aiLearning Challenge rehearsal

Author your own briefsThe framework, the template and the worksheet so each guide designs their own challenge briefs. Organization and validationRoles, materials list, time log and the instruments for learning from the rehearsal.

The thesis this rehearsal puts to the test

When the geography teacher sent us to draw the map of the mountain ranges, we copied the maps from books — and nobody called it cheating. The teacher was not expecting cartographers: she expected that, by reproducing the map, we would interact with the learning object — the ranges, the peaks, the rivers — and end up learning it.

AI in education, used well, is exactly that: an instrument for interacting with the learning object in creative ways. The artifact produced is not the point. That is why in this format process weighs 60 and product 40, and why no judge ever rewards the number of tools, the prompt engineering or the newest model. A beautiful artifact can hide an empty learning process — and the whole format is designed so that does not go unnoticed.

The golden rule of brief authoring: a generic AI answer must be insufficient by design. The event does not prohibit AI — it writes problems AI alone cannot finish.

Agenda — 3 hours

Time	Dur.	Moment	What happens
0:00	15 min	Welcome and the thesis	Why the Challenge exists: the artifact is not the point — the point is interacting with the learning object. Rules of the rehearsal.
See more What the room must hear: the artifact is not the point — the point is interacting with the learning object. Just like the teacher who sent us to copy the map: she was not expecting cartographers; she expected that by reproducing it we would end up learning the geography. That is why process weighs 60 and product 40, and why nothing scores by volume of AI. Minimum rules to make clear: only free tools, the same for everyone · no medium is mandatory · the record is filled in as you work · the defense is without AI · the work goes in rounds and the first version is shown next to the final one.
0:15	5 min	Revealing the brief	The envelope opens. The brief is unknown until this moment — that is how the real event works.
See more Why sealed: a brief unknown until the clock starts levels the field — nobody arrives with a prepared advantage — and reproduces how real problems behave. The choice among the catalogue's briefs is made that morning, not before. In the room: one seal is opened, projected. The others stay closed — and the guiding team's block is never opened with participants present.
0:20	15 min	Human-First Window (no AI)	Each team states the problem in its own words, names what it does not know and what it assumes. No screens: paper and conversation.
See more What it produces: the problem stated in the team's own words, with what is not known and what is assumed, on paper. That paper travels: it is the evidence the profile reads in its first dimension — a framing of one's own, sustained afterwards under the pressure of whatever the model proposes. Why before any screen: framing is the moment most exposed to homogenization. When the model proposes its reading, every version drifts toward it; protecting these minutes protects the diversity of everything that follows.
0:35	15 min	Divergence Gate	Three substantially different approaches, on paper. The team picks one and declares what it optimizes. What is discarded is kept: it is evidence.
See more Who does what — per the Challenge canon: the divergence is run by the team (three substantially different approaches, each with what it optimizes and what it sacrifices; what is abandoned is recorded and earns process credit, never loses it). The gate is verified by the guiding team — at the real event, the site's educator-mentors; never the judges. It is a scaffold, not a score. The verification asks one thing only: was the final direction chosen from a space of possibilities — or is it the first plausible answer with makeup on? Three polished products are not required: evidence of choice is. A team that does not pass diverges again; failing costs time, never eligibility. Rehearsal finding (2026-08-18), already incorporated: each of the three approaches names its concept and its candidate output format. Passing the Gate includes having chosen: on the way out, the team knows WHAT will exist at the end, not just what it is about. The guidance guides with questions and never touches the work — the "ask, never build" rule applies here as it does across the whole event. Why the gate exists: AI collapses collective diversity when divergence is not engineered — it is the most corroborated finding in the Challenge corpus, and this gate is its countermeasure.
0:50	10 min	Process Gate	Each team declares HOW it will work: a framework from the taught set, a combination, or its own way said in its own words — with the alternatives it looked at and why it chose that one. Neither which framework it picks nor how faithfully it follows it is ever evaluated.
See more What is declared. The team says out loud how it plans to work: it can take one of the frameworks taught in the Gym, combine several, or describe its own way in its own words. What is asked is that it has looked at alternatives and can say why it chose its own. What is NEVER evaluated: which framework it chose, or how faithfully it followed it. Only the deliberateness of the adoption is read. Skip notice. Skipping a step is allowed — but it is accounted for: a reasoned skip stands; a skip the team did not notice weighs toward a HOLD at the defense. At primary the gate is playful and pass-only: it is a conversation, nobody is ever held back. And at the defense: the team presents its work in the terms of its own framework — the one it declared here — and accounts for what it skipped. The process account is part of the defense.
1:00	50 min	Building with AI	Free tools available to everyone. The creation record is filled in while working, not at the end. ⚡ Mid-build, the perturbation arrives.
See more The rules of the block: free tools verified that morning; the guidance guides with questions and never touches the work; the creation record is filled in while working (consequential moments, not transcripts); the first version is saved exactly as it came out. The internal rhythm is the rounds: generate → audit against sources → redo with judgment, as many times as needed. Mid-block the perturbation arrives — identical for everyone, out loud and in writing — and is answered by adapting the design, not by trimming the ambition.
1:50	10 min	Preparing the defense	What will be demonstrated, who answers what. Every member speaks.
See more What is decided here: what is demonstrated live (demonstrable, not presentable), who answers what — every member speaks; a single spokesperson is a warning sign — and what the first version tells next to the final one. Anticipate the panel's question bank: which AI recommendation you rejected, what you abandoned and why, which assumption would destroy the solution, and how to explain it all without the slide's vocabulary.
2:00	40 min	Demonstration and defense	4 teams × ~9 min: live demo + panel questions. No AI present during the defense.
See more It is a gate, not a rubric row: it authenticates that the work is the team's — against AI autopilot and against mentor over-assistance, one mechanism for two failures. No AI at the table. Outcomes: Passed · In doubt (a second panel re-examines) · Not passed. In the rehearsal the panel rotates — one member from each team that is not defending, with the judges' sheet — so all twenty experience both sides of the table. The profile is recorded per dimension; it is never summed.
2:40	20 min	Debrief as educators	Hats switch: what did you feel as participants? What would you adjust as organizers? Where was the learning?
See more The hat switch is the rehearsal's second product: what you felt as participants, what you would adjust as organizers, where the learning was — and the question that closes the thesis: what was today's learning object, the artifact or the other thing? Capture it: the validation matrix on the organization page assigns each finding to its instrument and to the real-event decision it feeds. What is learned is recorded and proposed to the Challenge canon — it does not stay in the room.

