I want to start with a design question that sounds simple and isn’t: if you built a system whose entire point was to train students to own their own learning, would you grade it.
My first instinct, years ago, would have been yes. Grade everything. Grading is how you signal that something matters, or so I believed. It turns out that instinct is close to backwards for this particular case, and the reason why is one of the more interesting pieces of motivation research I have come across doing this job.
Autonomous Learning is a named system at Kalvium, sitting alongside HEROS, Live Books, and DOJO in the programme’s set of four core internal systems. It runs across all four years. And unlike the other three, it is deliberately not graded the way a subject is graded. This post is about what it actually trains, and why leaving it outside the grade book is the design decision, not an oversight.
What Autonomous Learning is actually built to train
Here is the plain description first. The system is built to teach four things. Solve real problems as part of everyday learning. Learn how to learn beyond marks and syllabus. Take ownership of projects and your own technical growth. Face challenges that build independence and problem-solving ability.
Read those four together and a pattern shows up. None of them describe a subject with content to be tested. All four describe a disposition, a way of relating to your own work, that either shows up or it doesn’t. You can test whether a student knows how a hash map works. You cannot meaningfully test, with a single exam, whether a student has internalised the habit of going and figuring something out before asking for help.
That distinction is the whole post. Sit with it for a second. A subject has a syllabus with a defined boundary. Autonomous Learning has no boundary by design, because the point is what a student does when nobody has defined the boundary for them.
Why grading it would work against the goal
Here’s where the research earns its place in this piece.
In the early 1970s, the psychologist Edward Deci ran a study that has been replicated and built on for over fifty years since. He gave two groups of people the same interesting puzzle. One group was paid for solving it. The other wasn’t. When the paid group’s payment stopped, their interest in the puzzle dropped below where it started. The unpaid group kept engaging with it during free time at roughly the same rate throughout. The finding got a name: the overjustification effect. When you attach an external reward to something people already found worth doing for its own sake, the external reward can crowd out the internal reason. The activity often doesn’t survive the reward’s removal.
Deci, working later with Richard Ryan, built this into a broader account called self-determination theory. Their claim, refined over decades of follow-up research, is that sustained motivation depends on three things being genuinely present. Autonomy is the sense that you’re doing something because you chose to. Competence is the sense that you’re capable of doing it well. Relatedness is the sense that it connects you to people who matter. Of the three, autonomy is the one most directly at stake in how you assess something.
Apply that to Autonomous Learning specifically. The entire point of the system is to build the internal sense that a student is directing their own growth, not performing growth for someone grading them. If you attach a score to that, you introduce exactly the external justification the research says can crowd out the internal one. Students start optimising for what earns the mark, which is a controlled, externally-directed behaviour, precisely the opposite of what the system is meant to build. You would be grading students into compliance on a system whose entire purpose is the absence of compliance as the driver.
That’s not an argument against grading in general. Essentials, the AICTE-compliant subjects, are graded through university exams, and that’s correct, because Essentials content genuinely has a defined boundary that exams can test fairly. The argument is narrower: for this specific system, grading the exact behaviour you’re trying to build would defeat it.
What Autonomous Learning is not: the difference from Mastery
A clarification is worth making here. It would be easy to conflate this with something else Kalvium also keeps outside the exam system: the Mastery layer.
Mastery is verified through codebases, deployments, and system artifacts rather than written tests, and I’ve written about that decision at more length elsewhere. But the reason there is different. Mastery is kept off exams because of a measurement problem: a written test cannot tell you whether a deployed system actually runs under real conditions, so the assessment has to be the artifact itself.
Autonomous Learning’s reason is closer to a motivation problem than a measurement problem. It’s not primarily that a test can’t capture the skill. It’s that testing the skill would change the skill. A student can demonstrate coding ability on a fair written or practical test without that test undermining the ability itself. Autonomy doesn’t work the same way. The moment you turn “direct your own learning” into “direct your own learning, worth 10 marks,” you’ve changed what you’re measuring. You’re now measuring compliance with a new instruction, not the disposition the instruction was trying to capture.
