Why We Started with One Program in One Municipality

By Anushka Appala and Dr. Janio Rosales

The name is national. The first step is not. NaciluzIA is a National Platform for Algorithmic Governance, and yet our first build is a three-month MVP for one social program, in one municipality, in Spanish and Q'eqchi', running on simulated data. To some ears that sounds like timidity — a national ambition that flinched. We see it the other way. The narrowness is the discipline, and the discipline is the point.

Let us make the case for starting small on purpose.

What "national platform" gets wrong as a first move

There is a pattern in government technology that ends the same way often enough to be predictable: a large, ambitious platform is announced, procured, and half-built, and then it stalls under its own weight — too many stakeholders, too many programs, too much sensitive data touched too early, and no single moment where anyone could point and say this worked. The ambition was real. The sequencing was wrong.

We are trying to avoid that failure by refusing to earn it. A first step should be small enough to finish, concrete enough to measure, and safe enough that failure teaches instead of harms. "National platform, phase one" fails all three tests. "One program, one municipality, simulated data" passes all three.

The four constraints, and what each buys

Every boundary on the MVP is doing work.

One MIDES program. Guatemala's Ministry of Social Development runs real transfer programs at real scale — its Bono Social already reaches 107,425 families across 322 municipalities (MIDES, 2024). We are not inventing a toy problem; we are choosing one well-understood surface from a system that genuinely exists and genuinely matters. That gives the proof a real shape without asking us to model the whole ministry.

One municipality. A single municipality is a complete unit — it has citizens, officials, a program, and a workflow — while remaining small enough to understand end to end. It is the smallest thing that is still the real thing.

Spanish and Q'eqchi'. Guatemala has 25 official languages — 22 Mayan plus Xinca, Garífuna, and Spanish (Academia de Lenguas Mayas / Ley de Idiomas Nacionales) — and a public sector that operates, in practice, almost entirely in Spanish. Building bilingual from the first day, in the largest Mayan language, forces the hardest requirement to the front of the line rather than deferring it to a "localization phase" that never comes.

Simulated data. This is the constraint that lets us move at all. Running on simulated data means no sensitive citizen record is touched, no privacy is risked, and no live benefit hangs on an unproven system. We get to prove the mechanism before we are trusted with the stakes — which is the correct order.

The one thing it proves is the thing governments buy

Strip away everything else and a single claim has to survive the three months: that the time to complete a social-program application drops from weeks to minutes, while 100% of decisions stay with a human.

That pairing is the entire product. The speed is what a ministry, a mayor, and a citizen can all feel immediately — the difference between a process that consumes days and one that resolves in a sitting. In a region where a single government transaction takes 5.4 hours on average, and up to 11 hours in some countries (IDB, Wait No More, 2018), minutes is not an optimization; it is a category change.

But speed alone is a familiar and dangerous promise, because the fastest way to clear a queue is to let a machine decide it. So the second half of the claim is load-bearing: every binding decision still runs through a deterministic rules engine and is signed by a named human official, recorded in a tamper-evident ledger valid under Guatemala's e-signature law (Decreto 47-2008). One hundred percent human accountability is not a slogan we attach to the demo — it is the thing the demo is built to prove holds up under speed.

Narrow and measurable beats broad and vague

A pilot that proves one honest, quantified thing is worth more than a platform that promises everything and demonstrates nothing. When the three months end, we want to be able to say something a government can act on: here is the program, here is the municipality, here is the time-to-complete before and after, and here is the record showing a human stood behind every decision. That is a sentence a ministry can take to a budget meeting. "We are building a national platform" is not.

From there the path is deliberate — a live pilot on real data, a measured recovery of welfare leakage, an executive commitment, multilateral funding to scale. Each step earns the next. But none of it is credible without the first, small, finished proof. The word national is a destination. One program in one municipality is how you actually start walking toward it.

Cada decisión, a la luz.

— Anushka Appala and Dr. Janio Rosales