Rules of play

Only free tools available to everyone. The same arsenal for every team: what differentiates the result is entirely human.
No medium is mandatory or privileged. Prototype, campaign, story, system, activity, performance — everything is admissible. What is required is that it be demonstrable.
Nothing scores by volume: not the number of tools, not the number of prompts, not the newest model. One short prompt that detects a deep flaw can evidence more capability than a chain of two hundred steps.
Zero identifiable personal data in any AI system. Scenarios use public, fictional or authorized information.
The creation record is filled in while working. Consequential moments — not transcripts. It is never graded by length.
Before building, each team declares its output format. The concept is not enough: passing the Divergence Gate, the team says WHAT will exist at the end — a script, a map, a campaign, a game, a guide — and that is the deliverable it demonstrates. No format is mandatory; declaring one is.
The solution carries its makers’ seal. Each team incorporates perspectives or elements of relevance and pertinence of its own — its classroom, its neighborhood, its experience — so the solution reflects them. A solution any other team could have made identically is not finished.
Validating the AI’s output is a central part of the process, not a formality. Reviewing, auditing and adjusting what the AI produces — errors, inadequacies, drift from instructions and from sources — is assessed work: the profile reads it in human-AI direction and in iteration.
The work goes in rounds, and the rounds are shown. No brief is solved with one request to the AI: each team presents its first version next to the final one at the defense, and names what changed between them and why.
The defense is without AI. A team that cannot account for its own work does not advance, whatever its profile.

The work cycle — iterative and progressive, by design

Every brief is worked the same way: in rounds that deepen. Each round is another pass over the learning object — that is where the learning lives, not in the artifact.

