Translation-only stabilization can be fragile to rotation and focus breathing

General lesson: a motion gate that aligns only X/Y translation can look tolerant in unit tests while remaining fragile in handheld use. High-frequency barcode imagery is especially sensitive to small rotation, scale, perspective, and focus changes, so calibrate those transforms separately rather than treating a single normalized pixel-difference threshold as “hand tremor tolerance.”

Applicability and limits: synthetic sweeps are useful for finding sensitivity boundaries, not choosing a production threshold by themselves. Confirm with real device frames because scores depend on barcode content, crop dimensions, JPEG artifacts, rolling shutter, autofocus, and sampling cadence. Raising the threshold alone can admit meaningful movement; limited rotation/scale registration or a shorter, consecutive-frame stability rule may be safer.

Evidence/example — 2026-09-16: the Shield motion-only scorer was executed on a deterministic 480×96 barcode-like crop. Translation alignment tolerated ±8 horizontal and about ±4 vertical sample pixels, but the current 0.12 threshold was crossed at roughly 0.26° rotation, 0.3% scale change, or Gaussian blur near σ=0.65 px. A combined 1 px X/Y shift, 0.25° rotation, 0.2% scale, mild brightness change, and low noise scored 0.151 and reset dwell. All 17 existing scorer tests passed; the benchmark is synthetic and does not claim real-iPhone calibration.