Datasets & Scenarios

A dataset is a named data bundle attached to a model: the values that fill the sets and params a JModel formulation declares. Keeping data out of the model means one formulation, many scenarios — "Q3 forecast", "peak season", "half fleet" — each solved and compared without ever editing the model itself.

Requires JModel: datasets fill a JModel source's open set/param declarations. A flat or imported model already carries its values inline, so scenarios don't apply to it — write the formulation in the JModel lens first. (The feature is instance-gated along with JModel.)

The Data tab

The workspace's Data tab manages the model's datasets:

• New dataset — create a bundle of sets and params values. Edit as a table (member lists, key→value entries) or as raw JSON.
• Generate skeleton from the model — pre-fills the dataset with every set and param the JModel source declares, so you only fill in values and can't misspell a symbol.
• Import file — upload a .dat (AMPL-style), .json, or .csv file; the parsed preview is validated before you save. For CSV, you name the single param it fills.
• Use / Stop using — pick which dataset the JModel lens compiles against while you work (or none, to use the model's inline values).

Every dataset is validated against the formulation with compiler-grade checks: a missing set/param, a key the model doesn't declare, a scalar where an indexed value is expected, or a wrong index arity is flagged inline — a typo never silently falls back to inline values.

A dataset is data only: deleting one never touches the model.

Scenarios: run and compare

The Scenarios section of the Solve tab turns datasets into runs:

1. Each dataset appears as a row with its latest run's status, objective, solver, and time.
2. Solve all launches every dataset (or launch them individually). Compilation happens server-side — the browser sends a reference, not a flattened model — and each run rides the normal async solve pipeline, so scenario runs show up in the model's history like any other execution.
3. Compare solutions — select two completed runs to diff their solutions: which variables differ and by how much (identical solutions are called out explicitly).

This is the classic what-if workflow: change the data, never the model, and let the runs accumulate side by side.

Via API

.../datasets/{dataset_id}/solve compiles the project's JModel source against the dataset server-side and returns an async task envelope; a compile error returns 422 with the message and source position. See the Models API.