Semogram Docs
Operators

Runs overview

Reading executions and making the approve-retry-correct call.

Every execution leaves a run: impact, narrative, evidence, diagnostics, and your recorded decision. Unreviewed runs are debt; reviewed runs are training data.

Your job

  • Lead with impact: what changed, what was produced, what needs a human.
  • Read the narrative against expectation, then evidence against freshness.
  • Classify failures before acting: transient (retry), structural (correct at its layer and rerun), gated (review the approval deliberately).
  • Never silently rerun a structural failure, and never leave a finished run undecided.

First time? Read your first run. Broken? Troubleshooting.

Image: run timeline with the failed node selected and its diagnostics open.

On this page