When should feedback arrive in a learning experience?

Every scenario designer decides, usually without noticing, when feedback arrives: at the choice, at the end of the path, or at the end of the module. The memory literature has an opinion, and it also has a surprise. The surprise is that the error itself is not the problem. Errors that are made and then corrected are among the best-documented boosters of learning. The problem is what happens when the correction is late, vague, or absent. This guide works through that finding and its consequences for timing, and it ends where the medium pushes it: toward correcting at the moment of choice.

The surprise: a corrected error beats an avoided one

Error avoidance looks like common sense. If you do not want errors on the test that counts, keep them out of the learning. Janet Metcalfe's review of the field, Learning from Errors in the Annual Review of Psychology, reports that this common sense is the rule in classrooms and that the laboratory evidence runs against it. For neurologically typical learners, errorful learning followed by corrective feedback is beneficial to learning. In tightly controlled studies, learners who guessed and generated a wrong answer before being shown the correct one remembered the correct answer better than learners who simply read it, provided the guess was related to the answer rather than a shot in the dark (Metcalfe, 2017).

That reframes a common worry in scenario design. A decision point with plausible, tempting suboptimal options can feel like a trap you are setting for the learner. It is closer to the opposite. The suboptimal option a learner is drawn to, and then has corrected, is the mechanism doing its work. The condition Metcalfe is careful about is the correction: a verdict alone, correct or not, gives almost no benefit. The feedback has to supply the reasoning behind the stronger option, not just register which one the learner picked.

Confidence is what makes the correction stick

There is a further finding that speaks directly to how decision options should be written. The benefit of correction is strongest when the learner was confident in the wrong answer. Metcalfe calls this the hypercorrection effect: high-confidence errors are corrected more readily than errors made with little conviction. The leading explanation is surprise. Being wrong when you were sure you were right captures attention and pulls it onto the correction (Metcalfe, 2017).

For a designer, that is an argument for options that a competent learner could genuinely believe in. A wrong option nobody would pick teaches nothing when it is corrected. A wrong option that feels right to someone acting on a common misconception sets up exactly the confident error whose correction sticks. The craft of building those options is its own subject, covered in Designing Decision Points. The point here is that the timing of the correction and the plausibility of the option work together: a confident error is a brief opening, and the correction has to arrive while it is open.

Why correct at the moment of the decision?

The most useful finding for a designer is this: naming the wrong idea a learner is drawn to and correcting it beats presenting the right idea cleanly. A correction at the moment of choice does exactly that, because the wrong idea has just been voiced by the learner's own decision.

So correct at the time of the decision. That is when the learner is still thinking about the choice they made, and delivering the expert rationale right then is when it lands. Wait until later and the reasoning that produced the choice moves out of reach: the learner moves on, the situation changes, and by the outcome that reasoning is a memory rather than still fresh in mind.

The evidence points to the same form. When Muller and colleagues tested four online treatments on Newtonian mechanics with 364 first-year physics students, the two that worked voiced the common misconception and corrected it, producing learning gains with effect sizes of 0.79 and 0.83 over a clean exposition of the same correct material; learners with low prior knowledge benefited most, and those with high prior knowledge were not held back (Muller, Bewes, Sharma, & Reimann, 2008). A verdict on its own is nearly worthless. Correcting at the moment of choice supplies that rationale exactly when the learner can still use it.

Where end-of-path feedback comes from

End-of-path feedback is not an arbitrary habit. It is inherited from the live exercise that digital scenarios descend from. In the live version, you gather people, run the simulation, and they make a stream of choices; afterward a facilitator leads a debrief that walks through what went right and what went wrong. That debrief works, and it works because it is a live conversation. People can ask why, push back, and hear the reasoning answered in real time.

