Everything examinable, in one sitting.

Tutorial material is examinable too, but the Figma operations are explicitly excluded. There is a reference page on ergonomics at 05 · Ergonomics — kept for the final exam, clearly not part of this paper.

Recurring archetypes across the four papers: draw an STN from a prose scenario (ATM, login CLI, toy camera, smart speaker) · Fitts' law parameters, and whether the law survives in 3D or microgravity · characterise a user's problem-solving (strategy + representation level) · explain a perception or typography failure · find the NLP failure in a dialogue log. Only the first four are in scope now; the NLP-failure archetype survives in a week-7 form (tokenisation, coreference, parsing), not in its speech-act or Grice form.

1. Usability engineering is software engineering. 63% of large projects overrun cost; the four user-related reasons are requested changes, overlooked tasks, users not understanding their own requirements, and poor user–developer communication. Pay a little now or a lot later.
2. HCI is the design, implementation and evaluation of interactive computing systems for human use — an input language for the user, an output language for the machine, and a protocol for interaction.
3. AA965 (Cali, 1995) typed ROMEO instead of ROZO and got no feedback on the chosen beacon: officially “pilot error”, actually poor HCI. Three Mile Island: a lamp said “valve closed” while it was open, 1,500 alarms, an on/off silencer.
4. Pioneer systems innovate but are commercially unviable (Xerox Star, $15,000); settler systems make the ideas succeed later (Macintosh, $2,500, developer toolkit, interface guidelines).
5. Fitts' law: T = a + b·log₂(2D/W). Difficulty rises with distance and falls with target width; ID belongs to the task, a and b belong to the device; IP = 1/b.
6. Well-defined problem = full specification of data, actions, strategies and goal states, representable as a problem space. Ill-defined = partial specification, e.g. a mission statement.
7. Difference reduction moves toward the goal one visible step at a time and gets trapped in local maxima; means–end analysis accepts sub-goals that temporarily increase the distance.
8. Colour is not a property of light; it is the eye's interpretation of frequencies. 3D is not a property of the retina; it is constructed from fixations, memory imputation, motion parallax and optic flow.
9. Gestalt: similarity, proximity, symmetry, continuity, closure, smallness, surroundedness. Patterns are constructed, not recognised — which is why a Dalmatian appears in dots and why a model can fail on the same image.
10. CRAP — contrast, repetition, alignment, proximity — plus unity and grids. One to two typefaces, two weights, two slants, four sizes, ≥2 pt apart; sans-serif on screen, ≤60 characters per line in print.
11. Dialog levels: lexical = shape of the control; syntactic = order of inputs and outputs; semantic = effect on the application's data. A good answer names the level before naming the fix.
12. STN: circles are states, arcs are actions/events, labels are action / condition. Draw escapes once at a meta level, or the diagram becomes spaghetti of duplicated behaviour.
13. NLP pipeline: tokenisation → POS tagging → parsing (parse tree) → semantic analysis → coreference resolution. Errors cascade, so accuracy drops with sentence length. Tokenisation is the hard stage across languages: Chinese has no spaces, Turkish packs a clause into one word.
14. Ambiguity is the reason all of this exists. The 121 most frequent English nouns average 7.8 meanings each; one ordinary sentence yields ~14 quadrillion possible interpretations — and speakers do not notice the ambiguity at all.