Power source on an activity (and ignoring bad Garmin estimates)

What "power source" tracks

Each Activity has a power_source field that records which device produced the power data in the file. The Hub uses the source for:

• Sanity checks (e.g. Garmin's wrist-based estimated power is much less reliable than a real meter).
• Calibration history, so a coach knows which device produced which numbers over time.
• Comparison, when an athlete has multiple power meters and you want to compare a known-clean one against another.

Recognised values

The POWER_SOURCE_CHOICES enum:

• '' (Auto / Unknown).
• stryd (Stryd, run power).
• garmin (Garmin, including wrist-based estimated power and Vector pedals).
• stages (Stages crank-arm meter).
• quarq (Quarq spider meter).
• favero (Favero Assioma pedals).
• wahoo (Wahoo).
• other.

Most files don't carry an unambiguous source, so the field is typically blank on first import. Set it by hand on the activity detail page if you want it tagged.

Ignoring Garmin estimated power

A separate flag, garmin_power_ignored, exists for the case where Garmin produced wrist-based or pedal-based estimated power that you do not trust. When set:

• The power metrics on the activity (avg_power, max_power, normalized_power) are kept on the record for visibility.
• Downstream calculators (Load by power, CP extraction, zone distribution) are told to ignore this activity's power and fall back to speed / pace or HR.

This is useful when:

• The athlete forgot to attach the real power meter and Garmin filled in estimated values from wrist or HR.
• A new device produced clearly wrong values during a calibration period.
• The athlete swapped pedals mid-season and old data should not pollute the model.

How to set the flags

Both power_source and garmin_power_ignored are editable from the activity detail page or via admin. Setting them does not delete any data; it just tells the system how to interpret the existing values.

See also: Activity quality flags, Metric exclusions, Load computation and override.