sensor-calibration-workbench

Sensor Calibration Workbench

Use this skill when the sensor technically works, but the readings are not yet trustworthy enough for the project.

Resources

• references/calibration-flow.md - end-to-end calibration workflow and evidence checklist
• references/common-failure-patterns.md - warm-up, scaling, drift, saturation, and environment mistakes
• references/persistence-and-revalidation.md - storing coefficients and deciding when recalibration is needed

When to Use

Use this skill when the request involves:

• volatile or implausible sensor readings
• "how do I calibrate this sensor?"
• load-cell factor tuning
• CO2, magnetometer, or color-sensor calibration
• storing calibration coefficients in EEPROM or flash
• deciding whether the problem is calibration, hardware, or environment

Do not use this skill when the sensor is not detected at all. That should route through hardware or bus bring-up first.

Workflow

1. Confirm the measurement problem:
   • unstable -> open references/common-failure-patterns.md
   • offset or scaling error -> open references/calibration-flow.md
   • values good once but bad later -> open references/persistence-and-revalidation.md
2. Identify the calibration class:
   • one-point or zero-offset
   • two-point scale calibration
   • multi-orientation or environmental calibration
3. Collect reference evidence before changing coefficients:
   • known reference values
   • warm-up state
   • ambient conditions
   • sample stability
4. Decide how calibration values will persist and how revalidation will be triggered after reboot, firmware update, or field drift.

Core Rules

• Calibration without a known reference is guesswork.
• Warm-up and stabilization time are part of calibration, not a side note.
• Do not mix hardware-fault symptoms with coefficient-tuning symptoms.
• Store both the calibration values and enough metadata to know when they became stale.

Verification

• Confirm readings converge toward a known reference after calibration.
• Confirm the calibrated values stay stable across repeated samples.
• Confirm stored coefficients reload correctly after restart.
• If the project has operating thresholds, verify those thresholds against the calibrated output rather than the raw sensor value.