GitHub

Tempo Framework — Overview

The Tempo Framework orchestrates TEMPO/TEMPO2 runs in a predictable, reproducible way. It is built from small, typed settings that describe paths, engine options, input modifications, what to capture/keep, runtime workspace layout, and logging.

Core ideas:

  • You configure a run using TempoRunSettings (composed of RunPaths, EngineOptions, InputModifiers, CaptureOptions, RetentionOptions, WhiteNoiseOptions, WorkspaceOptions, LoggingOptions).
  • Files are materialized into a job workspace before execution, according to WorkspaceOptions.
  • The engine runs with consistent flags and working directory selection.
  • Outputs are parsed into structured Julia results.

Data flow (typical)

  1. Build TempoRunSettings (or copy/modify with copy_with).
  2. Optionally validate(settings) to check inputs and par_output filename.
  3. Clean old artifacts in the job area (optional, cleanup_before_run=true).
  4. Materialize a job workspace (flat or split layout).
  5. Execute TEMPO/TEMPO2 with derived flags (capture residuals if requested).
  6. Parse engine outputs into result types.
  7. Optionally save artifacts/manifest and clean temporary files per policy.

Paths and layout

  • par_input and tim_input are file names relative to work_dir.
  • par_output is a filename-only token; the runner writes it under:
    • job_root/<par_output> for layout=:flat
    • job_root/output/<par_output> for layout=:split

WorkspaceOptions (highlights)

  • work_mode = :inplace | :jobdir and job_name control where the job root lives.
  • layout = :flat | :split and optional temp_dir control on-disk structure and execution cwd.
  • link_tim, snapshot_par control how inputs are staged into the job root.
  • cleanup_before_run, keep_tmp_on_success, keep_tmp_on_error control cleanup behavior.
  • write_manifest adds a small record of what was executed.
  • Optional I/O mirroring controls (e.g., io_mirror = :none | :full | (:depth_minus, N) if enabled in your build) can mirror job directories or outputs up the directory tree.

Results (high level)

Parsing builds structured results that can include:

  • Basic fit/engine outputs and internal iterations
  • Residual statistics
  • Optional white noise diagnostics (if enabled)

See the dedicated Results page for details.