Case study
Aguante: a health app built on the evidence
Every fasting app tells you that autophagy begins at 16 hours.
It doesn't. That number isn't supported by the human evidence, and the ketone curve people cite alongside it is close to flat between 12 and 18 hours. The claim spread because it's tidy and motivating, and because nobody checked.
Aguante is built on saying so.
Why that is a product decision
A health app that repeats a fabricated threshold has a structural problem: the moment a user learns the truth, everything else the app told them becomes suspect. In this category, trust is the whole product.
So the position is the differentiator. Aguante tells you what the evidence actually supports, which is less dramatic and considerably more useful.
It also means the content has to be right. When we found one false claim in the interface, fixing that one string was never going to be enough, because a false user-facing claim always has siblings. We grep the concept in both languages and fix the whole family. The version that ships a single correction is the version that keeps three more.
Bilingual, and the name means something
Aguante ships in English and Spanish, and it was written in both languages. Neither version is a translation of the other.
The name carries an obligation. Aguante means resilience and endurance, the act of holding on. Above all, it doesn't mean self-denial or punishment. That distinction is a brand rule, and it's enforced in every string in the app. A fasting product that drifts into deprivation language would be selling exactly the thing the name refuses.
What we built
A bilingual installable web app with fasting timers, evidence-backed explanations of what actually happens at each stage, and a consent and privacy layer built to match what the code really does instead of what a template says.
The testing lesson that came out of it
Three of four real bugs found in an audit of Aguante were invisible to state assertions. The app's internal state was correct. What a person actually saw was wrong. That produced a reusable visual-audit approach we now apply across projects: check what renders, and don't stop at what the state object claims.
Related: writing directly to localStorage doesn't move an app that has already loaded its
state into memory. A sweep that seeded eleven different states that way silently measured the
same state eleven times and reported clean. A check that would look identical if it were broken
proves nothing.
What this demonstrates
Research as a genuine product feature; defensible claims in a regulated-adjacent category; bilingual product craft with a brand rule enforced at the string level; and testing that verifies what a person sees.
Last updated