in high school, i was extremely good at school.

i don’t mean that in the vague, linkedin way where everyone was apparently a visionary at seventeen. i mean that i understood the system, optimized for it, and the system rewarded me. i graduated as a top ontario scholar and received the governor general’s academic medal. i also got accepted into waterloo’s computer engineering program, which felt like the natural conclusion to all of it.

at the time, i treated those results as fairly strong evidence that i was smart and capable. honestly, why wouldn’t i? grades were the clearest feedback loop available to me. learn some material, write a test or two, receive a number. repeat. when the number was high, the conclusion felt obvious: i understood the material and was progressing in the right direction.

four years of computer engineering at waterloo, six internships, and a lot of humbling projects later, i think grades are a much narrower signal than i once believed. not a useless signal – a narrow one. this distinction matters, and i’ve been trying to preach it to my peers at waterloo for quite some time now.

the system was not lying

i’m not trying to rewrite my high-school self as naïve or pretend the work meant nothing. those grades reflected real things: discipline, consistency, the ability to learn quickly, and a willingness to do the work. the awards were legitimate, and i am still proud of them.

a grade measures how well someone performed on a particular set of assessments, under a particular set of constraints, at a particular point in time. depending on how the course is designed, that can reveal a lot. it can show command of prerequisites, speed, precision, preparation, or the ability to reason through difficult material.

but it can only measure what the assessment actually asks you to do.

most graded coursework is intentionally bounded. the syllabus identifies the domain. the question usually has an intended interpretation. the necessary information is either available or expected to have been memorized. someone else has already decided that the problem matters, selected the constraints, and compressed the acceptable outcome into something that can be evaluated consistently.

sure, bounded problems are useful. they make it possible to teach fundamentals, isolate concepts, and give thousands of students reasonably comparable feedback. and obviously, this description does not apply cleanly to every part of academia. research, theses, and genuinely open-ended projects can be as ambiguous and judgment-heavy as any engineering work. the point i’m trying to make is that a large amount of what we call academic performance is performance inside a carefully constructed environment. engineering outside that environment has rarely felt as clean.

the difficult part – and usually the most fun – is often figuring out what the problem actually is: requirements are incomplete, constraints conflict, the first design is defensible but disproportionate, something works in isolation and fails under load, etc…. you can spend a week producing a technically impressive solution to a problem that never deserved one.

grades compress away the interesting parts

the way i see it, a grade takes a complicated process and reduces it to one scalar. i think we make peace with that because scalars are easy to store, rank, filter, and compare. it is also very funny because much of what gets compressed away is almost exactly what matters outside school lmao.

a grade says very little about whether i can:

  • notice that the obvious problem is not the real bottleneck;
  • work inside a codebase i did not design;
  • make progress when the requirements are ambiguous;
  • ask for help before uncertainty becomes wasted work;
  • explain a tradeoff well enough for other people to trust it;
  • decide that a clever solution is worse than a boring one;
  • own an outcome after the interesting coding is finished.

to be very clear, i’m not saying that someone with high grades lacks those abilities. what i’m saying is that the number itself usually cannot tell you whether they have them.

optimization changes the objective

i think the deeper problem here though, is that: measurements do not merely observe behaviour. once people care enough about them, they begin shaping behaviour. concretely, when the grade becomes the objective, naturally, you start asking questions like:

  • will this be on the exam?
  • how many marks is this worth?
  • what does the grader want?
  • what is the minimum work required for the maximum result?

lmfao, i have asked all of them at some point.

these are rational questions inside the school system. if several deadlines collide and one task is worth two percent, pretending that its weight does not matter would be stupid. the problem begins when this local strategy becomes a general instinct.

again, optimizing for grades does not automatically prevent learning. often, the two objectives align. understanding the material deeply may be the best way to earn the mark. but they are not identical objectives, and whenever they diverge, the metric creates some strange incentives. you memorize something long enough to reproduce it, then forget it. you avoid the interesting tangent because it will not be tested. you choose the safe project whose evaluation criteria are obvious. you learn to recognize the form of a solution without building the model underneath it.

the best parts of my engineering growth came from almost the opposite instinct: reading beyond what was required, building something before i knew whether it would work, tracing a performance problem across boundaries that were not technically mine, and going deeper into stuff i cared about because the questions became interesting – not because they were worth ten percent.

when every effort needs an assigned mark, ungraded curiosity begins to look inefficient. imo, that is a terrible habit to carry into an engineering career.

what grades are actually good for

grades exist for real reasons. universities need scalable ways to evaluate prerequisite knowledge, employers with too many applicants need cheap filters, students need feedback, and a well-designed exam can obviously reveal gaps that confidence alone will happily conceal.

the issue is not that a scalar is useless. the issue is forgetting how much information was discarded to produce it.

if a student has an exceptional grade in a difficult course, that is evidence of something. if a student has an average grade, that is also evidence of something. neither number tells you, by itself, how curious they are, how they behave when no solution is specified, whether other engineers trust them, or how quickly they improve after being wrong.

grades matter most when there is nothing better to measure. early on, there may be no shipped system, research output, internship feedback, or serious project to inspect. the transcript kinda becomes a proxy because the underlying qualities are expensive to observe directly.

as better evidence accumulates, the proxy should ideally lose authority.

a better objective function

don’t overcomplicate it, just ask different questions:

  • can i derive this from first principles rather than recognize the familiar pattern?
  • can i use it in a system whose boundaries were not chosen for my convenience?
  • can i explain where the model stops being true?
  • can i make a reasonable decision when every option has a downside?
  • can i notice when i am solving the wrong problem?

these questions are harder to score, which is partly why institutions rely on easier ones. but difficulty of measurement does not make the underlying quality less real.

sure, i still care about doing relatively well, i obviously wouldn’t flunk my entire transcript under the name of grades are overrated. i respect people who are excellent academically, and i am not retroactively converting every mediocre grade i received into proof of hidden genius lol. most of the time, a bad grade means i did not understand the material, did not prepare, or simply did poor work.

i just no longer let the number claim more than it measured. grades are not meaningless. they are a lossy compression of learning: convenient, comparable, and incapable of preserving everything that matters.

~ a.k