Round	What is done	What it demands of the learning object
1 · Anticipate	Before touching the AI: what we expect to come out, what we know, what will need verifying — and what format the result will have.	Stating it with what is already known — and discovering what is not.
See more Before opening the AI, the team puts in writing what it expects to come out, what it already knows and what it will have to verify. That prior framing is what the rubric looks for in its first dimension: a problem stated with judgment of one’s own, BEFORE the model proposes its own — and sustained afterwards under the pressure of what the model suggests. Anticipating also arms the audit: whoever predicted “the highest peak is around three thousand meters” has something to crash the answer against. Without prediction there is no surprise; without surprise, review is a formality. In the room: paper and conversation, zero screens. It is the Human-First Window in miniature, inside each round.
2 · Generate	The first version with AI, unpolished.	Translating intention into concrete instructions.
See more Translating intention into concrete instructions is already work with the learning object: you cannot ask well for what you do not understand. The first version is produced unpolished — it is material to think with, not a result. The temptation to watch: staying to polish the first answer. The first version exists to be audited, not defended. In the room: save v1 exactly as it came out — the rehearsal rule asks that it be shown next to the final one at the defense, naming what changed.
3 · Audit	Inspect the version against sources, data or reality — outside the model. Log every error found.	Really looking at it: every error detected is domain knowledge.
See more Inspect the version against something that is NOT the model: the authorized source, the classroom’s data, the person interviewed, the measurement. This is where the brief’s friction point lives — the place where the AI fails with apparent confidence — and every error detected is domain knowledge gained. Logging the errors is not bureaucracy: it is the evidence the rubric reads (human-AI direction is judged by decisions — what was rejected and why) and the input to the defense (“which AI recommendation did you reject most firmly?”). In the room: a living list of errors found, with the verification source next to each. Ink as you go, not memory at the end.
4 · Redo	Direct the correction: what changes, what is kept, what is discarded.	Deciding with judgment of one’s own, not accepting whatever comes out.
See more Direct the correction with judgment: what changes, what is kept, what is discarded — and why. Redoing is not repeating the request: it is deciding. The difference between “accepting whatever comes out” and “making what should come out come out” is exactly the profile’s human-direction dimension. Rounds 3 and 4 repeat as many times as needed. That the AI does not get it right the first time is the design working — and the mid-event perturbation lands here, as one more demanding lap. In the room: abandoned paths are kept, not erased: “show me something you abandoned” is a fixed panel question.
5 · Explain	Why the final version is better — and what the hardest error taught.	Being able to say it without the AI present: the proof that the knowledge stayed with the team.
See more Saying why the final version is better — and what the hardest error taught — with the tool switched off. It is the proof that the knowledge stayed with the team and not in the chat session: what the guardrails experiment measured by removing the AI at the exam. Explaining is also the gate: the no-AI defense authenticates that the work is the team’s. A team that cannot account for its process does not advance, however beautiful the artifact. In the room: every member explains something. An explanation without the slide’s vocabulary counts double.

Rounds 3 and 4 repeat as many times as needed — that the AI does not get it right the first time is the design working, not a setback. The creation record captures the laps; the rubric reads them (dimension 04: iteration and adaptation); the defense asks about them.

The stages of knowledge — so the work can be read

The rounds are the rhythm; these stages are the anatomy of knowledge work. They are a lens for reading one’s own work, never a sequence to perform: they can overlap, repeat and loop back. What is asked is to be able to point them out at the defense, out loud — in the record, tagging them is optional and never required — because in rehearsal we observed that when the stages are not named, they blur, and validation disappears first.

Stage	What is done — and what it demands of the team
1 · Research	Gathering the sources and the starting material. The AI can propose where to look; deciding where to look is the team’s.
2 · Source validation	Which sources deserve trust and why. An unevaluated source is not a source: it is a rumor with formatting.
3 · Data preparation	Extracting the data from the research · analyzing it · validating it · selecting what serves · and structuring it as the solution’s Structured Data Source. It is the most invisible stage — and where it shows most who really worked.
4 · Insight generation	Processing and analyzing the Structured Source to produce information and insight: what the data says that could not be seen at a glance.
5 · Insight validation	Does the insight hold against the sources and against reality? An unvalidated insight is an opinion with a chart.
6 · Presentation	The information and the knowledge take the format declared at the Gate. The presentation serves the insight — not the other way around.
7 · AI-output audit	Errors, inadequacies, drift from the instructions and from the sources. Transversal: it happens at every stage where the AI contributed, and it is a central part of the process — not a final formality.