These are two different systems, solving two different design problems, that happen to land on a similar answer: keep it out of the exam structure. It’s worth knowing they arrive there for different reasons, because conflating them would make the reasoning sound sloppier than it is.
So how does anyone know if it’s working
Fair question, and the honest answer is: not through a single number.
Live Books capture real project work, assignments, and hands-on tasks as they happen across the four years. That gives mentors a running record of what a student actually chooses to pursue, not just what was assigned to them. That record accumulates. A mentor looking at six months of a student’s Live Books entries can see, in a way a single test cannot show, whether initiative is a pattern or an anomaly.
HEROS runs underneath this, tracking learning activity across every session and milestone and flagging gaps early. Disengagement from self-directed work tends to show up as a gap in activity before it shows up as a missed deadline. HEROS is designed to catch it at the earlier point, so a mentor can step in with support rather than waiting for a term-end review to notice.
There’s also a Learning How to Learn course in the first semester, which matters more than it might sound like on paper. Four years of Autonomous Learning assumes students already have some vocabulary and technique for directing their own study. Most students entering a CSE programme straight from a school system built entirely around external structure, marks, ranks, defined syllabi, don’t arrive with that vocabulary automatically. Semester 1 doesn’t assume they do. It teaches it explicitly before asking students to run on it for four years.
None of this replaces the discipline layer, to be clear. Professional behaviour, being on time, meeting deadlines, working well in a team, is graded separately, and it should be. Showing up is a baseline expectation, not an optional autonomy exercise. What stays outside grading is narrower and more specific: the content of what a student chooses to pursue on their own initiative. Accountability for whether the system is working moves from a single score to sustained observation over time. That’s a slower signal than an exam mark. It’s also a more honest one for what’s actually being measured.
Where the evidence gets thin
I should be straightforward about the limits of what I’m claiming here.
Self-determination theory is well-established and has held up across a large body of replication since Deci’s original 1971 study, including in educational settings specifically. The general claim, that external rewards can undermine intrinsic motivation for tasks people already find worth doing, is about as solid as motivation research gets. What’s less settled is the specific implementation question. Is the exact mix of Live Books, HEROS monitoring, and mentor observation Kalvium uses the optimal way to sustain autonomous behaviour without a grade attached? That’s a design bet, calibrated through cohorts rather than proven in advance, and I hold it with real but not absolute confidence.
The honest version of the claim is this. The decision not to grade Autonomous Learning is consistent with what fifty years of motivation research says about attaching external rewards to autonomy-dependent behaviour. It is not a claim that the current implementation is perfect. It’s a claim that the direction is right.
The point of leaving it ungraded
A grade tells you what a student did to satisfy a requirement. It’s a useful signal for content with a defined boundary, and Kalvium uses it there, in Essentials, in the university exam system that produces the actual degree.
Autonomous Learning is trying to measure something a grade structurally cannot capture: whether a student has started to want to figure things out, independent of whether anyone is checking. The moment you check with a score, you’ve replaced the thing you wanted to see with a weaker substitute. A student performing initiative for an audience is not the same as a student exercising it because it’s theirs. Leaving it outside the grade book isn’t the absence of a design decision. It’s the design decision holding, even when the instinct to grade everything pulls the other way.
For a different but related reason, Kalvium’s most employable skills are also kept outside written exams. The parallel reasoning is in why Kalvium’s most employable skills are deliberately kept out of your exams. For the broader argument about what the learning science says most engineering programmes get wrong, see why most engineering programmes don’t produce engineers.
Arvind is Head of Programme Design and Delivery at Kalvium. He writes about the cognitive science of learning and how it shows up in the design of a live programme, grounded in named research and honest about where the evidence is still being worked out. Read more from Arvind or browse the B.Tech category.