About
Starisk is a small research project about the part of a Coursera course that the headline rating does not capture.
Why I built it
I started this while choosing my next course. I had taken a few 4.9 courses that were bad, some 4.7 courses that were good, and 4.8 courses that were excellent. The headline was always above 4.5, so it was not separating them.
The reviews were what separated them. Specifically, the ones other learners had upvoted carried more signal about whether a course was worth the time than the average did.
I tested the idea on a deliberate streak of good courses and bad ones. It kept holding: the upvoted reviews were pointing at a nearly independent dimension that the headline was not measuring. I suspected the headline was doing commercial work as much as measurement. Starisk does not claim to know intent — it just measures the gap.
Why publish it
When I was choosing, it was hard to find useful information about individual courses. The popular ones had Reddit threads. The niche, specialized, or newly published ones often had almost nothing. I wanted to make the signal I was already using easier to find.
I took many of my courses through Coursera's financial aid, so choosing carefully mattered before I applied. I wanted to know I would finish the course rather than merely tolerate it. Some of the best ones were as enjoyable as they were challenging.
This is also a research and portfolio piece. The crawler, the scoring model, the calibration test, and the site are all part of the work — not just a chart with a logo on top.
What the 29 means
I have completed 72 Coursera credentials (61 courses and 11 programs, plus 9 from other serious platforms). The 29 on the Method page are the subset I re-rated blind from memory to test the model. They are a calibration sample, not my transcript. The full list is on Background →.
That sample is small, and the Method page says so: n = 29, with roughly ±0.15 of uncertainty. It is directional evidence for a principled fix, not a proof.
What Starisk is, and isn't
It is
- What the most-upvoted written reviews say.
- A second number to read next to Coursera's.
- An honest record of where the two disagree.
It is not
- Objective quality, or a verdict on the teaching.
- An instructor rating.
- A product with accounts, or an affiliation with Coursera.
Who
I'm Matías Agelvis. You can find me at matiasagelvis.com, or write to hello@matiasagelvis.com.
The code, the data pipeline, and the calibration notes are in the repository. If you want the technical version of this page, read the method.