Research

Relevance

• A neuroscientist’s account of how memory emerges from changing connections between neurons, how the hippocampus organises experience, and what helps learning become durable.

Key findings

• Neurons connect into large networks, and repetition strengthens those connections: learning is a physical change in the brain, not just an accumulation of information.
• Stimulating the brain can elicit memories directly, and smell is the strongest and most direct trigger of human memory.
• The brain prefers multisensory experience: context helps memorisation, and context can be added artificially rather than only found.
• In-person learning is often more effective than digital learning because it supplies that multisensory context by default.
• Neuromodulators (adrenaline, serotonin and dopamine) promote learning by strengthening connections in the brain.
• They are released by attention and arousal, stress and threat response, reward and motivation, and curiosity and exploration; physical activity often activates them too.
• Familiar learning principles work by recruiting these mechanisms: reward systems, gamification, salience and cues, and emotional tagging.

Relevance

• Duolingo uses progressive disclosure in an effective way. When you start, you do not get a rundown of all the functionalities. You simply start answering questions, and as you progress, new mechanics are introduced organically.

Key findings

• The setup is framed as a bounded commitment rather than an open-ended configuration step: “Just 7 quick questions before we start your first lesson.”
• Those questions cover language, motivation, a daily time goal and starting point, and offer “Find my level” as an alternative to starting from scratch, so the entry point adapts to what the learner already knows.
• The first lesson begins with a real exercise, repeating a spoken sentence, with no walkthrough of the interface and no explanation of the app’s mechanics beforehand.
• Reward mechanics are revealed one at a time, and only after the first lesson is finished: first XP, accuracy and speed, then the streak, then the daily quest chest.
• The course path shows later units as visible but padlocked, so the remaining complexity is signalled rather than either hidden or made available.

Relevance

• After a quick onboarding you land right in a new document. Figma shows you around five core functions, then lets you explore on your own. The initial training is done through design documents, rather than videos or other separate learning material.

Key findings

• Setup is a short run of single-question screens (name, use, role, level of school, prior experience, plan, first project), and a preview beside them builds up as you answer: a cursor labelled with your name, then a file named after your choices.
• One question asks directly about prior experience (“Have you used Figma before?”), before any tutorial is shown. Later setup screens can be skipped.
• Onboarding ends in a document rather than a dashboard. Choosing “Jump right in” opens a canvas, with a template picker and an “Or Learn the basics” link placed on the file itself.
• The tutorial is itself a design file: “Figma basics” and “FigJam basics” sit in the file browser next to the user’s own files, and contain sample product pages and an area for practice designs.
• A five-step tour runs inside that file, each step anchored to the part of the interface it explains and carried out on real content: frames, text layers, the design panel, frame properties and UI kits. Steps can be stepped back through or closed.
• The tour ends by handing over control: “You’re all set!” offers “Practice more here” or “Start a new file”.
• A keyboard shortcuts panel lets you explore and see what functions you have used.
• Dev mode with green accent color and following frames are examples of ways to show the user they are in a different mode.

Relevance

• Stripe’s sandbox shows that a system with real consequences can be learned by trying things out, as long as the practice space is clearly separated from the real one. OpenBridge could do the same: a training mode that feels like the real interface, where mistakes have no consequences and it is always clear that you are training.

Key findings

• Setup asks a few questions about the business (name, location, website, what it does) and turns the answers into a recommended configuration, here accepting online payments, sending invoices and collecting tax, with the note “You can always change your setup later.”
• Before reaching the Dashboard, a screen presents the sandbox as “a safe environment for testing” where you “can try out features without any real money movement”, and lays out the path ahead in three steps: complete your setup, switch to a live account, copy your work over.
• Inside, a banner stays at the top of every page: “You’re testing in a sandbox. Changes you make here don’t affect real customers or payments.” A button to switch to the live account sits beside it.
• The first view in the sandbox points to a setup guide rather than to documentation: a checklist grouped by goal (set up payments, invoices, verify your account, tax), with a progress bar.
• Concepts are introduced as choices when you reach them rather than explained up front. Choosing how to accept payments offers payment links, a prebuilt checkout form or a custom flow, each with a one-line description and a link to compare them.
• Tasks create real objects in the sandbox: adding a product opens a form with a live preview that estimates the total from the pricing model, quantity and tax.

Relevance

• TaskLens is a design project that seeks to explore some of the same ideas as OpenBridge: unifying instruction and interaction into one layer instead of a separate manual or tutorial. They found that novices benefit from the staged guidance, but experienced users feel boxed in by it and route around it. This is something a training layer in OpenBridge should account for. Users already competent in the system need an easy way around the training, not just a way in.

