Challenges (Desafíos) — creation under authentic constraint, with verified human direction

What this is. The model of one of Smoother’s nineteen modalities. It describes what any Challenge is, from any organizer, in any edition and in any country. Where the model needs a fact that only a concrete edition can supply — calendar, venue, theme, the categories actually opened, partners — it names the class of data and does not state it: those are instance facts and they live on the event’s surface, never here.

The modality has no theme of its own — and AI is its environment, never its subject. “Cook the best meal”, “Sustainable social impact”, “Shared prosperity” or “Learning while having fun” are all legitimate Challenges. The theme is chosen by whoever organizes; the canon governs only its form. AI appears as the environment in which every challenge runs today, as the safeguard — never as its subject. And it degrades gracefully: a cooked dish is AI-resistant by nature.

A Challenge is not a course, and it is not a hackathon. Not the first, because it is not a structural container: it is a modality delivered inside a Learning Program or a Learning Experience (Programa, Experiencia), with its Learning Units (Unidades de Aprendizaje), its Learning Object (Objeto de Aprendizaje) and its rubric. And not the second, because the hackathon optimizes a one-day product, whereas the Challenge is a cycle whose final question is not “what did you build?” but “what are you capable of doing now, even with the tool switched off?”.

The six positions of the model

A well-authored Challenge is one in which a generic AI answer is insufficient by design. The event does not police AI use: it authors problems that AI alone cannot finish.
Process weighs more than product — approximately 60% process and 40% product — because a beautiful artifact can hide an empty process, and because polish bias is a known failure mode of the format.
Agency is authenticated orally, not declared in writing. The Creation Log (Registro de Creación) is evidence for the jury, never a scored count; what decides advancement is the Live Defense (Defensa en Vivo): whoever cannot account for their own decisions did not make them. The same instrument detects AI autopilot and adult over-assistance.
Capability is demonstrated when the tool is switched off, through a transfer task after the event. If the transfer does not appear, that is the finding and it is published.
The Challenge is guided, not minimal-discovery. That is the format’s intellectual vulnerability, and it is answered with scaffolding: prior preparation, rehearsal, mentors present, process gates, guidance that fades but does not disappear.
Equity is structural or it does not exist. Only freely and publicly available tools; no privileged medium; intellectual property belonging to the person taking part, absolutely; and mandatory local reframing of the theme.

The nine resistance properties

The authoring core. Every brief satisfies them before being published — it is a signed checklist, not an aspiration: authentic context · incomplete information · competing values · local or live information that is in no training corpus · consequences and trade-offs with named costs and named parties · demonstrable artifact · disruption · provenance recorded lightly · defense that survives live questioning. The tenth requirement is a negative test: if the authoring team can name the correct answer, the brief is not finished. And every brief declares its friction point: where AI predictably fails on this problem. A brief without a declared friction point is not publishable.

The properties guarantee that a generic answer does not win; they do not guarantee that a real team finds where to start. Hence the structure floor: a suggested route, the list of contents of the deliverable, open candidate formats, and the requirement of relevance of one’s own — never scored, brief-specific or rejected, and faded by category.

Authoring prohibitions: no required medium · no obligatory method worksheet · no museum of pedagogical frameworks · no poster ethics — ethics lives inside the content, in a load-bearing dilemma the team cannot go around · no real personal data · no volume tasks.

The cycle and the mechanics

Five phases on the CEMI Impact Arc (Arco de Impacto CEMI) — Engage → Enable → Inspire → Empower → Connect — with two flows: a single-day sprint, or a guided build over several weeks. Hard rule: the judged finale is live, in every category and in every flow. Reviewing a video authenticates nothing.

Five signature mechanics run in all categories, calibrated and never omitted: the Initial Human Window (Ventana Humana Inicial) — the first minutes after the reveal, without AI, with an individual record before any group convergence; the Divergence Gate (Compuerta de Divergencia) — three substantially different approaches before producing, verified by the mentors and not by the jury; the Disruption (Perturbación) — a new condition delivered by the organizer, identical for all teams, announced live; the Mentoring Market, with consultation tokens equal for everyone; and the post-event AI-free transfer task, which is neither a punishment nor a gate and whose result goes back to the person and to nobody else. And a sixth piece with a status of its own, the Process Gate (Compuerta de Proceso): each team declares the framework it is going to work with — one of those taught, a combination, or its own way, named in its own words — and why. The deliberation of the adoption is read; fidelity, never.

The four Categories (Categorías) are developmental bands, never cognitive ceilings: it says Category and not “level” deliberately. One theme, four cuts of the same problem, designed with a low floor, a high ceiling and wide walls — and a ranking between categories is never produced.

