Trace one failed Run
The text of a sample reply in the developer preview says the Run failed. A teammate opens the workflow diagram, points to a model step and says, “That's where it failed.” What evidence would convince you that the model was called at all?
Follow one identity
To see the example in the browser, open the default read-only preview using the
fixture setup and sign-in instructions.
You can also follow the same records in the
fixture test.
Find test_fixture_bridge_orb_and_loom_share_one_registered_revision and read from the first
Bridge request.
- In Bridge, expand the Circle beside the synthetic reply labelled Retained reply. It gives you the Run ID for this turn. The Run status reads Not retained in this reply.
- Open the reply’s … menu and choose View run in Orb. Match the Run ID, read the status and look at the evidence
attached to it. In the test, find the assertions on
statusandsummary. What does the sample itself tell you was executed? - Follow Exact Pattern into Loom. Match the workflow name and revision to the ones recorded for the Run. The diagram tells you which steps this revision permits. Which record would you need to say that a particular step took place?
Account for the failure
Open the code that prepares the sample
and find _seed. It creates a Run, writes RunStatus.FAILED to the ledger and attaches
that Run ID to the conversation. The fixture supplied the status itself. No model ran.
Try the same reasoning on a harder case. Suppose a real Run is marked failed, its Scroll contains a model step, and the event for that step is missing. You know how the Run ended. Whether it reached the model is still an open question. The Orb guide explains how to tell what the system captured, what the current view loaded and where the record has a gap.