The discernment test: at the end, the team can order its own work and say what was researching · extracting · validating · analyzing · manipulating · presenting · auditing. If everything feels like one mass of “using the AI”, the stages were not worked — they were crossed.

The taught framework set — and the right to use none of them

In the Gym and in this kit, several ways of structuring the work are taught. They are taught as a repertoire, not as a mandatory method: at the Process Gate each team declares which it takes, how it combines them, or describes its own way in its own words.

Framework	What it offers
The AI-augmented work lens	Framing first, then working through the knowledge stages of the table above — with the double validation (of sources and of insight) and the transversal AI audit as its hallmark. It is one among several, not the default option. A Smoother design position — ours, not research-backed.
Design thinking	As non-linear abilities — empathize, define, ideate, prototype, test in whatever order the problem asks — never as a fixed recipe of stages to be exhibited. External practice tradition, adapted.
The inquiry cycle	Ask, investigate, interpret, conclude, ask again. Useful when the brief is above all an open question. Documented practice, public domain.
The engineering design cycle	Define the problem and its constraints, design, build, test, improve. Useful when something has to work. Documented practice, public domain.
…or your own	Always available. Describing your own way in your own words counts exactly as much as taking any of the above.

These frameworks are taught here — in the preparation — and never appear inside a brief: the brief poses the problem, not the way to solve it. And neither which framework a team chooses nor how faithfully it follows it is ever evaluated — only the deliberateness of the choice, declared at the Process Gate.

The briefs — sealed until the clock starts

On the day, one is used (decided that morning); the rest remains as the rehearsal’s catalogue. All share the same architecture: an identified learning objective — without one, a brief can be designed and solved without touching any knowledge or curriculum (rehearsal finding) — authentic context, incomplete information, values in tension, local information the AI does not have, a demonstrable artifact, and an ethical dilemma there is no escaping.

BRIEF A Your own classroom — sealed

The activity AI cannot finish

Context. You are a design team from your own school. The leadership asks for a real answer, not a speech: students already use AI to complete assignments, and the current assignments — answering what, where, when, who — AI solves in seconds, just as a search engine has for three decades.

The brief. Design and demonstrate a learning activity for a subject and grade you actually teach, in which AI is the instrument with which the student interacts with the learning object — and in which a generic AI answer is insufficient by design.

Local information you must supply (the AI does not have it): the concrete reality of your classroom — the real group, its real resources, what actually happens when you set an assignment.

The tension you cannot dodge: the activity must also work for the student who has no device or connection at home. If your design excludes them, that exclusion is part of the result.

What is demonstrated at the defense: a portion of the activity, run live with the panel playing students.

Learning objective: By the end, each participant masters designing an activity where a generic AI answer is insufficient by design, anchored to a curricular objective of their own subject.

Deliverable, format and route

Format — choose one and declare it at the Gate: a ready-to-use activity script · a slide sequence with instructions · a printable work station · a short instructional video — or another you propose.

The deliverable must contain: the learning objective · the instruction as the student would read it · the AI’s exact role (what it does and what it does NOT do) · the path for those without a device · how learning is recognized.

Suggested route: 1) fix subject, grade and objective · 2) three concepts with candidate formats (Gate) · 3) draft with AI · 4) audit it against your real classroom — the group, the resources, what actually happens · 5) final version and rehearsal of the live demonstration.

BRIEF B The community next door — sealed

A problem with a first and last name

Context. Real problems do not come in textbook statements: they live in the experience of concrete people.

The brief. Interview a member of another team (15 minutes, inside the Human-First Window) about a concrete, everyday problem of their community or sector. Design and demonstrate, creating with AI, a solution that person would recognize as their own — not a solution that would read the same in any country.

Local information you must supply: the interview is the source. What the person said — and what they did not say and had to be asked again — is in no model.

The tension you cannot dodge: the most persuasive version of your solution (the campaign, the story, the image) risks exposing or caricaturing the person and their community. Decide where the line is, and let the decision show in the design.

What is demonstrated at the defense: the artifact doing what it says it does, and the interviewed person on the panel, with the right of reply.

Learning objective: By the end, each participant masters gathering real local information (the interview) and translating it into a solution the community recognizes as its own — inquiry, design and the ethics of representation.

