Day 80: Measuring calibration error honestly
How wrong is it, really?
Before the full tape-measure validation (Days 82–84), characterize calibration error in a controlled way: measure known objects or a person with known dimensions across varied distances and poses. This tells you the *baseline* error the calibration introduces, separate from model error. You'll express it as a range ('±2–4 cm on shoulder width under good conditions'), not a single optimistic figure.
Illustrative: calibration error grows as pose conditions degrade — the honest story to document.
Error bars are a feature, not an admission
Stating '±3 cm under typical conditions, worse off-angle' is stronger than claiming '98% accurate' — it shows you measured, understood the failure modes, and can communicate uncertainty. That's the mark of an engineer who ships honest systems, and it's precisely what the Day-90 checkpoint and interviews reward over a suspiciously round accuracy claim.
Key terms
- Mean absolute error (MAE)
- The average absolute difference between measured and true values — a natural error metric for measurements.
- Error characterization
- Systematically measuring how error varies with conditions, expressed as ranges rather than a single figure.
Why report measurement error as a range tied to conditions (e.g. "±3cm ideal, ±8cm off-angle") rather than a single accuracy figure?