Key findings

• TaskLens automatically builds a simplified version of a software's interface for one specific task. You describe the task in plain language, and an AI breaks it into steps, picks out only the tools needed for those steps, and generates a custom control panel inside the software.
• The idea is to put instructions and the tools themselves in the same place, instead of a separate manual or tutorial you have to keep switching between.
• In a test with 32 beginners using Blender, people using the simplified interface felt less overwhelmed and finished tasks faster than people using the normal interface: one task took about 12.7 minutes instead of 15.1, another took about 20.4 minutes instead of 25.
• Beginners using the simplified interface also understood the underlying concepts better, watched fewer tutorial videos, and explored more tools on their own.
• A separate test with 8 experienced users was more mixed. They liked having less clutter and appreciated the built-in explanations, but some felt boxed in by the fixed steps, since expert workflows aren't usually that linear. A few gave up on the simplified interface partway through and just used keyboard shortcuts instead.
• The researchers note some limits of their own work: the difficulty level is chosen by the user rather than adjusted automatically based on how they are doing, it has only been tested in one piece of software (Blender), and it does not work well for open-ended, exploratory tasks that cannot be broken into clear steps.

Relevance

• They concluded that generic interfaces and guidelines across systems support learning, and reduce the amount of training needed. This is especially interesting in connection with OpenBridge, which lays such a foundation for learning.

Key findings

• Progressive disclosure through a multi-layered interface defers advanced or rarely used features to a later stage, so a novice starts on a reduced-functionality layer and moves up once competent.
• Tested on 42 military users on a real UAV mission-planning tool, comparing a layered interface (unneeded features greyed out) against the full interface.
• No statistically significant difference in task performance or awareness of features between the two groups: individual differences between users were the dominant factor and masked any effect of the interface itself.
• Even so, the layered group showed more consistent performance across users, and people who scanned menus methodically found functions later more reliably than those who stumbled onto them by luck.
• Because individual differences dominated, the authors argue for a customised approach (matched to a user's own exploration style) rather than one layered design for everyone. Progressive disclosure suits quick learning and infrequent or multi-application use; a full interface paired with explicit search-strategy training suits users who need durable, deep recall.
• Their stated conclusion: implementing generic interface design heuristics, guidelines and principles across systems supports learnability and should reduce training demand.

Relevance

• Not all believed “truths” about learning are correct when studied closer.

Key findings

• Ericsson, Krampe & Tesch-Römer's (1993) famous violin study concluded that accumulated deliberate practice alone accounts for expert-level performance differences.
• In their replication study, Macnamara & Maitra did not reproduce these results, and found that accumulated practice did not correspond to skill level the way the original study reported.
• They found an effect of practice, but it was smaller than in the original study (η² = 0.26 here versus η² = 0.48 originally). Practice still matters, but it is less decisive than the original claim suggested.
• Teacher-designed practice was also found to be less relevant to improving that popular belief.
• The authors conclude that deliberate practice cannot, on this evidence, account for why some individuals reach the highest levels of expert performance and others do not. Amount of practice is not the whole story.

Relevance

• The article takes the findings of Ebbinghaus and explores how spaced repetition should actually be applied to enhance retention. This concrete application of the principle could be valuable to implement in the OpenBridge training interface.

Key findings

• “The spacing effect is one of the oldest and most reliable findings in research on human learning” (Carpenter et al., 2012, p. 2), with early demonstrations going back to Ebbinghaus (1885/1913).
• Retention improves when study sessions are spaced apart in time rather than massed in immediate succession: the same total study time, scheduled differently.
• The optimal gap depends on how long the knowledge has to last: roughly 10–20% of the delay before it is needed. Tested after 7 days the best gap was 1 day; after 35 days, 11 days; after 70 days, 21 days.
• A longer gap is not automatically better. Wait too long and forgetting offsets the benefit, so there are diminishing returns rather than a single best interval.
• Material that has been learned and then forgotten is relearned far faster than material met for the first time, so regular re-exposure keeps knowledge cheap to restore.
• Spacing benefits diverse kinds of learning, but the authors note it has still not been systematically built into curricula: the research has not produced clear enough instructions for practice.

Relevance

• Often times learning conditions that feel difficult helps improve learning, but learners don’t tend to notice this effect themselves. How can we implement desirable difficulties without hurting the confidence of the learners?

Key findings

• Performance (that can be observed during practice) and learning (the durable change the practice is meant to produce) are different things, and current performance is often an unreliable index of whether learning has actually occurred.
• Bjork & Bjork call this difference retrieval strength (how accessible information is right now) and storage strength (how durably it is learned).
• The conditions that make performance improve quickly during practice often fail to support long-term retention and conditions that slow apparent learning down often optimise it.
• Examples of these “desirable difficulties” can be:
  • Varying the conditions of practice, rather than keeping them fixed and predictable, improves later recall and transfer. For example studying material in two different rooms beats studying it twice in the same room.
  • Interleaving different topics or skills rather than practicing one at a time (blocking), produces worse performance during training but far better long-term retention and transfer.
  • Producing an answer yourself (The generation effect), rather than being shown it, produces more durable learning than being presented the same information.
  • Testing or attempting to retrieve knowledge, even with no feedback, is often more effective for long-term retention than rereading the material.
• Despite this proved effect, learners often report that they thought they performed better during the easier conditions.

Relevance

• The article proves testing as an effective strategy for learning. This is something that a training interface should incorporate.

