Comprehension
Can an agent understand content, context, and constraints?
Content carries explicit meaning, units, relationships, and rules, so an agent does not have to infer them from layout or tone.
Humans read context from layout, imagery, and convention. A strikethrough price next to a red badge means a discount. Small grey text below a button probably contains a condition. Agents can sometimes infer these cues, but every inference is a chance to be wrong.
Comprehension asks whether the meaning a designer intended survives the trip to a machine. Prices need currencies. Dates need time zones. Availability needs a definition. Constraints like "ships to EU only" or "requires an annual plan" need to be stated as data, not implied by a footnote.
Comprehension also covers state: what the agent is looking at right now, how it relates to the user's goal, and what has changed since it last looked.
What good looks like
Signals
- Key facts are available as typed, structured data alongside human-readable copy.
- Units, currencies, locales, and time zones are explicit.
- Constraints and eligibility rules are expressed as data, not fine print.
- Domain terms are defined once and used consistently.
- Current state (cart, session, workflow step) can be read directly.
Anti-patterns
- Meaning carried only by color, position, or iconography.
- Prices without currency, or quantities without units.
- Critical conditions hidden in tooltips, images, or modals.
- The same concept named differently on different pages.
Evaluation checks
Use these questions when reviewing a task. They feed the AX evaluation.
- C1Are prices, dates, and quantities machine-readable with explicit units?
- C2Are eligibility rules and constraints exposed as data?
- C3Can an agent read the current state of a session or workflow without rendering the UI?
- C4Is terminology consistent between the interface, the data, and the documentation?