Interaction can be modeled as repeated translation between the user's language and the system's, with loss possible at every step

What it is

The interaction framework splits one interaction into two translation passes: the user translates intention into action (the execution side), and the system translates internal state into a readable display (the evaluation side). The wider the gap between intention and action, or between display and interpretation, the more is lost—these are the gulf of execution and the gulf of evaluation.

Why it happens

Loss occurs at each language conversion: to "brighten this photo," the user must find the right control and understand its meaning (first translation); the system renders the brightness change as a slider position and thumbnail (second); the user reads the display back against the original intent (third). A terminology mismatch, missing feedback, or ambiguous cue at any link pulls the final interpretation away from the intent.

Studying it

Run task analysis along the intention-action-display-interpretation chain: for a given goal, log where users pause, err, or seek help on each translation. In usability tests, "can't find the entry" marks execution-side loss and "can't read the result" marks evaluation-side loss. Cognitive walkthroughs ask stage by stage whether users can name the next action and interpret the feedback, systematically locating the failures.

Where it stops holding

The framework is descriptive, not an error-rate predictor, and it assumes users act with a clear intention—it explains exploratory browsing and aimless use poorly. Gulf size is relative to a user population and task context; no interface is simply "high-gulf" in the abstract.

Applying it

• In reviews, walk both translation passes: can the target user get from intent to action, and from display back to outcome?
• Align terminology with user language: name controls in task vocabulary, not in translated system concepts.
• Give every operation interpretable feedback so the evaluation pass requires no memory or guesswork.
• In tests, record how users restate the task and diff it against interface vocabulary term by term; each mismatch is a loss point.

Related

• Same group: B4.17.2 State transition diagrams can enumerate a UI's reachable states and illegal transitions · B4.17.3 Combinations of parallel controls explode the state count; exhaustive modeling only works locally · B4.17.4 A mode is one operation yielding different results in different states; mode errors can be located on the state graph · B4.17.5 Unreachable states and dead-end states are defects that the model detects mechanically
• Nearby: B4.10 Direct manipulation · T1 Interface Copy and Content
• Search terms: gulf of execution · gulf of evaluation · interaction framework