Tasks are split into standard units such as keystroke, point, home, draw, mental preparation, and system response

What it is

The Keystroke-Level Model decomposes a skilled task into a finite set of standard units: keystroke, pointing, pointer movement, homing the hand between keyboard and mouse, drawing, button press, mental preparation, and system response. Each category gets a time constant, and summing the sequence gives an error-free skilled completion time. This card is about the taxonomy itself — it exists separately not to repeat "how to estimate with KLM," but to answer a more upstream question: why exactly these categories, no more and no fewer, and how this classification was drawn in the first place.

Why it happens

This taxonomy holds together on a somewhat unintuitive judgment: the number of independent degrees of freedom in human operation that actually affect duration is quite limited. Pointing and drawing both look like "the hand is moving," but they are split into separate units because pointing only cares about the distance from a starting point to an endpoint (describable with a model like Fitts's law), while drawing cares about the shape and length of the entire path — the two follow completely different timing regularities, and merging them would lose critical information. Conversely, "moving the hand to the keyboard" and "moving the hand to the mouse," despite pointing in different directions, are merged into the same "homing" category, because both share the same degree of freedom that determines duration — the distance and muscle memory needed to shift the hand from one position to another. Whether a grouping is reasonable is not judged by whether the actions look alike, but by whether they share the same physical or cognitive mechanism that determines timing. How the categories are chosen also determines which design differences the model can see at all: if two alternatives differ in the count of some unit — fewer keystrokes, a closer pointing target, fewer hand switches, a shorter blocking wait for system response — the model reflects those differences clearly; but if the difference lies along a dimension this list does not cover, the model is simply blind to it.

Where it stops holding

This unit list is an approximation summarized from desktop keyboard-and-mouse scenarios, not a universal taxonomy of operation. Touch gestures, voice input, gaze control, and gesture-based interaction each involve degrees of freedom entirely different from a desktop setting, and directly applying these six categories to describe them misses exactly the factors that determine how long these newer input methods take — a new taxonomy needs to be built for new input methods, rather than forcing new actions into old categories. More complex cognitive processes, recovery behavior after an error occurs, and switching between multiple tasks also cannot be reliably compressed into some unit plus a time constant — these phenomena carry too much inherent variability, and forcing a model onto them only manufactures a false sense of precision. The choice of decomposition granularity should be guided by "what layer of difference does the design question at hand need to see," not by mechanically reusing all six categories from this list every time.

Applying it

• Write a separate unit-sequence table for each candidate alternative, listing each unit's category, occurrence count, and triggering condition.
• When comparing alternatives, focus on keystroke count, total pointer travel distance, number of hand switches between devices, drawing path length, and blocking wait duration, rather than comparing total time in the abstract.
• Calibrate the timing of only the categories that genuinely dominate the current task, rather than indiscriminately reusing every old constant.
• How to check: when reporting a result, state which unit taxonomy was used, the assumptions behind each category, the corresponding interface version, and exactly what level "skilled user" refers to — without this information, readers have no way to judge whether the decomposition actually aligns with the design question at hand.

Related

• Same group: B4.12.2 The placement of the mental preparation unit is governed by heuristic rules and is the main source of estimation error · B4.12.3 System response time counts toward the total only when it blocks the user · B4.12.4 Absolute constant values are questionable, but the differences between alternatives are relatively reliable · B4.12.5 Touch, voice, and gesture lack agreed-upon constants and must be measured before use
• Nearby: B4.05 Keystroke-Level Model · A10 Reaction Time and Movement Time
• Search terms: KLM operators · standard unit · time constant · operator taxonomy