How predictable are KUCCPS degree cutoffs?

01The question

CareerNext shows students an expected cutoff for the next intake. It comes from a rule: a weighted average of past cutoffs, blended with last year's figure, with two adjustments ("floors") and a range around it.

Before replacing that rule with a model, we wanted to know:

1. How much do cutoffs actually move from year to year?
2. Does our rule beat the simplest possible guess: "same as last year"?

02The data

• 2,178 degree offerings from CareerNext's database, with cutoffs for 2021–2025. Cutoffs are cluster points out of 48.
• About half of degree programmes don't fill in a given year (54–60%). For those, anyone who meets the cluster minimum gets in.
• From 2025 the portal leaves those programmes' cutoffs blank. So a blank means "the minimum was enough", not missing data.

03How we tested

We hid each year's real cutoffs (2023, 2024 and 2025), predicted them from the earlier years only, and compared. We tested CareerNext's live model against simple methods, including "same as last year".

We scored each method by its typical miss: half of its predictions were closer than this, half further, in cluster points (lower is better). We also checked how often CareerNext's range contained the real cutoff, and how often its likely/unlikely labels were right. For 2025, misses are scored only on programmes that filled.

04What we found

Findings:

1. Cutoffs above the minimum move a little over one point a year. About one in three moves more than two, and whole years shift together. 2025 rose by about half a point.
2. CareerNext's model is ahead of "same as last year" every year, but only just: by 0.07 to 0.18 points.
3. Extending a trend is the worst method in all three years. A rise one year doesn't reliably mean another rise.
4. Treating courses that usually don't fill separately helped. Likely/unlikely labels were right 89%, 88% and 84% of the time, up from 87%, 86% and 81%.
5. The wider range holds about two-thirds of real cutoffs (61–67%). It's a typical range, not a safe one.
6. Trend direction isn't a forecast. Last year's direction repeated only 33–42% of the time.

05What it means for CareerNext

Done since our first test: the range is wider, and courses that usually don't fill are treated separately. Next:

• Show trend arrows as history ("rose last year"), not as a forecast.
• Judge every change on all three test years together.
• Set the bar: any future model must beat "same as last year" on years it hasn't seen. If it can't, we don't ship it.

06Limits

• Three test years only, and the model's lead is small.
• 2025's minimum wasn't published, so for blank programmes the exact cutoff is assumed. The likely/unlikely results hold either way.
• Degree courses only. Diploma, KMTC, TVET, and TTC courses have no official cutoffs to test.
• This measures how well we predict cutoffs, not whether students made better choices.

07Status

Concluded. The question is answered. The fixes are next, in CareerNext, and they'll be measured with the same test.