I Built an Agent to Predict My Son's Final Exam Topics (This Year It's My Topolone's Turn)
Let me set the scene, because it matters. My oldest — we’ve called him Topolone since he was three, something like “big little mouse” that just stuck — has his maturità in a few days. That’s the Italian final high-school exam: two long written tests plus an oral, the thing that decides whether you graduate. He’s at a liceo scientifico, the science-track high school. First written exam Thursday June 18, second one Friday June 19. He’s taller than me now, by the way, and he never lets me forget it whenever I fumble a password.
And here’s the honest part: things at home had gotten pretty tense. Not because of him. Because of me. I’m exactly the kind of parent who, instead of just staying calm, starts circling his kid every twenty minutes asking “did you go over Pirandello? And integrals? You know there’s probability too, right?”. Honestly, I was the anxious one — he, all things considered, was handling it with a lot more composure than I was.
At some point he looked at me and said the sentence every nerd dad deserves to hear at least once: “Dad, instead of stressing me out, why don’t you do something useful with all those computers?”
Touché.
So I did the only reasonable thing: I built an agent
Instead of breathing down his neck, I locked myself in the office and built a small agent. The idea was simple: half of Italy already does this “exam topic betting pool” — the tototracce — by hand every June, cross-referencing literary anniversaries and historical milestones across dozens of websites. It’s exactly the kind of tedious, repetitive, “read a ton of sources and summarize” work an AI agent is good at and an anxious dad is terrible at.
The setup: the agent runs on Claude Opus 4.8, does a round of web research (Skuola.net, Studenti.it, Geopop, Il Fatto Quotidiano and a few others), reads through the prediction articles, extracts the recurring names and themes, and weighs them by two factors that actually matter in this game:
- How “round” the anniversary is — the Italian Ministry of Education loves round numbers: the 100th anniversary of this, the 80th of that.
- How long a given author has been absent from the exam — the longer they’ve been “overdue,” the higher they climb.
Then I took the qualitative signals from those articles — “clear favorite,” “practically locked in,” “the long-awaited one” — and turned them into relative percentages. Nothing scientific, to be clear. It’s an attention index, not a bookmaker’s odds. But at least it’s organized anxiety.
One honest disclaimer before the numbers, the same one I gave my son: this kind of prediction is more an exercise in hope than an actual forecast, and the Ministry has a long, proud tradition of catching everyone off guard. So treat these tables as a compass for revision, not an oracle.
First written exam — Italian (Thursday, June 18)
The structure has three categories: A (literary text analysis), B (argumentative essay), and C (reflection on a current-affairs topic). For each category the Ministry offers multiple prompts to choose from — the percentages below indicate how likely a given name or theme is to show up among the options, not that it’ll be the only one.
Type A — Literary text analysis
2026 is a wild year for literary anniversaries, which the agent loved (I loved it a little less, since it means my son has to review all of them).
| Author / work | Why it’s on the list | Probability |
|---|---|---|
| Grazia Deledda | 100 years since her Nobel Prize (the only Italian woman to win it), never appeared on the exam before | 35% |
| Giovanni Verga — I Malavoglia | 145 years since publication, top favorite in polls | 30% |
| Luigi Pirandello | double anniversary: 90 years since his death and 100 years since One, None and a Hundred Thousand | 25% |
| Gabriele D’Annunzio | absent from the exam since 1999, “the long-awaited one” | 15% |
| Giovanni Pascoli | a recurring author in recent years’ exams | 8% |
| Eugenio Montale | missing for 14 years (last appeared in 2012) | 8% |
| Carlo Collodi | 200 years since his birth | 5% |
| Filippo Tommaso Marinetti | 150 years since his birth, founder of Futurism | 5% |
Translated for my son: review Deledda, listen to your father. A hundred years since her Nobel, the only Italian woman to ever win it, and she’s never come up. That’s the kind of coincidence the Ministry can’t resist.
Type B — Argumentative essay
| Topic | Why it’s on the list | Probability |
|---|---|---|
| 80th anniversary of the Italian Republic and women’s right to vote | the June 2, 1946 referendum, a “round” anniversary considered nearly locked in | 45% |
| 40th anniversary of the Chernobyl disaster | April 26, 1986, themes of energy/environment/tech responsibility | 30% |
| 25th anniversary of 9/11 | a historical turning point, geopolitics and global security | 15% |
| 800th anniversary of the death of St. Francis of Assisi | the relationship between humans and nature, the Canticle of the Creatures | 15% |
| 40th anniversary of the Palermo Maxi Trial | rule of law, the fight against the mafia, Falcone and Borsellino | 12% |
The 80th anniversary of the Republic got the highest score in the whole analysis, and here the agent and a parent’s gut feeling agree: a round anniversary, women’s suffrage, the Constitution. It doesn’t get more “locked in” than that.
Type C — Current-affairs topic
And, of course, the topic closest to my own day job wins this category.
| Topic | Why it’s on the list | Probability |
|---|---|---|
| Artificial intelligence (young people, work, misinformation) | the single most-mentioned topic across every source | 50% |
| Climate crisis and eco-anxiety | a topic Gen Z cares deeply about, both environmentally and psychologically | 25% |
| Geopolitics and the right to peace (tied to Article 11 of the Constitution) | ongoing conflicts, possibly linked to the 80th anniversary of the Republic | 20% |
| Mental health and youth distress | social media pressure, performance anxiety | 18% |
| Women in STEM (e.g. Margherita Hack) | overcoming gender stereotypes in science | 15% |
| Migration and identity | globalization, cultural belonging, coexistence | 12% |
There’s something beautifully absurd about using an AI agent to predict that the most likely current-affairs topic is… AI. If it actually comes up and my son picks it, I’ve already told him we have first-hand material at home.
