Estimate by summing time constants for standard operator units

What it is

The Keystroke-Level Model (KLM) decomposes a skilled task into standard operators — keystrokes, pointing, mouse movement, dragging, button presses, mental preparation, and system response — assigns a time constant to each, and sums the sequence to estimate error-free completion time. It is the lightest-weight member of the GOMS family: GOMS needs a goal hierarchy and selection rules before it can decompose a method, while KLM skips that layer, assuming the method is already known and the path is unique, doing only "how long does this known path take" — in exchange for a modeling cost far lower than full GOMS analysis.

Why it happens

KLM can simplify complex human operation into a sum of fixed-category time constants because of one key assumption: as long as the task path is known, the user is sufficiently skilled, and no errors occur during execution, the operation sequence itself is determined — and only a determined sequence can be broken into independent units, each timed and then summed. This assumption compresses "predicting duration," an otherwise complex problem involving cognition, motivation, and environment, into a pure arithmetic exercise: number of keystrokes times the keystroke constant, plus number of pointing motions times the pointing constant, plus the count and duration of mental preparation and system waits. Decomposition requires first laying out the entire input stream in time order, then labeling each step's category, count, and duration; system response is a special case, since it is not determined by the user's action and needs to be measured directly or estimated from historical data rather than assigned a fixed constant. This method is naturally suited to comparing design variants for keyboard-and-mouse tasks, especially when two alternatives differ in step count, input volume, or where waiting occurs — KLM turns those differences into concrete numbers.

Studying it

A study must first define the task, device, input method, and the path skilled users actually take, then encode that path using categories such as K (keystroke), P (pointing), B (button press), H (homing the hand to a device), M (mental preparation), and R (system response). Record the repetition count for each operator category — system response should use real measurements rather than an assumed value — sum everything to get a predicted total time, and compare it against skilled users' real completion time. If a notable gap appears between predicted and measured values, trace the error source item by item: is the touch target smaller than expected, slowing pointing; did predictive text or autocomplete change the keystroke count; how far off is actual system wait distribution from what was assumed; or does the actual strategy users take simply differ from the path assumed during decomposition.

Where it stops holding

KLM does not predict learning, errors, hesitation, collaboration, multitasking switches, or more complex cognitive load, all of which fall outside what its "single skilled, error-free path" premise can cover. Touchscreens, voice interaction, gaze control, and networks with substantial latency variation each have operational characteristics different from a traditional keyboard-and-mouse setting, and directly reusing constants calibrated on a desktop scenario will systematically diverge from the real value — new input categories or recalibrated constants are needed rather than simply carrying old ones over. If the task's operational path itself is unstable — different users take entirely different sequences to reach the same goal — then no matter how accurately the constants are calibrated, the summed result will be distorted, because the model was never computing "how long the goal takes on average," only "how long this specific path takes."

Applying it

• Build a separate operator-sequence table for each candidate design, labeling every operator's category and repetition count.
• When comparing designs, focus on differences in total operator count, number of mental preparations, long movements, and system waits, rather than looking only at the single total-time figure.
• Calibrate key time constants with a small number of genuinely skilled users, rather than reusing old values from the literature or other products without verification.
• How to check: present the KLM time estimate alongside error rate, discoverability, and recovery cost in the same report, so readers do not mistake a single duration figure for a complete judgment of whether the design is usable.

Related

• Same group: B4.05.2 Rules for inserting mental preparation operators determine estimate accuracy · B4.05.3 Constants came from particular devices and eras and require recalibration
• Nearby: A10 Reaction Time and Movement Time · B4.04 GOMS
• Search terms: Keystroke-Level Model · KLM · task time estimation · operator unit