Self-Assessment
A written report can be polished well past the point of reflecting real understanding — that’s true of any writer under deadline pressure, and it’s become far easier with generative AI tools in the mix. The computer science capstone’s answer wasn’t to police the writing more strictly; it was to build assessment around formats that are much harder to fake.
The central mechanism is Level of Expertise Dialogue: an unscripted, notes-free, in-person conversation about your own work. A memorized pitch survives about one follow-up question. What’s left after that is a genuine test of whether the understanding behind the work is real — which is exactly why this format carries real weight in the course, alongside (not instead of) the writing itself.
The same instinct — test understanding through live, low-stakes conversation rather than through scored artifacts alone — shows up in a smaller, friendlier form too: hobby talks. Early in the semester, before any of the real stakes had built up, each student gave a short talk on a hobby they actually knew well — chosen deliberately, since hobbies are very often genuinely technical: strategy in a favorite game, the mechanics of a craft, an approach to training. There was no grade attached to it. Feedback was given live, in the room, in a deliberately friendly environment — real, specific comments on what worked and what could improve, not a score. The point wasn’t to evaluate the talk; it was to give students a safe place to practice being watched, questioned, and critiqued before any of that had real consequences attached to it.
Both formats rest on the same belief: self-assessment isn’t really about assigning a number to how good something is. It’s about building the ability to stand behind your own understanding, out loud, in front of other people — a skill that a grade can’t teach, and that only gets built by actually doing it, repeatedly, in an environment where getting it wrong the first few times is expected and fine.