Skip to content

Task YAML Structure

Task card mental model

A Jarvis-HEP run does three things:

  1. Draw a parameter point.
  2. Run a workflow that turns that point into observables.
  3. Compute a LogLikelihood from those observables.

YAML structure (overview)

New to YAML? Start here: YAML format overview

Quick checklist (required vs optional)

  • Required: Scan, Sampling
  • Recommended: EnvReqs
  • Choose one workflow backend: Calculators or Operas
  • Optional: LibDeps, Utils

A typical card is organized like this:

Scan:               # run name + output location
Sampling:           # how points are drawn + objective
LibDeps:            # shared backend deps installed once (optional)
Calculators:        # external-program workflow (optional)
Operas:             # in-process workflow (optional)
EnvReqs:            # platform + dependency contract (recommended)
Utils:              # helper functions (optional)

Scan (run location and output layout)

Scan defines the run name and where outputs are written.

Minimal (copy-paste)

Scan:
  name: "MSSM_Run"
  save_dir: "&J/outputs"

For a standalone project, Jarvis writes outputs under the project root using <TASK-NAME> = Scan.name:

&J/outputs/<TASK-NAME>/

Path placeholders and resolution: Placeholder in path resolution

Common options

Scan:
  name: "MSSM_Run"
  save_dir: "&J/outputs"
  sample_directory:
    limit: 200
    width: 6
    archive_samples: true
  • sample_directory: configures how sample data is stored or displayed.
  • limit: 200: caps the number of samples kept in the directory at 200.
  • width: 6: sets the display width to six items per row or column.
  • archive_samples: true: archives samples that exceed retention limits, removing them from active view.

Output directory layout

The output layout follows the CLI contract: Command line tools

Common locations:

  • &J/outputs/<TASK-NAME>/SAMPLE/: per-sample working files
  • &J/outputs/<TASK-NAME>/DATABASE/: structured outputs and converted products
  • &J/logs/<TASK-NAME>/: run logs
  • &J/images/<TASK-NAME>/: flowchart and plots

Sampling (how points are generated)

High-level shape (reference):

Sampling:
  Method: "Random"        # sampler name
  Variables: []            # scan variables
  Bounds: {}               # sampler-specific controls (optional)
  LogLikelihood: []        # objective definition

Details: Samplers; Variables

LibDeps (shared external backends)

LibDeps lists external program packages that are not part of the per-sample workflow.

Use it for backends that you install once (per machine or per environment) and then reuse across scans.

High-level shape (conceptual):

LibDeps:
  # shared backend packages

Details: Library dependencies

Calculators (external executables backend)

High-level shape (conceptual):

Calculators:
  # one or more calculators / steps

Details: Calculators

Operas (in-process operators backend)

High-level shape (conceptual):

Operas:
  # modules/operators and mappings

Details: Operas

EnvReqs (platform and dependencies)

High-level shape:

EnvReqs:
  OS: []
  Python: {}
  Check_default_dependences: {}   # optional

Details: Environment Requirements

Utils (helper functions)

Common uses: interpolation tables, reusable functions used by expressions.

Details: (if you have a Utils page later, link it here)

Expressions and I/O