparison tooling for energy drift, norm drift, density, and runtime deltas
• Decision profiles that score runs against an explicit objective, constraint set, and ranking policy
• Durable experiment exports with copied run evidence, comparison JSON, report HTML, manifest, and bundle ZIP
• Evidence artifacts:
  • config.yaml
  • run.json
  • metrics.json
  • energy.csv
  • environment.lock.json
  • artifacts/final_density.npy
  • artifacts/final_state.npz
  • report.html
  • manifest.sha256.json
  • evidence_bundle.zip
• Verification tooling for manifests and config digests
• Replay support for deterministic reruns

Quickstart

Install the published package from PyPI:

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install qs-dmss

Run the bundled demo config from the installed package:

qs-dmss run-demo

Launch the bundled installed-package demo campaign:

qs-dmss campaigns run-demo

For source development, install the checked-out repository in editable mode:

python -m pip install -e .[dev]
qs-dmss run configs/demo.yaml

Reviewers who want to validate the published release without a source checkout can use the wheel-first path in docs/reviewer-wheel-quickstart.md.

Start the local cockpit:

qs-dmss cockpit --host 127.0.0.1 --port 8001

Then open http://127.0.0.1:8001 in a browser.

Inside the cockpit you can:

• Launch a single run from a checked-in or edited config
• Launch a parameter sweep across interaction strength, timestep, step count, amplitude, width, or seed
• Launch a template-defined decision campaign that expands into a reproducible multi-parameter run matrix
• Compare multiple runs side by side with shared experiment metadata
• Save a comparison into the experiment registry and reopen it later with report and bundle downloads
• Load an objective-driven template and see the recommended winner directly in the comparison view

Verify the generated evidence bundle:

qs-dmss verify runs\<run_id>

Replay a prior run using the captured config:

qs-dmss replay runs\<run_id>

Persist a saved experiment bundle from two or more runs:

qs-dmss experiments export <run_id> <run_id> --label "comparison bundle"

List saved experiment artifacts:

qs-dmss experiments list

Launch the decision campaign defined by a template:

qs-dmss campaigns run configs/demo.yaml

Or launch the bundled installed-package demo campaign:

qs-dmss campaigns run-demo

The checked-in demo template now includes a decision profile:

• objective
• constraints
• ranking
• campaign

That means sweeps, experiment exports, and template-driven campaigns can now return a replayable recommendation instead of only raw metric tables.

Container Runtime

Build the container image:

docker build -t qs-dmss .

Run the cockpit in Docker:

docker run --rm -p 8001:8001 qs-dmss

The image installs the built wheel, starts qs-dmss cockpit --host 0.0.0.0 --port 8001, and exposes the health endpoint at http://127.0.0.1:8001/api/health.

Project Layout

configs/                 Checked-in example configs
schemas/                 JSON schema for run configs
src/qs_dmss/             Package source
tests/                   Smoke and reproducibility tests
runs/                    Run ledger outputs (generated)
experiments/             Saved comparison artifacts (generated)

Development

Run the smoke tests:

pytest

CI lives in .github/workflows/ci.yml and validates:

• the editable install and test suite across Python 3.10 through 3.13
• static cockpit JavaScript syntax
• source distribution and wheel build metadata
• installed-wheel run-demo smoke test
• Docker build plus live /api/health and /api/configs probes

Release-candidate versioning and distribution artifact rules live in RELEASE.md.

PyPI distribution details and Trusted Publishing provenance live in docs/pypi-distribution-readiness.md.

Citation

Citation metadata lives in CITATION.cff. GitHub uses this file to populate the repository citation prompt, and Zenodo can use it when archiving GitHub releases.

For formal research references, prefer the Zenodo DOI citation:

• Project DOI: 10.5281/zenodo.20074924
• First archived release DOI: v0.1.3 / 10.5281/zenodo.20074925
• Software Heritage archival: pending; add the SWHID after Software Heritage reports a completed archive for the release.

Zenodo citation notes live in docs/zenodo-citation.md.

Current Scope

This branch intentionally focuses on the package/evidence/reproducibility spine first. Optional accelerator backends, UI layers, plugin expansion, and broader enterprise modules can now build on a stable execution loop:

configure -> run -> measure -> bundle -> verify -> replay

The cockpit adds the first browser-native product layer on top of that loop:

configure -> launch -> inspect -> verify -> replay -> download

The experiment registry now makes comparison durable too:

select runs -> compare -> save -> report -> bundle -> reopen

The decision layer adds recommendation semantics to that flow:

select template -> launch campaign -> score runs -> recommend winner -> export evidence

The campaign layer now automates the search plan too:

select template -> expand campaign -> run matrix -> score variants -> recommend winner -> reopen bundle