Skip to main content

Failed Message Insights page (WebApp)

The Insights page (/Insights, under DIAGNOSE in the sidebar) groups the failures in a time window so you can see what kinds of problems occurred and where to look first (Spec 043). It counts failure occurrences: every failed attempt. An event that failed three times counts three times, and an occurrence stays in its window after the event later completes or is skipped.

The page only reads. It never starts an AI analysis and never resubmits or skips anything.

What the page shows​

  • Range: 1h, 12h, 1d (default), 3d, 7d, 30d or a custom range of up to 90 days, plus an endpoint filter.
  • Group by
    • Error patterns (default): failures grouped by normalized error text, as before. This always works, with or without failure intelligence.
    • Likely failure causes: grouped by the category saved by failure intelligence, such as Transient dependency or Missing reference data, plus Uncertain, Not analyzed and Not eligible.
    • Recovery guidance: grouped by the guidance saved with each result, such as Retry may help or Change likely required. Guidance describes the failure when it was analyzed. It is not a current retry instruction, and it is never recomputed with today's thresholds.
  • Coverage: how many occurrences have a saved analysis. Classified means a known cause and confident guidance. Uncertain means the result is inconclusive or unrecognized. Not analyzed means no saved result; this includes failures that can no longer be analyzed because their event has since completed or been skipped. Not eligible means the endpoint is outside the analysis allow-list.
  • Groups: each shows its count, coverage, endpoints, event types, latest occurrence and top error patterns. and N more lists the rest.
  • View failures opens a paged list of the group's failed attempts with their analysis, saved cause and guidance, when that result was made (Analyzed at), and the event's status now. A marker shows when a newer analysis is running, its outcome is unknown or it failed. Each row links to the event's details with that failed attempt's failure intelligence card outlined and scrolled into view; the card names the provider and model and notes that threshold settings were not recorded with the result.

Likely causes, recovery guidance and coverage need saved classifications. When they aren't available the page keeps error patterns and says why: intelligence is off, its storage can't be read, the message store can't identify failed attempts, or the selection has more failures than the enrichment limit (10,000 by default; narrow the range or pick an endpoint). A link that asks for causes or guidance in such a selection opens on error patterns. When saved classifications can be read but no provider is configured (read-only mode), the page says that new analysis is unavailable, so Not analyzed failures stay that way until a provider is configured.

Saved classifications are read again when the page opens and on Refresh. New failures can take up to 30 seconds to appear. Changing the range or endpoint clears the page until that selection has loaded; Refresh keeps the current content, dimmed, until the new read arrives. If a selection can't be loaded, the page says so and offers Try again. The range, endpoint, grouping and open group live in the URL, so a view can be shared and coming back from event details returns to it.

Access​

Insights is cross-endpoint, so it requires the site Reader role. The endpoint filter doesn't lower that requirement.

API​

GET /api/metrics/failed-insights, GET /api/metrics/failed-insights/occurrences and GET /api/metrics/failed-insights/patterns; see WebApp REST API for parameters and the NimBus:Insights limits.