Second written exam — Math, science-track high school (Friday, June 19)
There’s one extra certainty here: the Ministry confirmed back in January that the second exam for the science track, applied-sciences track, and sports track will be Math only, no Physics. The structure is the usual one: 2 problems (pick 1) + 8 short questions (answer 4), 6 hours total.
Here the percentages aren’t about “which exam will show up” — the exam is the same across all of Italy — but about how likely a given topic is to appear across the problems and questions, based on how central it’s been in recent sessions and official mock exams. And since there’s real historical data here, these numbers are far more reliable than the literature predictions above.
| Topic | Probability of appearing |
|---|---|
| Function analysis (limits, asymptotes, derivatives, max/min, inflection points) | 90% |
| Differential calculus and theorems (Rolle, Lagrange, Cauchy, L’Hôpital) | 80% |
| Integral calculus (areas between curves, volumes of solids of revolution) | 75% |
| “Real-world” / modeling problem (optimization, applied contexts) | 70% |
| Probability and statistics (combinatorics, Bayes, binomial, normal distribution) | 55% |
| Differential equations (growth, radioactive decay, tank drainage) | 50% |
| Analytic geometry in 3D space (planes, lines, spheres, distances) | 45% |
The message for Topolone is unambiguous: function analysis is basically guaranteed. If there’s one thing you can’t walk in unprepared for, it’s that. Everything else is a matter of decreasing priority, not an excuse to skip whole chapters.
The moral (for me, not for him)
In the end the agent did its job: it turned my scattered anxiety into a tidy table, and gave me something concrete to put on the kitchen table instead of the usual barrage of questions. My son looked at it, raised an eyebrow, and said “okay, not bad” — which, in nineteen-year-old, is roughly a standing ovation.
But here’s the thing I actually learned, and I’m writing it down because in a few years it’ll be his younger brother’s turn. No model, however powerful, tells you the thing that actually matters: that it’ll be fine either way. That this exam is a rite of passage, not a verdict on who you are. That ten years from now he won’t remember which topic came up — he’ll remember his father, instead of nagging him, locking himself in the office to build him a little toy just to say I’m here with you.
So yes: use these tables to prioritize your revision, never to skip whole sections of the syllabus. The Ministry is famously good at catching everyone off guard.
And Topolone — good luck for the 18th and 19th. Dad’s here. Even when he’s annoying.
Update (June 27): how it actually went
The exams happened, and I have to be fully honest about it: the Ministry didn’t follow a single gram of our statistical reasoning. As tradition dictates, it did its own thing. The agent had built an attention index around round anniversaries and authors “overdue” for years — the Ministry responded by picking authors who had no anniversary to celebrate at all. The exam-topic betting pool remains an exercise in hope, not an oracle, exactly as I’d warned my son.
Type A: a complete whiff
The literary-analysis prompts were Cesare Pavese (“Passerò per Piazza di Spagna”) and Vitaliano Brancati (“I piaceri”). Zero for eight. No Deledda with her Nobel centennial, no Verga, no Pirandello with his double anniversary. The instructive part, in hindsight: neither Pavese nor Brancati has a “round” anniversary in 2026 — the Ministry broke, on its own, the very heuristic half of Italy’s prediction pool runs on, my agent included. In the year with the most round literary anniversaries in recent memory, it picked two authors completely off every anniversary list.
Type B: a partial hit
This one went a bit better. Prompt B1 is a speech by Giuseppe Saragat to the Constituent Assembly, dated June 26, 1946 — not the women’s-suffrage referendum text we had on the list, but still within the orbit of the 80th anniversary of the Republic, the topic the agent had ranked at 45%, the highest score in the whole analysis. Right in spirit, wrong on the specifics. The other two prompts — Bianucci on scientific creativity and Furedi on borders — weren’t on any table, though Furedi’s touches the same nerve of identity and borders we’d quietly included among the possibilities.
Type C: the real plot twist
Here the agent whiffed completely, and in the most instructive way possible. We’d put artificial intelligence at 50%, the climate crisis at 25%, then mental health, geopolitics, women in STEM. What actually came up were two completely different prompts: Wenke Husmann’s “Funziona a meraviglia,” an invitation to rediscover wonder and awe, and Mario Calabresi’s “Alzarsi all’alba,” a meditation on the value of hard work in contemporary society. No AI, no climate, no performance anxiety — exactly the topics everyone, agent included, was betting on. While public discourse was looking at the “hottest” topics of the moment, the Ministry went the opposite direction: wonder and effort as an antidote to the noise.
Second written exam — math: here the numbers hold up
And this is where the difference I’d already flagged shows up: built on real historical data instead of anniversaries, the math prediction held up far better. Problem 1 modeled water-level changes in Lake Bracciano over 11 years — a piecewise function, a sinusoidal model, classic calculus theorems: exactly the “real-world / modeling problem” we’d put at 70%. It turned out so laborious and ambiguously framed that most candidates picked Problem 2 instead — classic function analysis, comparing a parametric rational function with an absolute-value one, a common tangent line, inflection points and area: our 90% on “function analysis” landed dead center.
Among the short questions, probability and combinatorics showed up twice — a without-replacement draw using cards from scopone, the Italian card game, and a pure combinatorics question: our 55% was on target. Analytic geometry in 3D space also appeared, with a sphere and a tangency condition to a plane: the 45% confirmed. The only topic that didn’t show up at all was differential equations, despite the 50% we’d given it.
The takeaway for next year’s revision is the confirmation of what I already wrote above: for Italian literature, the prediction pool stays an emotional compass, nothing more. For math, where the history of past exams actually matters, the statistical index worked. The Ministry can still catch you off guard on an author — and this year it did, completely — but it rarely surprises you on function analysis.