Coding and AI for teens in Ica, Peru

INEI figures: the 2017 city count is in table 1.9 of Perú: Perfil Sociodemográfico 2017; the 2025 counts are in table 1 of the Censos Nacionales 2025 demographic indicators. INEI had published no 2025 city count by November 2026. Units overlap and we never add them.

From cuadros 11.13, 11.6 and 11.5 of INEI's Compendio Estadístico Perú 2026 (tomo 1), based on ENAHO, the national household survey. INEI prints decimal commas. The 2015 baseline is our choice.

MINEDU, CNEB plan de estudios for secundaria, under the Jornada Escolar Regular and the Jornada Escolar Completa. The last row is our reading of the full plans. Read November 2026.

import csv
from collections import Counter

pairs = Counter()
with open("DataSet1_Agroindustria_ICA_2023_rev.csv", encoding="utf-8") as fh:
    for firm in csv.DictReader(fh, delimiter=";"):
        pairs[(firm["tamanio_emp"], firm["exporta"])] += 1

exporters = pairs[("MICRO", "SI")] + pairs[("PEQUENA", "SI")] + pairs[("MEDIANA", "SI")] + pairs[("GRAN EMPRESA", "SI")]
big = pairs[("GRAN EMPRESA", "SI")] + pairs[("GRAN EMPRESA", "NO")]
print("Firms that export:", exporters)
print("Large firms:", big)
print("Large and exporting:", pairs[("GRAN EMPRESA", "SI")])

Eight firms are both large and exporters. Divide by the 10 large firms and you get 80%: most large firms export. Divide by the 22 exporters and you get 36.4%: most exporters are not large. Same eight firms, two denominators, two true and very different statements.

import csv
from collections import Counter

pairs = Counter()
with open("DataSet1_Agroindustria_ICA_2023_rev.csv", encoding="utf-8") as fh:
    for firm in csv.DictReader(fh, delimiter=";"):
        pairs[(firm["tamanio_emp"], firm["exporta"])] += 1

def share(part, whole):
    return round(100 * part / whole, 1)

sizes = ["MICRO", "PEQUENA", "MEDIANA", "GRAN EMPRESA"]
exporters = sum(pairs[(s, "SI")] for s in sizes)
for s in sizes:
    firms = pairs[(s, "SI")] + pairs[(s, "NO")]
    print(s.ljust(13), "firms", str(firms).rjust(3),
          "| of these, export:", str(share(pairs[(s, "SI")], firms)).rjust(5), "%",
          "| of all exporters:", str(share(pairs[(s, "SI")], exporters)).rjust(5), "%")

Data: Ministerio de la Producción (PRODUCE), Empresas Agroindustriales Ica, 2023, on datosabiertos.gob.pe, licence ODC-By. Shares are our own calculation. The file spells small as PEQUENA, without the ñ. Firms by province: Ica 284, Chincha 140, Pisco 39, Nazca 27 and Palpa 4.

The micro row is the surprise. Only 1.3% of micro firms export, yet they are more than a quarter of all exporters, because there are 457 of them. A small rate applied to a big group can outweigh a large rate applied to a small one. Statisticians call the size of each group the base rate, and ignoring it is one of the commonest reasoning errors there is.

AI models make the same slip. A model trained to spot exporters that leans on "large" will be right about large firms and still miss most exporters, because most of them are not large. Our teenagers learn to ask "a percentage of what?" before they trust any number, theirs or a machine's.

1. Without code, ages 13 to 14. In a class of 30, 4 students play the cajón and 3 of those 4 are in 4° de secundaria; the class has 12 students in 4°. Work out both percentages, and say which one a music teacher would want.
2. First Python. Run Version 1, then add the line that prints both percentages for large firms with the right labels.
3. Ages 15 and over. Change Version 2 so it groups by province instead of size. Which province has the highest share of exporters, and is it also where most exporters are?
4. Going further. Write a function both_ways(a, b) that takes any two yes/no columns and prints the share in each direction, then test it on wine makers and exporters.

A practice problem of ours. A register lists 30 firms: 6 export, and 4 of the exporters are large; there are 5 large firms in all. What share of large firms export, and what share of exporters are large? Then: if you picked a firm at random and were told only that it exports, would you bet it is large? (Ours, written for this page. Answers: 80% and about 66.7%; yes, slightly, since two in three exporters are large here, unlike in Ica's real register.)

Routes are our suggestions. Grades follow MINEDU's normative age on 31 March: 12 in 1° de secundaria, 16 in 5°.

Every route keeps reasoning in front: tests before trust, explanations in the learner's own words, and no copying from AI. Teenagers heading for university admission exams, which are in Spanish, can see how those exams work on our page Peru university admission exams explained; we teach alongside school and never promise admission.

Peru keeps UTC minus 5 hours every day and has not used daylight saving since 1994 (tz database, America/Lima; INACAL publishes legal time). India keeps UTC plus 5 hours 30 minutes, so India stays 10 hours 30 minutes ahead of Ica.

Neither clock changes, so a slot booked in April is still the same slot in November. Families book in their own time and the mentor sees the matching India hour. Many Ica teenagers start in the January to March holidays and keep their slot when classes resume.