Deliverable, format and route

Format — choose one and declare it at the Gate: a campaign (posters + messages) · a narrated service prototype · an illustrated story · a one-page proposal — or another you propose.

The deliverable must contain: the problem in the interviewed person’s words · the solution and who uses it · how they would recognize it as their own · the ethical line you decided not to cross, visible in the design.

Suggested route: 1) interview (in the Human-First Window) · 2) three concepts with candidate formats (Gate) · 3) build with AI · 4) audit against what the person SAID — not against the model’s generic neighborhood · 5) validate with the person before the defense.

BRIEF C The maps, 30 years later — sealed

The maps exam, 2026 edition

Context. The classic exam: draw the map of the country’s mountain ranges. Today AI “draws the map” in seconds — and draws it plausibly wrong. This brief is today’s version of that exam: the map is still not the point.

The brief. Produce with AI: (1) a map of the country’s mountain ranges with the ten highest peaks correctly placed and altitudes verified against authorized sources — not against the model’s memory; (2) a line connecting the peaks in descending order of altitude; (3) an infographic comparing our ranges with the Andes, the Alps and the Himalayas; (4) a “selfie” of a team member at the summit of the highest peak — environmentally and visually correct for that real summit.

Local information you must supply: the authorized verification sources and the criterion for arbitrating when the sources disagree.

The tension you cannot dodge: the final product must also be usable by a student with a visual disability. Accessibility is not an annex: it is part of the design, and it costs.

What is demonstrated at the defense: the pieces, plus the record of what the AI produced on the first attempt, what was wrong, and how you detected it.

Learning objective: By the end, each participant masters the physical geography worked (ranges, peaks, altitudes) and the method of verifying AI outputs against authorized sources.

Deliverable, format and route

Format — choose one and declare it at the Gate: a comparative poster (map + line + infographic) · a visual presentation · an atlas notebook · an interactive piece — or another you propose.

The deliverable must contain: the map with the ten verified peaks · the descending altitude line · the comparison with the Andes, Alps and Himalayas · the environmentally correct selfie · the list of the AI’s errors with the source that exposed them.

Suggested route: 1) authorized sources first, prediction on paper · 2) three treatments with candidate formats (Gate) · 3) generate v1 · 4) audit peak by peak against the sources · 5) redo until the map satisfies the evidence — not the model.

BRIEF D The impossible timeline — sealed

History you can defend

The brief. Choose a period of Dominican history you teach. Produce with AI your own version of the period: an illustrated timeline where every date, figure and scene is verified against sources outside the model — and where two accounts that do not agree (the textbook’s and another source’s) appear side by side, with the discrepancy explained.

Local information: the sources you use in your classroom, and the real discrepancy you find between them.

The tension: telling the period honestly requires including those the school narrative usually leaves out. Decide who you reintroduce, and let it show.

Demonstrated: the timeline, plus the plausible errors the AI produced (anachronisms, invented scenes) and how you detected them.

Learning objective: By the end, each participant masters working historical facts with sources, handling conflicting accounts and detecting plausible anachronisms in generated material.

Deliverable, format and route

Format — choose one and declare it at the Gate: a mural timeline · a visual card narrative · a digital exhibition — or another you propose.

The deliverable must contain: the facts with their source · the discrepancy between accounts side by side, explained · who you reintroduced into the narrative and why · the AI’s anachronisms detected.

Suggested route: 1) period and classroom sources · 2) three treatments with candidate formats (Gate) · 3) generate · 4) verify date by date, image by image · 5) assemble the final version with the discrepancy as protagonist.

BRIEF E The ecosystem that does exist — sealed

Natural science without invented species

The brief. Build with AI your own version of a local ecosystem (a visual guide, a game, a trail) where every species shown actually lives there, verified against sources; compare that ecosystem with one from another continent; and turn the comparison into an activity your students could do.

Local information: the chosen ecosystem (the schoolyard counts) and the authorized source for its species.

The tension: the most spectacular version of the material (the megafauna, the exotic) is the least faithful to the real ecosystem. Fidelity versus spectacle — decide and defend.

Demonstrated: the material working, plus the inventory of species or details the AI invented and you discarded.

Learning objective: By the end, each participant masters the local ecosystem worked (real species, with sources) and the auditing of AI-generated science content.

