Stages can skip and loop rather than form a strict linear sequence

What it is

The seven stages of action can begin anywhere: users see the interface before forming a goal, revise a plan immediately after performing, or return to the goal level after comparison fails. Skilled users compress most handoffs, and exploratory behavior may lack a final explicit comparison. This card follows directly from the seven stages themselves: reading only the first card, it is easy to mistake the model for a pipeline that must run in strict order — this one corrects that misreading. The seven stages are analytical dimensions, not an execution order.

Why it happens

Real action deviates from strict linearity because the goal itself can be redefined mid-action, unlike a pipeline whose starting condition is fixed once and for all. Autocomplete can merge the originally separate "plan" and "perform" stages into one motion — before the user has fully decided what to type, the input method has already laid out candidates, effectively letting the system pre-empt part of the planning stage. An error message can bounce a user straight from "perform" back to "specify," skipping the perception and interpretation that should sit in between. An unexpected search result can make a user revise their original goal on the spot, in which case "compare" effectively happens before the goal re-forms, inverting the order entirely. These jumps are possible because the seven stages are, at bottom, analytical labels used to slice behavior, not a physical sequence behavior must pass through — one stage can fully nest a smaller subloop of its own. A single "save" action, for instance, can contain three smaller subloops — finding the save button, confirming the save location is correct, waiting for the save-complete feedback — and each subloop can itself be decomposed into a full pass through all seven stages.

Where it stops holding

Nonlinearity does not make the model useless. If an analysis abandons the stable slicing the stages provide, diagnosis easily degrades into a vague "the experience is bad," losing exactly the precision the model was meant to offer — nonlinearity means the order and boundaries between stages are flexible, not that the concept of a stage can be discarded. The model also does not fit completely aimless browsing with no goal at all, since its starting point is "a goal has formed"; once that starting point is absent, the rest of the stage breakdown loses its frame of reference. For several parallel activities happening at once, the model will not automatically tell you which goal to slice the behavior sequence against. For automated agents and human-agent collaboration, it also needs to be stated explicitly whether each stage is carried out by the user or delegated to the system or agent — otherwise a diagnosis of "stuck at this step" cannot distinguish a human problem from an automation problem.

Applying it

• Preserve a re-entry point at every stage in the design: a visible place to restart, a plan that can be undone and redrafted, an operation that can be interrupted partway through, and a reviewable history of past results — letting users cut back or jump out from any point in the middle instead of being forced to redo everything from the start.
• Let users adjust parameters directly from the feedback they currently see, instead of forcing them back to a home screen to reset the goal — the goal may legitimately change because of what the feedback revealed, and the interface should accommodate that change rather than block it.
• When annotating recordings, allow one segment to map to multiple stages and allow marking loops and nesting, rather than forcing complex behavior into one single path just to keep the annotation tidy.
• How to check: test whether a user can recover smoothly after being interrupted mid-task — verify the system preserved the stage they were at, the choices already made, and the last result obtained; if recovery requires redoing stages already completed before the user can continue, the interface has implemented genuinely nonlinear behavior as if it were a linear flow.

Related

• Same group: B4.01.1 Goal, plan, specify, perform, perceive, interpret, and compare are seven stages · B4.01.3 The model locates which link contains the problem
• Nearby: H1 Interaction Patterns and Flows · I1 State, Time, and Response
• Search terms: action cycle · nonlinear interaction · exploratory behavior · subgoal loop