FAQ

Plain answers to the questions newcomers ask first, with links into the documentation for depth.

🧩 Basics

What is DispatchAtlas?

DispatchAtlas is a toolkit for scheduling and optimization across distributed-computing environments. It bundles a benchmark atlas (curated scheduling problem instances), a solver registry (dispatching rules, metaheuristics, exact adapters, and the NDSO family), a campaign engine (reproducible batches of solver runs), an analysis layer (statistical summaries and figures), and a public portal that presents the exported evidence. Start with the quick start and the tutorials.

What is DispatchAtlas not?

It is not a production dispatch service or a cluster scheduler daemon. It evaluates scheduling algorithms on recorded benchmark instances and publishes the resulting evidence; it does not schedule live workloads. The architecture overview shows where each package boundary sits.

How does DispatchAtlas relate to existing schedulers and frameworks?

It compares against them rather than replacing them. The platform comparison and the platform-comparison exports in analysis exports document how rival frameworks differ in capabilities, features, and inspectable artifacts — with every row labeled by its evidence source, and without converting capability rows into performance claims.

📚 Terms To Know

What is the "benchmark atlas"?

The curated catalog of scheduling problem instances the toolkit evaluates solvers on. Each benchmark family declares its taxonomy, generator profile, citations, and scale envelope; materialized instances are validated, characterized, and wrapped in provenance envelopes. See the benchmark model.

What is an "evidence bundle"?

A directory of statistical tables, accessible SVG figures, and a supplement of excluded or withheld rows, all derived from one completed campaign and filtered to one evidence tier. See the evidence bundles page.

What is an "evidence tier" (or "disclosure tier")?

One of four fixed ids — core, speed, quality, platform — that controls how much evidence an export may expose. Lower tiers cannot surface higher-tier mechanisms or withheld rows, and the filter fails closed rather than leaking. The tier policy is executable, not editorial; see analysis exports.

What is a "campaign"?

A configured set of solver runs: benchmark instances crossed with solvers and objectives, executed under explicit seeds, stop criteria, budgets, and checkpoints, with one JSON record per run. The campaign engine owns this layer, and the second tutorial runs one end to end.

📊 Results And Evidence

Where do results come from?

Campaigns run locally in the experiments workspace, which keeps recipes in source control and heavy outputs on the machine that produced them. The portal then consumes committed, disclosure-filtered exports derived from those run records — never the workspace itself.

Is the current portal data final evidence?

No. Current portal bundles are smoke-development evidence for interface checks, documentation examples, and preview pages. Comparative claims wait on campaigns that pass the statistical and release gates described in readiness.

Does the site run campaigns?

No. The site consumes static benchmark, solver, gate, and analysis export bundles. Campaign execution remains in the dispatchatlas.lab package.

Does the recommender claim the best solver?

No. It explains capability fit, exclusions, budget assumptions, and confidence labels from metadata and available evidence — see the solver recommender.

Which policies filter what an export may expose?

Executable disclosure policies in dispatchatlas.analytica filter evidence bundles, result datasets, and companion pages before anything is published. Forbidden rows are excluded and recorded in a supplementary exclusions manifest with the reason.

When are downloads public?

Downloads stay release-gated until license, data-use, disclosure, package, and quality checks pass — see downloads.

🤝 Using And Citing

How do I cite DispatchAtlas?

Use the interim software entry on the citation page and refresh it once release metadata is available.

Where do I get help?

Open an issue or discussion on the GitHub repository, or start from the community page. The troubleshooting page covers the most common failures first.