Coding and AI for teens in Cajamarca, Peru

INEI, Perú: Perfil Sociodemográfico 2017, table 1.9 (city, people counted in 2017), and INEI, Censos Nacionales 2025, Indicadores demográficos, table 1 (total population at 15 September 2025). No 2025 city count had been published by November 2026.

The district is an administrative unit, not the same area as the city INEI described in 2017, and its figure is eight years newer, so 242,285 against 201,329 is not a measure of growth. Comparing numbers that count different things is a mistake data scientists learn to avoid early, and it is the same mistake the project lower down guards against.

How Peru's main admission exams work, from the universities' own pages, is explained in Peru university admission exams explained.

Route and outcomes are ours. MINEDU sets normative ages on 31 March, which put a 1° de secundaria pupil at 12 and a 5° pupil at 16; our courses are for 13 to 18, so a 12 year old in 1° joins once they turn 13.

Our national guide, coding classes for teens in Peru, compares the options in more depth; Spanish-speaking parents can read clases de programación para adolescentes.

import csv

def guess(theory, practice):
    return theory + practice / 2

match = 0
rows = 0
with open("unc-cursos-pregrado-2026-I.csv", encoding="latin-1") as listing:
    for course in csv.DictReader(listing):
        rows += 1
        t = int(course["HorasTeoria"])
        p = int(course["HorasPractica"])
        if int(course["Creditos"]) == guess(t, p):
            match += 1
print("Rows:", rows, "Rule fits:", match, "Misses:", rows - match)

Two thirds right is not bad for a guess, but 961 misses is too many to ignore. Printing a few of them shows the cause: most have an odd number of practice hours, so the guess gives a half, such as 3 theory and 3 practice giving 4.5. The university's credits are whole numbers. So it must round, but which way?

Rather than argue, Version 2 tests three rival rules side by side: the exact guess, the guess rounded up and the guess rounded down. It counts the misses for each, and then lists the rows that even the strongest of the three cannot explain.

import csv
import math
from collections import Counter

rules = {
    "exact": lambda t, p: t + p / 2,
    "round up": lambda t, p: math.ceil(t + p / 2),
    "round down": lambda t, p: math.floor(t + p / 2),
}
misses = {name: 0 for name in rules}
left_over = Counter()

with open("unc-cursos-pregrado-2026-I.csv", encoding="latin-1") as listing:
    for course in csv.DictReader(listing):
        t = int(course["HorasTeoria"])
        p = int(course["HorasPractica"])
        c = int(course["Creditos"])
        for name, rule in rules.items():
            if c != rule(t, p):
                misses[name] += 1
        if c != rules["round down"](t, p):
            left_over[(c, t, p)] += 1

print(misses)
for (c, t, p), n in left_over.most_common(3):
    print(n, "rows: credits", c, "theory", t, "practice", p)

Course list for 2026-I from the Universidad Nacional de Cajamarca, reused under its Open Data Commons Attribution licence. The rules are our guesses and every count is ours. Hours are weekly hours as the university's data dictionary defines them. Four sections carry 0 credits, among them Deporte and Danzas Típicas Peruanas.

Rounding down wins clearly: it fits 2,683 of 2,882 sections, about 93%. But 199 sections still break it, and they are not random. The largest group, 122 sections with 2 hours of theory and 3 of practice, gets 4 credits where the rule says 3. Perhaps those courses follow an older study plan, or a school sets its own credits; the file does not say, and an honest analyst says "we do not know" rather than inventing a reason.

This is how an AI model should be judged too. A model that is right 93% of the time is useful, and the 7% it gets wrong is where the interesting questions are. Teens who learn to list the exceptions, instead of hiding them, become the people who can be trusted with real data.

1. Ages 13 to 14, on paper. For five rows of the table, work out the guess by hand, then round it both ways. Which rule would you have bet on before seeing the counts?
2. First Python. Run Version 1, then print the first ten rows it misses, with their hours. Write down the pattern you see before reading on.
3. Ages 15 and over. Group the 199 remaining misses by professional school. Are they spread out, or concentrated in a few schools?
4. Going further. The file also gives the number of students in each section. Add them all up: you get 52,488. Explain why that is a number of seats, not of students, and why the two are different.

Peru is fixed at UTC minus 5 hours and has used no summer time since 1994 (see the America/Lima zone of the tz database; legal time comes from INACAL). India is fixed at UTC plus 5 hours 30 minutes, which keeps it 10 hours 30 minutes ahead of Cajamarca.

Because both clocks stay put all year, a booked slot never drifts. The family books in Cajamarca time and the mentor sees India time. The long summer break from January is a natural moment to begin, and the same hour can carry on into the school year.