Source Health and Lifecycle Triage

A federated query is only ever as dependable as the sources behind it. SemanticFed registers each warehouse, database, API or file store once, then probes it continuously for reachability and latency. This concept takes the stream of probe results and turns it into a triage surface: a single Data Sources page, reached from Registry in the left navigation and subtitled as health and connectivity lifecycle, that shows which connections are holding up, which are slipping, and what should happen to one that keeps failing.

Without that page, the first sign of a sick source is usually indirect: a report errors, a number looks short, and somebody spends an afternoon tracing the problem back to an endpoint that started timing out. The design brings the signal forward and attaches a decision to it. Summary cards at the top count connected, degraded and disconnected sources beside a tally of probes run in the last twenty-four hours, so the shape of the estate is legible before any row is read.

Below the cards, a table lists each source with its connector type, a status badge, how long ago it was last probed, the latency that probe measured, and whether push-down is enabled for it. The sample rows span the lifecycle deliberately: healthy database sources answering in tens of milliseconds, one degraded API source at several hundred, a disconnected replica with push-down switched off, and a draft object-store source that has never been probed. Expanding the degraded row reveals a probe history for its last dozen checks, with a latency trend that climbs into a spike, a count of consecutive unreachable results, the upstream timeout message returned, and the time of first failure.

Beside the table, a Proposed Action card reads that sample evidence back as a recommendation: sustained probe failures justify a lifecycle transition, the source owner will be notified, and a small label names what triggered the proposal. Apply Transition and Dismiss keep the last word with a person. Source health threads through the rest of the product, since it explains why a query run slowed, informs which push-downs the planner can rely on, and appears among the actions agents may take under workspace permissions.