Skip to main content

Methods and runs

A method is a repeated workflow identified in instrument data. A run is one execution of that method.

Chipper reconstructs runs from the logs it receives. One run can span more than one file, and a file can contain more than one run. When Chipper receives a fragment without enough context to prove that it is complete, it labels the run as partial instead of presenting it as a complete execution.

Methods​

Open View → Methods to compare method health over a selected time window. The dashboard and table show activity before a process model is trained, including:

  • Run count and current activity.
  • Error rate.
  • Open alerts.
  • Average duration.
  • Recent runs and related SPC controls.

Open a method to see its overview. If a trained process model is available, the page also adds step analysis, failure modes, and any expert-defined flow attached to that model.

The Methods surface is enabled account by account. If you do not see it, contact your account owner or Chipper Support.

Runs​

Open View → Runs to search by run name, method, instrument, file, outcome, score state, or start time. Runs are available even when process-model scoring is not enabled.

A run detail page ties together:

  • Identity, method, instrument, start time, duration, and outcome.
  • The latest available process-model verdict and conformance score.
  • Recent events produced by monitoring rules.
  • A time-based activity view.
  • Run steps and the log evidence that supports them, when available.

Use the outcome and evidence to distinguish an execution problem from an expected completion, then follow linked events, alerts, and files for the raw context.

Process models​

A process model is Chipper's learned, per-method representation of how a workflow executes: its steps, timing, equipment, and recovery behavior. It adds a verdict and deeper step-level analysis to a run.

Process models do not simulate future instrument state and are not required for basic observability. Files, methods, runs, durations, events, and alerts begin appearing sooner. A tuned model generally needs around three months of repeated run history, depending on method frequency and data quality.

For the end-to-end investigation flow, see Understand the run.