Hello Zindi team,

While validating a no-ID solution for the University Hackathon by Ngao Labs:

Financial Inclusion competition, I found strong evidence that Train and Test

were split chronologically inside each `country × target` class rather than by

an ordinary random split.

The evidence is reproducible without external data.

The competition page does not explicitly ban the provided `uniqueid`, but it

does reserve disqualification for data leaks or practices that compromise a

solution's inherent value. I have therefore not uploaded any submission using

this pattern.

Could you please confirm in writing whether either of the following is allowed?

1. using the numeric portion/order of `uniqueid` as a model feature; or

2. applying a deterministic rule based on the discovered Train/Test boundary.

If useful, I can privately provide exact counts and a minimal reproduction to

the competition team.

Thank you.