The digital translation of that debrief is paragraphs of after-the-fact text. It talks at the learner about what they did and what they should have done instead, detached from the moments where the choices actually happened. It arrives after the fact, and it does not land the way a facilitator in the room does. The problem compounds as a design branches: a learner who has meandered through many forks cannot easily locate where they went wrong, because the correction sits at the end of a path rather than at the decisions that produced it. This is the structural difference examined in Branching vs. Guided Scenarios. The relevant consequence for timing is simple. What a room full of people and a facilitator accomplish through live conversation, a digital experience has to accomplish another way, and correcting at the moment of choice is that way.

The repetition alternative, and why training rarely gets it

There is one setting where the end-of-path problem takes care of itself: massive repetition. Branching combined with heavy replay genuinely works. Video games are the proof. Failing repeatedly is the fun, the replay costs nothing, and a player will run the same encounter dozens of times until the optimal path is obvious. Correction is not needed at each choice because the sheer volume of attempts surfaces what works.

Training has neither the time nor the appetite. Replaying a workplace scenario enough times to discover the optimal paths is possible, and a motivated learner occasionally does it, but almost nobody invests the time. The repetition model that rescues branching in games is not available to a program a learner runs once or twice. So the question is not whether replay could solve the timing problem in principle. It is what to do given that the replays will not happen.

The design argument for correcting at the choice

Here is the position, offered as a design argument rather than a research finding about branching. A learner who rides a wrong path to its ending, and only later meets a better one, comes away holding two experiences of the same situation with roughly equal weight. Nothing about the first experience marks it as the wrong one; it played out, it had an ending, and it is now a memory alongside the other. Retrieving which of the two was right, later and under pressure, becomes its own problem. What reliably builds recognition of the optimal choice is not exposure to both, but repetition paired with correction: corrected encounters accumulate until the pattern of the right choice is easy to recognize.

There is also a memory mechanism in the background. New learning that is similar to earlier learning can disturb the earlier memory while it is still consolidating. Sosic-Vasic and colleagues found that a second, similar learning task in the minutes after a first one reduced memory for the first by up to twenty percent, and that similarity between the two was what gave the interference its bite (Sosic-Vasic, Hille, Kröner, Spitzer, & Kornmeier, 2018). That is a general, well-documented effect, and while it is not a study about branching scenarios, it does suggest a downside of two competing scenario experiences. The lesson is preventive. Correcting at the moment of choice avoids creating the competing experience in the first place. When the wrong reasoning is voiced by the choice, corrected on the spot, and revised in a single move, there is one consolidated memory of the situation rather than two that later have to be told apart.

The approach, and where AliveSim sits

The approach this all points to is a learning loop: put the learner in a situation, let them commit to a real decision, and correct the reasoning at the moment they commit, while it is still theirs to inspect. The correction names what was tempting about the suboptimal choice and why the stronger option is better, which is the form of correction the evidence rewards. Done at every decision, it turns a scenario into a series of confident errors met by timely corrections, which is close to a direct reading of what the errors literature recommends. The mechanics of writing that correction are covered in Decision-Point Mentoring, and the question of whether a well-timed loop also feels more engaging is taken up in What Does 'More Engaging' Actually Mean?.

AliveSim's Guided Scenarios are built around this loop. The learner enters a realistic situation, chooses among plausible options, and receives expert mentoring at the moment of each choice rather than a text debrief at the end. That structure is the reason a Guided Scenario can correct the reasoning while it is still present, and it is what AliveSim's approach to bridging knowledge to performance puts into practice. The timing is not a feature added on top of the design. It is the design.

References

• Metcalfe, J. (2017). Learning from errors. Annual Review of Psychology, 68, 465–489.
• Muller, D. A., Bewes, J., Sharma, M. D., & Reimann, P. (2008). Saying the wrong thing: Improving learning with multimedia by including misconceptions. Journal of Computer Assisted Learning, 24(2), 144–155.
• Shute, V. J. (2008). Focus on formative feedback. Review of Educational Research, 78(1), 153–189.
• Sosic-Vasic, Z., Hille, K., Kröner, J., Spitzer, M., & Kornmeier, J. (2018). When learning disturbs memory: Temporal profile of retroactive interference of learning on memory formation. Frontiers in Psychology, 9, 82.

Published July 18, 2026 · 9 min read