Dashboard numbers look wrong

Symptom

A widget shows a number that doesn’t match what you expected — cost too low, latency too high, a count that seems off. Almost always the data is right and the query is reading it differently than you assumed: the time range, granularity, aggregation, or filters change what a widget reports. Check those four before suspecting the underlying traces.

• A metric looks far higher or lower than reality.
• Two widgets that “should” match don’t.
• A number changed when you only changed the time range or granularity.

---

Quick checks

• The widget’s time range matches the window you have in mind.
• The granularity (bucket size) is what you expect — per-hour and per-day give different numbers.
• The aggregation (sum / average / median) answers the question you’re asking.
• No stray filter (model, status, attribute) is silently narrowing the data.

Causes and fixes

Cause	What you see	Fix
Time range / granularity	A number changed when you only changed the window or bucket size	A chart reflects the selected window and bucket. Set both to match your expectation — average latency per hour and per day differ from the same traces.
Aggregation mismatch	Two widgets that “should” match don’t	Sum vs. average vs. median answer different questions — confirm the widget uses the one you mean.
Filters narrowing the data	A metric looks far lower than reality	A widget filter (model, status, attribute) silently excludes traces; clear it to compare against the full set.
Eval sampling	An eval-based metric covers fewer spans than total traffic	If a metric is built on evals run at a sampling rate, it covers a subset of spans, not all of them.
Timezone	An apparent gap or spike at a day boundary	Day boundaries follow the dashboard timezone — the boundary effect, not missing data.

Diagnostic checks

Open the widget editor and read its time range, granularity, aggregation, group-by, and filters. Then cross-check one value against the trace explorer for the exact same window:

• Apply the same time range and filters in the trace explorer.
• Count the matching traces (or read the latency/cost column) and compare to the widget.
• If the two agree, the widget config — not the data — explains the number.

Minimal smoke test

Set the widget’s time range and granularity to match your expectation, clear extra filters, and confirm the value lines up with a trace-explorer count for the same window. They should reconcile within the rounding of the chosen aggregation.

Escalate

If a value still can’t be reconciled with the trace list for the same window, contact support@futureagi.com with the dashboard, the widget config, and the window.

Prevent recurrence

• Label widgets with their aggregation and window so readers don’t misread them.
• Keep one “all traffic, no filters” reference widget to sanity-check the others.

Next steps