Key findings

• The “testing effect”: being tested on material has a greater positive effect on retention than spending the same time studying or re-reading it.
• In the comparison, one group studied the material twice while another studied it once and then took a test.
• Retention was then measured at three intervals (five minutes, two days and one week) rather than immediately after studying alone.
• After five minutes, repeated studying produced better recall than testing; but after two days and one week, the tested group retained substantially more, even though repeated studying had left students more confident they would remember.

Relevance

• Situation awareness is arguably the central design goal for any bridge interface, so a definition this precise is worth having on hand before deciding what training should actually build toward.

Key findings

• Situation awareness (SA) is the perception of the elements in the environment within a volume of time and space, the comprehension of their meaning, and the projection of their status in the near future.
• The article asks what shapes situation awareness, and how this affects the decisions made in a workplace.
• Endsley's model describes three levels of situation awareness.
• Level 1 (perception): perceiving the essential properties of the task environment, either directly through the senses or mediated through digital sensors. This is the raw data: what you see, hear, or read off a screen.
• Level 2 (comprehension): combining the elements from level 1 with your own knowledge to build an understanding of the situation. In other words, not just registering data, but knowing what it means for your goals.
• Level 3 (projection): using the elements from levels 1 and 2, together with knowledge of the situation's spatiotemporal aspects, to forecast a future state or consequence. This is the highest level, and the one that makes operationally useful decision-making possible.

Relevance

• The article distinguishes between difficulty inherent in the material and difficulty imposed by the way the material is designed and presented. For complicated and highly interconnected material, such as the material OpenBridge is often used to communicate, this distinction is particularly relevant. The findings suggest that instructional design should minimise unnecessary cognitive load where possible, allowing learners to focus their limited working-memory capacity on understanding and learning the essential information.

Key findings

• Learning is described as the acquisition of schemas that organise knowledge and can eventually become automated through practice.
• Learning places demands on working memory, which has limited capacity. Cognitive load therefore affects how easily new knowledge can be acquired.
• Sweller distinguishes between intrinsic load and extraneous load:
  • Intrinsic load is determined by the complexity of the material itself. It is closely related to element interactivity: the number of elements that need to be understood and processed together. Because of this, material with many interacting elements is inherently more difficult to learn.
  • Extraneous load is caused by how the material is presented and can be influenced through instructional design.
• The effect of minimizing extraneous load depends on the complexity of the material. For difficult material with high element interactivity, unnecessary cognitive load can interfere with learning, but when element interactivity is low, reducing extraneous load may have little effect on learning.
• The split-attention effect shows that when related information is presented separately, learners may have to mentally integrate the different sources. This creates unnecessary cognitive load. Presenting related information together can therefore reduce extraneous load.
• The redundancy effect shows that more information is not necessarily better. When information is already sufficiently explained by itself, adding text that simply repeats the same information can create unnecessary cognitive load.

Relevance

• The paradox names exactly the failure mode a training layer inside OpenBridge would have to survive: an operator who is motivated to get the task done will route around anything that reads as instruction, no matter how well designed.
• “Training wheels” and the “Minimal Manual” are the earliest concrete examples of training built into the interaction itself rather than delivered beside it: the closest precedent here to what this course is trying to design.

Key findings

• People taking up a new system are motivated and want to get something done, not passive or lazy, so they skip the manual and the introduction and go straight to the task, and so never make the small up-front investment that would make them far more effective later.
• Production bias: the user’s goal is throughput, not learning. Anything that looks like a detour from getting the task done, including instruction that would pay off later, gets skipped.
• Assimilation bias: users interpret a new system through what they already know. Useful when the old knowledge transfers, but misleading when it doesn’t: irrelevant similarities can blind people to what the new system is actually doing, or stop them from recognising a feature that doesn’t map onto anything familiar.
• Carroll and Rosson argue these biases are not defects in the learner to be corrected, but fundamental, mutually reinforcing properties of how people learn, which is why they treat them as paradoxes needing a design tradeoff, not a “bad design” problem to be eliminated outright.
• Their “training wheels” interface shows every function of the full system but structurally disables advanced, error-prone choices during early learning, so a likely mistake is blocked outright rather than merely discouraged or explained.
• Their “Minimal Manual” design works with the production bias instead of against it: instructions are deliberately left incomplete, with “On Your Own” sections that push the learner to apply what they just did to a new problem of their own choosing.

Key findings

• New users in advanced systems often become frustrated and confused by the errors they make in the beginning of training.
• The article explored whether removing parts of the interface that often trigger “error states” when used incorrectly, during training, could make the training more effective.
• This training environment had only the basic functions of the system, which enabled the user to learn the most important things significantly faster.
• In the published study, the control group spent almost a quarter of their time recovering from the error states that the training interface blocked off, and the training group also scored better on a comprehension post-test.

Relevance

• Ebbinghaus (german psychologist) discovered the forgetting curve and the spacing effect, contributions to the study of memory that laid the foundation for the study of human learning and memory.

Key findings

• Ebbinghaus conducted experiments on himself.
• He discovered that the amount of information retained decreases over time, with the steepest drop occurring immediately after learning.
• His research showed that spacing out study sessions over time, rather than cramming, improves long-term memory retention.