Parallax offset and latency-caused ink lag are independent sources and easy to confuse

What it is

When nib and ink disagree, people use one sentence: “the ink is not under the nib”. There are two sources. Parallax is geometry: present at rest, changes with viewing angle. Latency lag is time: lengthens with speed, closes on a stop. Both can be present at once. Filling parallax with prediction, or curing latency with bonding, twists the wrong knob to the end and still leaves a seam. Diagnosis has to split them.

Why it happens

Parallax is a function of gaze, cover thickness, and the two-plane calibration, independent of whether the hand moves. Latency is Δt times speed along the path tangent, independent of where the head is. An orthogonal experiment therefore exists: still → parallax only; head-on fast stroke → parallax minimal, latency dominates; slanted still → latency zero, parallax dominates. Stacked in slanted moving writing, the seam’s direction is neither purely tangential nor purely along gaze, and looks like “random bias”. Prediction extrapolates along the tangent and can only fill the tangential component of latency; the lateral component from a slanted view is beyond it, and at corners the tangential patch becomes a barb. Geometric calibration translates the whole path and does nothing to a latency seam that lengthens every moment. Calling both seams “inaccuracy” lets hardware and systems teams bounce the bug.

Studying it

Four cells: still head-on, still slanted, moving head-on, moving slanted. In each, measure seam length and direction (relative to path tangent, relative to gaze projection). Then prediction only, and calibration table only.

Independent variables: motion, viewing angle, prediction, geometric table. Dependent measures: seam length, angle from tangent, which intervention works in which cell.

Prediction should mainly help moving head-on; the table should mainly help still slanted. Little crossover support independence.

Where it stops holding

Very slow and head-on, both seams shrink inside nib width and the problem “vanishes”, then returns in another setting. An external display or projection may stack another delay and another optic; the four cells must be done on the final viewing surface. Nib occlusion undercounts static parallax; diagnose with a thin nib or a side view.

Applying it

• Split following bugs with the four cells before assigning work: still slanted to bonding/calibration, moving head-on to pipeline/prediction.
• Do not title a defect with a single “offset pixels”; state pose and whether the pen was moving.
• Prediction and calibration may both be on; acceptance still passes the four cells separately.
• How to check: a still slanted seam should shrink after loading calibration and should not shrink after turning prediction on; moving head-on the reverse. A seam that remains after a hard stop is parallax; one that closes is latency. With both interventions on, moving slanted should be the smallest residual, not a cell that was sacrificed.

Related

• Same group: C5.12.1 Glass-to-display gap causes parallax · C5.12.2 Parallax offset changes with viewing angle · C5.12.3 Offset needs per-model factory calibration
• Adjacent: C5.08 Inking Latency · C5.11 Inking Latency and Predictive Rendering
• Search: parallax versus latency · nib ink misalignment · gap diagnosis