The Lab
Where we explore what's possible, before it becomes a solution.
Four rules for every entry.
- 01
One question, real data.
No invented numbers. When a product has no users yet, we test on made-up data and label it clearly. The code behind every figure can be re-run.
- 02
A simple baseline first.
Nothing counts as working until it beats the obvious guess.
- 03
Limits, stated.
Every entry says what it can't tell you.
- 04
Negative results too.
If an idea doesn't work, we publish that.
Data analysis · Concluded · 5 October 2026
How predictable are KUCCPS degree cutoffs?
We tested CareerNext's cutoff forecasts against what actually happened in 2023, 2024 and 2025. Simply repeating last year's cutoff was hard to beat, and extending a trend was the worst approach every year.
Related: CareerNextRead the entryTechnical note · Concluded · 4 October 2026
When a tenant mistypes their Paybill account number, where does the payment go?
We tested LandlordPMS's M-Pesa matching on a synthetic estate. It absorbed every formatting mistake, but a typo that spells another real unit went to the wrong tenant.
Synthetic dataRelated: LandlordPMSRead the entryData analysis · Concluded · 4 October 2026
How much TVET choice does a student have without leaving their county?
We counted every registered TVET course offering by county. A typical county offers about a quarter of the 221 courses, and five counties hold over a third of all offerings.
Related: CareerNextRead the entry
We publish an entry only when the work behind it is finished.
Have a problem we should explore?
Some of the best questions come from people who live with the problem.