I keep coming back to a paper published in 1885.
Hermann Ebbinghaus, a German psychologist, ran a series of experiments on himself. He memorised lists of nonsense syllables and then tracked how much he could recall over time. The result was a curve, now called the forgetting curve, that showed something most teachers would rather not know.
We forget almost everything we learn, within days, unless something specific is done about it. The decay is not gradual. It is exponential at first, then flattens. By the time a week has passed, most of what was learned once is gone.
This finding is 140 years old. It’s been replicated, most recently by Jaap Murre and Joeri Dros in 2015. The general shape of the curve is one of the better-established findings in memory research. It’s the kind of result that should sit at the centre of how any serious learning environment is designed.
Most engineering programmes act like it doesn’t exist.
What Ebbinghaus actually found
The 1885 experiment is worth a careful look, because it gets misquoted often.
Ebbinghaus didn’t claim we forget at a fixed rate. He showed that the rate depends on three things: how the original learning happened, how meaningful the material was, and crucially, what happened between the first learning and the test. With no retrieval in between, the loss was steep. With retrieval, the curve flattened. Each time the material was successfully recalled, the next round of decay was slower.
This second part, the flattening, is the part that matters for programme design. The forgetting curve isn’t a verdict on the human brain. It’s a description of what happens when nothing intervenes. Intervention changes the curve.
That distinction is everything.
What the next century added
Ebbinghaus identified the phenomenon. The next hundred and forty years filled in the mechanism.
Two findings stand out.
Henry Roediger and Jeffrey Karpicke published a series of studies starting in 2006 on what they called test-enhanced learning. The setup was simple. Students studied a passage and were then either tested on it or asked to re-read it. The students who re-read felt more confident. They reported feeling like they had learned more. When tested a week later, the students who had been tested in the original session, not the ones who re-read, recalled more.
This is the testing effect, and it’s been replicated dozens of times. Pulling information out of memory strengthens the trace. Putting it back in by re-reading doesn’t, or does much less. Most students re-read because it feels productive. They confuse fluency with learning. The research is clear that those aren’t the same thing.
Robert Bjork at UCLA spent decades on a connected idea. He coined the phrase desirable difficulties to describe design choices in learning that feel harder in the moment but produce better long-term retention. Spacing study sessions out instead of cramming. Interleaving topics instead of blocking them. Using retrieval tests during learning, not only at the end. Each of these makes the experience of learning harder. Each of them, in carefully controlled studies, produces stronger retention later.
Read those two paragraphs again. The implications are uncomfortable.
The way most of us were taught to study, re-reading our notes the night before the exam, is one of the least effective strategies. The thing students avoid, namely being tested while they still feel underprepared, is the thing that actually builds durable knowledge. The cram works for the test. It doesn’t work for the career.
What a typical engineering programme does instead
Pull up the curriculum of a standard four-year B.Tech and look at how it tests learning.
Most programmes run one major exam at the end of each semester. The student studies for a few weeks, sits the paper, and moves on. The next time that material comes up, if it comes up at all, is one or two semesters later, in a course that assumes the foundation’s intact.
By the time a third-year student is asked to apply data structures to a real problem, the first-year course where they “learned” trees and graphs is eighteen months gone. The forgetting curve had its way long ago.
Why does this structure persist? It isn’t because anyone tested it against the research and found it works. It persists because it’s convenient. One exam per semester is easy to schedule. It produces a clean grade. It fits the academic calendar that exists for reasons that have nothing to do with how memory works.
The trouble is that students notice. They notice they have to relearn material every time it shows up again. They notice the syllabus they “completed” in second year is hazy by the time they need it in the third. They blame themselves. Most of them assume they’ve just got bad memories.
They haven’t. They’ve got the same memory the rest of us have. The structure they’re in is doing the forgetting for them.
What we changed when we took the research seriously
I lead programme design and delivery at Kalvium, which means I get to make this question operational rather than theoretical.
A few of the choices we have made, and the research they are grounded in.
Retrieval is built into the rhythm, not saved for the exam. Students do short retrieval-style exercises through the week, not as graded tests but as the actual mechanism of practice. A concept introduced on Monday comes back as a recall prompt on Wednesday, then again the following Monday, then in a project the following month. The spacing isn’t decorative. It’s the design choice that flattens the curve. This draws directly on what Roediger and Karpicke established.
Capstone projects start in semester two, not year four. A first-year student building a small full-stack application is being asked to retrieve and apply concepts from across that semester. The act of building is a sustained retrieval exercise. It’s harder than a written exam in the moment, which is the point. This is the desirable difficulties principle in action.
Assessments are distributed. Rather than a single end-of-semester paper, students are evaluated on continuous output, with multiple smaller retrieval points. This is operationally harder for us to run. It’s also, on the evidence, dramatically better for the student.
The cohort is small enough to interleave properly. Interleaving topics, the practice Bjork found to be more effective than blocked teaching, only works if the people running the sequence can track what each student is doing. At scale it collapses. Smaller cohorts let us hold the structure.
None of this is novel. Every one of these design choices is sitting in a research paper that’s been publicly available for years. What is uncommon is the willingness to let the research, rather than the academic calendar, shape the programme.
Where the evidence gets thin
This is the part most marketing copy about “evidence-based learning” leaves out.
Some questions are well-settled. The basic shape of the forgetting curve is one. The benefit of retrieval over re-reading is another. The general value of spacing is a third.
Other questions are still contested, and intellectual honesty requires saying so.
The optimal spacing interval, for instance, isn’t a solved problem. Different studies find different intervals depending on the material and the target retention horizon. Anyone selling you a precise spaced-repetition schedule that “just works” is overselling the evidence.
The transferability of these findings is also debated. Most studies test recall of specific facts or vocabulary. Engineering involves transferring patterns across problems, which is a harder cognitive task than recalling definitions. Whether the testing effect transfers cleanly to that kind of learning is still being worked out. We design as if it does, because the underlying mechanism (effortful retrieval) plausibly applies, but we hold that with appropriate humility.
The role of individual differences is another open question. Some students seem to benefit from spaced retrieval more than others, for reasons that aren’t fully understood. A programme that takes the research seriously has to be honest about variance, not just averages.
I think it’s important to say all of this clearly. The strongest claim we can make is that the design choices we’ve made are consistent with the best available research. We can’t claim they’re optimal. Anyone who claims that, in any educational context, is overstating what the evidence supports.
Why this matters more than it looks
The argument I’m making isn’t that Kalvium has discovered something new. The argument is the opposite.
The forgetting curve has been sitting in textbooks for 140 years. The testing effect has been replicated for two decades. The desirable difficulties research has been around since the 90s. These aren’t speculative findings on the edge of cognitive science. They’re foundational results that most engineering education has chosen, implicitly or explicitly, to ignore.
When a parent asks me whether Kalvium is doing something different, this is the honest answer I try to give. We aren’t ahead of the curve. We’re reading research that’s been available to anyone for a long time, and we’ve arranged the programme so that the research, rather than the calendar, is the constraint. That’s a less dramatic claim than most education marketing makes. It’s also defensible.
I think the more interesting question, and the one I want students and parents to start asking, is the reverse. Not “what new and exciting pedagogy is this programme using?” but “what does this programme do about the forgetting curve, the testing effect, and the spacing effect, all of which have been known for a long time?”
Programmes that have a good answer to that question are doing the work. Programmes that don’t are running on the academic calendar and hoping for the best.
Hoping, on this particular question, is not enough. Ebbinghaus made that clear in 1885.