Glossary · ObservabilityAlso called monitoring, alertingChecked 27 September 2026
What is observability, and how is it different from monitoring?
Monitoring tells you something is wrong. Observability lets you work out why without reproducing it. The practical test is a question: when a customer says last Tuesday's run was wrong, can you answer from what the system already recorded, or do you have to run it again and hope it misbehaves the same way?
In one lineAnswering why, from what was already recorded.
Why it matters to you
For an AI system this is harder than for ordinary code, because the interesting failures do not throw. A step that returns an empty result, a provider that answers 200 with a rate-limit body, a model that answers confidently and wrongly: none of those raise an error, so a dashboard of error counts shows nothing at all.
What works is picking one number per pipeline that only moves when the pipeline is healthy, and reading it on a schedule. On the WhoFits lead engine that number is the yield rate, the share of enriched leads that turn up a website, because a silently blocked search engine returns an empty result with no error at all.
The other half is splitting surfaces by what each can see. Mooney sends errors to five places because each one sees something the others cannot, and a decision table names which to open first per symptom. After every release a smoke test fires a deliberate crash to confirm the pipeline still works, because monitoring that is never tested dies quietly.
If a proposal used this word and did not explain it, ask us.
The first conversation is with the people who would build it. If there is a fit, the Consult comes next.