Readings

Where the thresholds that force a waiting interface to exist come from — about one second before a user notices a delay, about ten before they lose the thread. This week's case sits well past both.

The standard argument for progress bars. Read it as the position the rest of this course argues with: it is about how waiting feels, and says very little about whether the indicator is telling the truth.

A real upload lifecycle written by the people who operate it: initiate, upload parts, complete, and the states in between where a part exists but the object does not. Read the Multipart upload process section.

Worth reading before you write your own. The platform already has a determinate/indeterminate distinction in one attribute, and it maps onto the state question more closely than most people expect.

Two paragraphs holding the whole of this week's question: the platform hands you a boolean saying whether a denominator exists, and total is meaningless when it is false.

What a browser actually reports during an upload, at what granularity. Note what is not in it: nothing about what the server does after receiving the bytes.

Re-read the value attribute paragraph specifically. Omitting value is how the platform spells indeterminate, and it is deliberately not the same as value="0".

A widely used indeterminate treatment, argued for on perception grounds. Read it against this week's constraint and decide whether a skeleton admits a fractional reading — the article does not ask that, which is why it is worth arguing with.

Read the Basic (simple) exponential smoothing section only. The one-line recurrence is the whole of the third estimator, and the article is explicit about what the smoothing factor assumes.

Search for --progress-bar. A thirty-year-old progress meter shipping on millions of machines, documented unusually bluntly about what the display can and cannot promise. Worth seeing a design that chose austerity.

The researchers' own summary of the perceived-duration experiments, including effect sizes. Read the numbers rather than the headline: notice how large a difference had to be before it was detectable, and compare that with what you could detect today.

Re-read in week 6. These thresholds are an order of magnitude larger than the perceived-duration effects above — a useful sense of proportion before deciding how much easing is worth.

The required and optional attributes, and what a missing aria-valuenow means. Audit line 2.

For when the number needs words around it — which, on a staged upload, is most of the time.

The success criterion behind audit line 3. Read the Intent section; the examples are a good model for what counts as a status change worth announcing.

The mechanics for audit line 4. Note that no-preference is a separate value from the query not matching.

A timeout is a time limit imposed on a user, and this is the criterion that governs it. The exceptions matter as much as the rule.

When a failure warrants an assertive announcement rather than week 7's polite live region. The guidance on not stealing focus is the part that applies here.

The platform's own timeout mechanism, and useful before week 9 — it is the same object that will implement your Cancel.

The platform's cancellation primitive. Read what abort() guarantees and, more importantly, what it does not: it stops the client listening, which is not the server stopping work.

The Aborting a multipart upload part specifically. A real service documenting that parts persist and cost money until explicitly aborted — precisely the failure the Cancel guarantee is written to exclude.

Short and pointed for this week: a meter reports a measurement, a progressbar reports progress toward completion, and using one for the other misstates what the value means. Check your register for a value that is really a meter.

Read this if any of your rows traced back to a timer. It documents what a clock reading is, and why it is not evidence that work occurred.

The standard argument for small-n usability testing. Read it critically and note the scope of the claim: it is about the share of findable usability problems surfaced, not about how often anything happens in a population — exactly this week's existence-proof-versus-rate line.

Re-read after week 6. Compare the participant numbers and controls in that work against what you ran today, and you have the size of the gap between an experiment and an afternoon.