Activity quality flags and the needs-review queue

What quality flags are

When an Activity is uploaded, the parser runs basic data quality checks. Anything it spots gets recorded on Activity.quality_flags (a JSON list of issue objects) and, if serious, the needs_review flag is set so the activity surfaces on the coach dashboard for attention.

What gets flagged

The auto-extract logic flags MetricHistory entries when they look anomalous against history:

• Anomalous metric value: the auto-extracted value (e.g. best 20-min power) is more than 20 percent higher or lower than the athlete's previous value for the same metric type. Marked with a review note like "Modeled value 312 is 23% higher than previous (254)".
• Discrepancy against coaching value: the extracted value disagrees with the canonical "current" value on MetricHistory by more than a tolerance set per metric type.

Activities themselves can also be flagged in quality_flags for issues like:

• Power data dropouts.
• HR strap glitches.
• GPS gaps.
• Duration vs distance mismatches.

(The exact flag list is driven by the parser; it evolves as new data quality cases are added.)

Where flagged items show up

• The activity row on the coach dashboard gets a small warning indicator.
• The activity detail page shows the full quality flag list with detail.
• Flagged auto-extracted MetricHistory rows are surfaced for coach review at /coach/metric/<metric_id>/review/. See Reviewing metric submissions.

How to clear the flag

• Open the metric review page and accept or decline the value. The needs_review state on the metric clears on decision.
• On the Activity itself, clearing quality_flags and flipping needs_review is possible via the admin if you want to retire a flag without further action.

Why this matters

The quality system is an early-warning layer for the downstream calculators. Without it, a single bad data import can pollute the athlete's critical power model, CSS history, or zones. The flag plus review queue keeps a human in the loop before the bad data becomes canonical.

See also: Reviewing metric submissions, Metric exclusions, Load computation and override.