Constants came from particular devices and eras and require recalibration

What it is

KLM time constants came from empirical measurement in early keyboard-mouse environments. Modern touch, pointer precision, input methods, system latency, network waits, and user proficiency differ; reusing old constants can create systematic bias and requires recalibration. Together with the previous two cards, this completes the three steps of a proper KLM analysis: sum the standard units (card one), correct the special mental-preparation operator with an insertion rule (card two), and finally confirm that every constant used in the sum actually came from a measurement environment that matches the current scenario — an error at any one of the three steps makes the final number untrustworthy.

Why it happens

Time constants drift across devices and eras because they were never pure physiological constants to begin with — they are the joint result of human motor capability and a device's response characteristics: keystroke time depends on keyboard layout and key travel, pointing time depends on target size and pointer precision, and a touchscreen tap is further affected by finger occlusion and mis-tap correction algorithms, all factors that shift systematically as hardware generations change. System response is a special case — it is often not a stable single value but a distribution that fluctuates with network conditions, server load, and cache hit rate, and plugging in a single measurement or a median from an old paper amounts to forcing a quantity that should be described as a distribution into a point estimate, losing exactly the variance information most needed to judge how reliable the estimate actually is. Different user populations also differ systematically in movement speed and cognitive preparation pace, so using constants measured on a different product in a different era to predict a current target population's performance is, in effect, using one mismatched distribution to approximate another.

Where it stops holding

Not every new project needs all six operator categories fully remeasured — if the goal of the analysis is only a relative judgment, such as which alternative has fewer steps or more waiting, reusing old constants for a rough relative comparison still has value, since constant error acts on both alternatives at once and largely cancels out. Rigorous recalibration is genuinely needed for scenarios that will be used to make an absolute time commitment (promising a business stakeholder that a flow completes within some number of seconds), which are far more sensitive to constant error. Touchscreens and environments with substantial network variability especially cannot reuse constants calibrated on desktop keyboard and mouse, because the gap between these environments and a traditional desktop setting already exceeds what counts as negligible error — continuing to use the old values does not produce an approximation, it produces a systematically wrong one.

Applying it

• First identify which operator categories genuinely dominate the current task's time, and prioritize recalibrating those, rather than spreading equal effort across measuring all six categories.
• Measure with five to ten skilled users matching the target profile, recording the median and the spread of the distribution for each operator category, not just a single average.
• For system response specifically, record both the P50 and P95 percentiles, and note the network conditions, cache state, and device model at measurement time, so readers know under what conditions the number holds.
• How to check: attach the source of each constant, the sample size, the confidence range, and key assumptions whenever reporting an estimate; rerun calibration whenever the platform version, input method, or a key component affecting response speed changes, rather than letting one calibration result persist long after the environment it was measured in has changed.

Related

• Same group: B4.05.1 Estimate by summing time constants for standard operator units · B4.05.2 Rules for inserting mental preparation operators determine estimate accuracy
• Nearby: A10 Reaction Time and Movement Time · K1 Platforms and Device Forms
• Search terms: KLM constants · recalibration · device latency · response time distribution