Attention

Cognitive load

Working memory has a small budget, and interface noise spends it before the task does.

cognitive load: How much a person has to hold in mind at once to get through a task. theory splits the demand on working memory into three parts: intrinsic load from the difficulty of the task itself, extraneous load from how the material is presented, and germane load from the effort of building a durable mental model. Only extraneous load is pure waste. Design work is mostly about cutting extraneous load so the budget goes to the other two.

How it shows up in software

Every field, label, tooltip, banner and modal spends from the same budget. The split attention effect appears whenever a person must hold information from one part of the screen to understand another: an error message far from the field that caused it, or a help panel that covers the form it explains. Redundancy is a cost too, since restating the same thing in two formats makes people reconcile them.

Using it well

Put help where the difficulty is. An inline hint beside the field beats a paragraph at the top or a link to documentation.
Place errors on the field that caused them, in words that say what to do next.
Strip redundancy. If a label, a placeholder and a tooltip say the same thing, delete two.
Accept intrinsic load where the task is genuinely complex, and spend your effort on removing presentation noise instead of oversimplifying the task.

Where it turns manipulative

Hiding the terms that govern a purchase behind a link and calling it load reduction. Simplification that removes information people need to decide is concealment.
Raising extraneous load deliberately on the path you want unchosen, such as a cancellation flow that adds screens and surveys.
Removing power features because a first-time user finds them heavy, which offloads all the complexity onto the people using the product daily.

Where you have seen it

Stripe Checkout
Validation appears inline against each field as it is completed, so a correction never requires reconciling a list of errors at the top with the fields below.
Linear
Issue creation shows a small set of fields with the rest behind an explicit expand, keeping the common case low load without removing options.
Google Flights
Filters update the result set in place, so the person does not hold prior results in memory to compare against.

What the research says

Sweller, 1988Well evidenced
Learners solving problems by means-ends analysis learned less than learners given worked examples, because the search consumed working memory that was not available for learning.
Sweller, van Merrienboer and Paas, 1998Well evidenced
Assembled the split attention, redundancy and worked example effects into a single account of instructional design under working memory limits.
de Jong, 2010Mixed evidence
A critical review noted the three load types are hard to measure independently and are often inferred from the outcome they are meant to explain.
The design implications hold up better than the three-way split does as a measurement framework.

Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.