Recognition rate and bodily feel must be logged separately for each hand morphology

What it is

Finger-count evaluation cannot ship one overall recognition rate and one mean “was it tiring.” It has to keep two columns per morphology stratum: did recognition pass for this stratum, and did holding the shape pass for the people in it. A small-palm stratum may recognize poorly and not ache yet. An arthritic stratum may recognize well and collapse immediately. One composite number hides the failing stratum inside the passing one. Stratification is not a statistical layout choice. It is the evidence format for whether this channel holds for the people you claim.

Why it happens

Recognition error and bodily cost do not come from the same place on the same hand. Size and distance hit the gap in the image. Tendon coupling and joint disease hit the hold. Mean recognition is pulled up by the majority young-adult hand. Mean comfort is pulled up by short tasks. Average those two and a product infers “counting works.” After stratification you see children systematically merging 3 and 4, and older adults unable to hold 4 for ten seconds. Without strata, improvement goes to the wrong place: more data for already-recognized large hands, no calibration for small ones, no edge trigger for limited joints.

Studying it

Define strata in advance: palm-width quantile, age band, syndactyly or joint limitation, left/right hand. Each stratum needs an n that can carry its own confusion matrix and time-to-collapse, not a post-hoc slice that finds two children. Recognition and bodily feel must be taken on the same trials: show the count (recognition), then hold (feel). Do not test recognition on one group and fatigue questionnaires on another. Publish a table, not a pie. The control is the unbinned overall accuracy, to show how the headline hides stratum failure.

Where it stops holding

Tiny samples make hard strata unstable; restrict the claimed population in advance (adult near-field only) rather than conclude from a two-person child stratum. Prostheses and missing digits need their own stratum, not an “other” bucket. Cultural reports of comfort vary widely; the bodily column should lead with time-to-collapse and compensatory angle, with rated ache as support. Body-scale skeleton strata belong to another group. This table takes the hand only.

Applying it

• Write the ship gate as “recognition and hold pass in every stratum,” not “95% overall accuracy.”
• The report template includes palm width, age, and limitation axes. A test missing an axis is unfinished.
• When a stratum fails, pick the lever by column: calibration or distance cap for recognition failure, flash-then-drop for hold failure. Do not announce the channel verified from the headline number.

Related

• Same group: C4.09.1 Extended-finger count can express a choice or a quantity · C4.09.2 Palm size and shape change which canonical forms are recognized · C4.09.3 Holding a finger combination has extra postural cost
• Adjacent: C4.05 Three dimensions of gesture evaluation · C4.13 Full-body pose and skeleton tracking
• Search: stratified evaluation · disaggregated accuracy · hand morphology