The model applies only to highly skilled performance; novice behavior is not parallel

What it is

CPM-GOMS's parallel structure assumes users have already automated the method, feedback, and subgoals involved. A novice typically reads a hint, then finds the control, then confirms what it means, then acts — most activities run serially with long pauses in between; the same interface produces a different critical path for the two kinds of users. This card sits next to, but is a different limitation from, the GOMS-family card "the analyzed object is a fixed method": that card limits whether modeling can even be applied to a given user population at all (has the method stabilized). This card goes further and is specific to CPM-GOMS's parallel structure: even once the method has stabilized, whether the user has reached the level of automaticity that lets perceptual, cognitive, and motor resources be occupied simultaneously is a separate threshold — a fixed method does not imply operation has automated to the point of being parallelizable.

Why it happens

Parallelism holds only for skilled operation because it requires the perceptual, cognitive, and motor processors to be able to handle different subtasks independently, and that independence is precisely a product of extended practice: an expert can let the eyes pre-search for the next target while the hand is still moving toward the current one, because "searching for what's next" no longer requires the expert to actively think about it with working memory — it has become a low-cost background activity that happens incidentally. A novice has none of that available: they must spend working memory understanding what a label means, judging whether the current state matches what is expected, and confirming whether this step should be taken — these alone consume the entire capacity of the cognitive processor, leaving no margin for anything that could be called parallel. As a result, a batch of activities an expert can layer on top of each other gets forced apart into a sequence of serial steps for a novice, interspersed with long pauses that never appear for an expert at all. This also explains why the same interface produces a completely different critical path for experts and novices: the interface has not changed — what determines "which activities can run in parallel," the user's level of automaticity, simply does not satisfy the same assumption for the two populations.

Where it stops holding

Skill itself is task-specific, not a label that transfers everywhere once attached to a person: an expert who breezes through financial reports every day will show novice-like serial behavior the moment they switch to an unfamiliar graphics editor, because the two tasks require automating completely different content. A finer boundary: even for the same skilled user completing the same task, only some segments may have been practiced enough to reach automated, parallelizable processing, while other, relatively unfamiliar segments remain serial — a user long practiced at routine data entry can still fall back into a step-by-step, serial confirmation mode when hitting a rarely occurring exception-handling branch. So rather than pinning a simple "expert" or "novice" label on an entire user population, skill should be labeled per task segment; the same model can perfectly well describe one segment with a parallel structure and another with a serial one.

Applying it

• For novices, occasional users, and experts, either build separate models for each, or at minimum validate the same model against real data for all three — testing on experts alone and assuming it applies to novices is not sufficient.
• Before doing any parallel optimization, first identify which segments of the task have been repeated frequently enough to have become automated, and assume a parallel structure only for those; treat the rest as serial.
• For a scenario dominated by novices, focus optimization on discoverability and timely feedback rather than copying an expert path to compress steps — a novice has no use for the parallel structure inside an expert's path.
• How to check: measure the same task's completion time and operation sequence with real user samples at different experience levels; if the novice sample's critical path diverges clearly from the expert sample's, the task genuinely has a skill-level boundary, and the model cannot be applied across populations unchanged.

Related

• Same group: B4.13.1 Perceptual, cognitive, and motor resources can be occupied in parallel during skilled performance · B4.13.2 Total time is determined by the critical path, not the sum of all operation times · B4.13.3 Shortening an operation off the critical path does not change total time · B4.13.5 Critical-path analysis explains why removing steps did not make the task faster
• Nearby: B4.11 GOMS Model · Q2 Usability Evaluation
• Search terms: skill level · novice behavior · automated segment · task-specific expertise