Skip to main content
StackNext is an evidence source, not an answer generator. Every claim about runtime behavior should point to a route, trace id, error id, span, duration, status code, query or finding id.

​
Start with the sample

In the dashboard, the Overview page shows the sample window and what the current sample observed. In an MCP client, call get_project_summary first. Check meta.connection.instrumentation: waiting means no app is connected. Read meta.completeness and the sample window before drawing conclusions.

​
Explore the dashboard

HTTP requests view listing routes with status, duration, error and database query counts HTTP requests view listing routes with status, duration, error and database query counts

​
Investigate an error

  1. Open Errors or call list_errors with a route, name or query filter.
  2. Read the latest matching error and its redacted stack.
  3. If resolvedTraceId is present, inspect the request timeline and the failing span.
  4. Check stacknext_list_diagnostics with category: "errors" for a repeated fingerprint or a 5xx response without a correlated error.
  5. Report what is proven, what is correlated by the route/time fallback, and what remains unknown.
An error grouped by fingerprint shares a normalized signature, not necessarily a cause. A missing trace link means the correlation is approximate or absent; it is not proof that the error happened outside a request. Server errors view grouping one error by fingerprint with its route and correlated trace id Server errors view grouping one error by fingerprint with its route and correlated trace id

​
Investigate a slow request

  1. Use Requests sorted by duration, find_slow_requests, or list_requests with sort: "slowest".
  2. Open the request and compare self time across spans.
  3. Inspect database queries inside the trace. Drizzle durations are null; never treat them as zero.
  4. Check slow-request and slow-span findings. A slow child span may be suppressed when the whole request is already reported.
  5. Test whether the pattern is isolated by filtering the same route or method.
A parent span includes its children. Never add parent and child durations together.
Trace detail for a slow route showing the span waterfall, the selected span and its attributes Trace detail for a slow route showing the span waterfall, the selected span and its attributes

​
Inspect a trace

get_trace returns the span tree, missing parent ids, errors and associated queries. A missing parent is a completeness gap, not proof that the span is unrelated. Use exact span ids when describing a link. Unknown or verbose-only span types are retained and labelled rather than mapped to a guessed feature. See instrumentation coverage. Traces view listing captured traces with status, service, duration, error and span counts Traces view listing captured traces with status, service, duration, error and span counts

​
Use the CLI

The dashboard, CLI, export and MCP all read the same collector sample. A snapshot is frozen and cannot answer questions about current behavior.