The scoring curve maps q-diff range (sensor movement variability) to a raw score (0-10). Lower q-diff = more stable = higher score.

• Drag points on the chart to adjust the curve visually
• Edit exact values in the knot table below the chart
• Add/remove points to change curve complexity
• Between points, scores are linearly interpolated
• Values outside the curve range are clamped to the nearest endpoint

Production uses 6 knot points. Fewer points = smoother curve, more points = finer control.

After the curve produces a raw score, the rescale function maps it to the final 0-10 range. This controls score distribution shape.

• Linear (default) — even spread between low and high bounds
• Power — power >1 compresses high scores, <1 compresses low scores
• Sigmoid — S-curve that clusters scores near the midpoint
• Exponential — emphasizes differences at the high end
• None — use raw score directly (no rescaling)

Low/High Bounds define the raw score range that maps to 0-10. Scores outside bounds are clamped.

• Start Row — skip initial sensor samples (warmup period). Default 200 (~2 seconds at 100Hz)
• Stable Threshold — q-diff below this is considered perfectly stable. Lower = stricter
• Outlier Percentile — use this percentile of q-diff to ignore spikes. 98 = ignore top 2%
• Aggregation — how to combine multiple time windows into one score: Min (worst window, strictest), Max (best window, most lenient), Mean, or Median
• Target Windows — time windows in seconds to analyze. Each window is scored independently, then aggregated

Production defaults: start=200, threshold=0.02, percentile=98, aggregation=min, windows=3,5,7

• Save/Load presets to keep parameter configurations you want to compare
• Import/Export as JSON files to share configurations
• Run Analysis processes all ~800 cached CSV files (~40 seconds)
• Score Calculator lets you test a single q-diff value against current settings instantly
• Rebuild Cache re-extracts data from CSV files (only needed if new test files are added)

Typical workflow: load production preset, adjust one parameter, run analysis, compare distributions. Nothing writes to the database.