Evaluation

The modality invents no evaluation. A Challenge declares its own Learning Object, stable across editions and categories; from it comes the rubric — rows = components of the Learning Object, columns = the four canonical levels — and the output is a multi-component profile, never a single grade. The weights express emphasis, never coefficients of a total, and no document or interface may present a weighted sum.

Aggregation is by vote, never by score. The jury nominates by named award category, and the submissions with the most nominations advance; no submission ever receives a composite number. Seven award categories, each traceable to a rubric row. The guard against polish bias is structural: visual finish and interactivity are recognized as mentions outside the advancement path.

Accreditation firewall. Accreditation, endorsement, approval, validation or certification is never claimed by anyone, for anyone. A certificate of participation is recognition, never certification; certification is separate and optional, and it certifies the Challenge’s Learning Object by producing a profile.

The Gym, the boundary and the preparation

A championship is a performance event and by itself does not supply the conditions of practice. The Challenge Gym (Gimnasio de Desafíos) is the layer that does supply them: short, open and free challenges, unscored and non-qualifying, where each mechanic is rehearsed at least once. Within it — and only there — the person orienting may demonstrate by doing, under two binding limits: only on a decoy, and nothing produced in a session may appear in a submission. The three decoy rules are a defense by construction, not a detector.

The mentor boundary: they may ask, explain, criticize, point at an idea, question an assumption; they may not operate the device, write the instructions or the code, produce an asset, determine the concept or edit the submission. And the preparation of the Learning Facilitator (Orientador de Aprendizaje — the role that accompanies the process of the person who is learning) is a verified condition, with its own Gate (Compuerta), its per-site roster and a deliberate separation between who verifies and who holds the veto — so that nobody has to say no to their own site.

Evidence

Level: Mixed, and explicit. Research-backed: the distinction between performance and learning — AI can improve immediate performance without improving learning, and part of the advantage disappears or reverses when the tool is withdrawn; homogenization, which is the most corroborated finding of the corpus and the reason the Divergence Gate exists; the superiority of guided instruction over minimally guided instruction, above all in beginners; the four C model; and the Consensual Assessment Technique, with its known limits. Contested or with a caveat, declared with its limit: the magnitude of deliberate practice, under replication debate; the “creative scar”, moderate evidence with the risk unmitigated in the literature. Synthesized: the assembly — the nine properties as a signed checklist, the cycle, the four Categories, the gate architecture, the award↔row mapping and the measurement model.

Original / design position, declared as such: the Guiding Position “The Almighty Gym”; the CEMI Impact Arc, created by Carlos Miranda Levy; the five signature mechanics as a set; the Process Gate — that requiring a framework declaration and an oral account of the process should improve the result is our choice, not a finding — and its deliberation sub-anchor; aggregation by vote; and the separation between verification and veto.

What is unresolved, and is said rather than hinted at: detecting a Gym leak is unresolved. Material made there precedes the clock and leaves no trace in the Creation Log. The application route exists; a structural detector does not exist, and it is not claimed to exist. Neither the minimum team size nor the number of consultation tokens is ruled either, and the measurement lacks a control group by construction.

Prohibited wording on every surface and in every language: “proven method”, “guaranteed method”, “scientifically guaranteed”, “accredited by”, “endorsed by”, “certified by” a framework owner, and every uncovered form of “the world’s first”. The labels of techniques and frameworks are never upgraded.

Vocabulary

This model speaks in Smoother’s own voice: Sujeto de Aprendizaje / participante / aprendiz (learning subject / participant / learner), never “estudiante” or “alumno” (student, pupil); Orientador de Aprendizaje, never “docente” or “profesor” (teacher, professor); Unidad de Aprendizaje, never “módulo” (module); and a Desafío is a modality, never a “curso” (course). The exception in the capacity rule operates per artifact surface: when material addresses schools, families, educational authorities or sponsors, and in consent and safeguarding instruments, the institutional register is permitted because that is the register with which the institution names its own relationship with the person who studies. The exception never reaches Smoother’s voice nor anything a participante reads.

Learning Models · The Learning Object · Evaluation · Scenarios · Role Simulation · Memory and Learning · Artificial Intelligence in Learning · Learning Activities · Study and Practice Techniques · Riddles and Puzzles · Task Routing · Discovery · Classification · Onboarding · Professional Frameworks · Evidence Framework

Source

protocol/smoother-desafios.md — state of 2026-09-17 (open section of the CHANGELOG, after 0.40.0).