Cookbook
Data quality report

Data quality report

When an answer feels thin or the assistant says it cannot see a metric, first ask icuvisor for a data-quality report. This read-only report explains whether the issue is sparse history, Strava-restricted activity imports, missing stream metadata, HR/TRIMP-only load, stale wellness sync, missing sport thresholds or zones, or missing race/calendar context.

When to use this

  • “Why can’t you see my data?” or “Why did that answer ignore my workout details?”
  • Before a coaching prompt that depends on streams, thresholds, wellness, or race calendar context.
  • After reconnecting intervals.icu, changing device sync, or importing historical activities.

The recipe 

Use icuvisor to explain why you may not be able to see enough of my data.

Run get_data_quality_report for 2026-01-01 through 2026-01-31. If I name a sport,
set the sport filter to that sport; otherwise inspect all visible data.

Then:
- Summarize the overall data-quality status in one paragraph.
- List each warning or critical diagnostic by code and plain-language meaning.
- Tell me the next action for each issue, naming the icuvisor follow-up tool when one is returned.
- Do not fetch raw activity streams unless the report specifically recommends checking one activity ID.
- Do not guess missing values; say whether the problem is restricted source data, missing streams,
  HR/TRIMP-only load, stale wellness, missing thresholds/zones, sparse activity history, or calendar/race data.

What icuvisor does

StepToolWhy
1get_data_quality_reportReads bounded activity, fitness/load, wellness, profile, and calendar probes and returns actionable diagnostics.
2Follow-up tools from report recommendationsUse only the targeted tool named by each diagnostic, such as get_activities, get_activity_streams, get_training_summary, get_wellness_data, get_athlete_profile, or get_events.

A good answer looks like

Data-quality status: warning. icuvisor can see recent activities, but stream-dependent and load-dependent answers may be incomplete.

  • restricted_source: 5 Strava-imported activities look like restricted summaries. Re-import those workouts directly from Garmin/Wahoo/Coros or use available summary fields only.
  • missing_stream_metadata: visible activities do not advertise stream channels. For a specific workout, inspect that activity with get_activity_streams only if stream-level analysis is required.
  • trimp_or_hr_load_available: load appears HR/TRIMP-derived. Treat it as neutral training load, not as power-based TSS.
  • stale_wellness: the latest wellness row is older than the report end date. Refresh the wellness bridge or choose a window ending on the latest synced row.

I should not estimate hidden watts, HRV, or race context from those gaps.

Variations

  • “Run this for only Ride data from the last 90 days.”
  • “Before my race-week review, check whether you can see race events and recent wellness.”
  • “Explain which data gaps I should fix first before asking for a training-plan review.”

Why this prompt works

  • One bounded preflight. The report checks common visibility failures without downloading raw stream samples by default.
  • Actionable diagnostics. Each diagnostic includes a code, severity, evidence, and next action so the assistant can explain the cause instead of apologizing vaguely.
  • No guessing. The prompt tells the assistant to name the data gap and stop rather than fabricate hidden activity, wellness, load, or calendar values.
Weekly training reviewReadiness check