Deliverable, format and route

Format — choose one and declare it at the Gate: a visual field guide · a game · a signposted trail · a mini-atlas — or another you propose.

The deliverable must contain: the verified species with their source · the comparison with an ecosystem from another continent · the derived activity your students could do · the inventory of what the AI invented and you discarded.

Suggested route: 1) ecosystem and species source · 2) three concepts with candidate formats (Gate) · 3) generate · 4) audit species by species · 5) try the derived activity with another team before the defense.

BRIEF F Corner-store mathematics — sealed

The neighborhood’s numbers, checked by hand

The brief. With real prices and quantities the team supplies (the corner store, the week’s shopping, the electricity bill), produce with AI your own version of a leveled mathematical problem situation — and verify by hand every operation the AI presents, documenting the wrong ones. Compare two versions of the same problem (one generated, one corrected) and explain what changed.

Local information: the real prices — the model does not know your neighborhood’s corner store.

The tension: the most realistic problem exposes the finances of a recognizable family. Anonymize without the problem losing its truth.

Demonstrated: the problem situation playable before the panel, plus the list of arithmetic or context errors the AI produced.

Learning objective: By the end, each participant masters the operations worked with real data and the manual checking of AI-generated arithmetic — calculation as judgment, not as faith.

Deliverable, format and route

Format — choose one and declare it at the Gate: a printable set of problem situations · a card or board game · a leveled worksheet — or another you propose.

The deliverable must contain: the real data anonymized · problems at at least two difficulty levels · the generated and the corrected version, side by side · the arithmetic or context errors found, with the calculation that exposed them.

Suggested route: 1) gather the real data · 2) three concepts with candidate formats (Gate) · 3) generate · 4) verify every operation by hand · 5) play the problem among yourselves before playing it for the panel.

GUIDING TEAM ONLY Perturbations and friction points — do not open with participants present

At the real event this lives in the authoring record, never on the participant’s page. It is published here collapsed because this rehearsal trains both roles. The perturbation is delivered mid-build (~1:20), identical for every team, out loud and in writing.

Perturbation per brief

Brief A: “The leadership has just banned phones in the classroom. Your activity must work all the same.” (The assumed resource is cut; the reasoned adaptation is the evidence.)
Brief B: “The person most affected by the problem cannot read. Your solution must reach them.” (From the canonical perturbation menu; it stresses the literacy assumption.)
Brief C: “Two authorized sources give different altitudes for one of your peaks. Decide which to use and defend the criterion.” (New evidence contradicts an assumed datum — and it is a real disagreement between sources, not an invented one.)
Brief D: “A newly surfaced primary source contradicts a central date or fact of your timeline. Incorporate it.”
Brief E: “Your material must now also work offline, on paper or off-screen.”
Brief F: “The family budget in the problem is cut in half. Recalculate, and the problem must remain teachable.”

Declared friction point (where the AI fails, predictably)

Brief A: the AI proposes template activities from the internet’s average; it does not know the real group or its constraints. The friction appears when landing on the concrete classroom.
Brief B: the model was not at the interview. It will fill the gaps with plausible stereotypes of the “generic neighborhood” — detecting and correcting that is the learning.
Brief C: generators produce plausibly incorrect maps and landscapes: misplaced peaks, altitudes from memory, tropical summits with eternal snow. Verification against sources is the real learning object.
Brief D: historical images with plausible anachronisms and accounts that average conflicting versions into one smooth “truth”. Detecting the smoothing is the learning.
Brief E: the model populates ecosystems with species that do not live there (or do not exist) with total apparent confidence.
Brief F: arithmetic with intermittent errors and “reasonable” prices that are not the neighborhood’s. The manual check is the point.

The creation record — consequential moments

It is not a transcript or an exhaustive log: it is the record of the moments where judgment was exercised. It is never graded by length, and completing it by stage is never required. One row per moment; the stage column (research · extract · validate · analyze · manipulate · present · audit) may be marked — it is an aid for reading one’s own work, not a requirement, and it is never counted:

Stage (optional)	Moment	Human decision	AI contribution	What was rejected and why

The defense — no AI at the table

Each team demonstrates its work and answers panel questions. Every member answers — a single spokesperson is a warning sign. The defense is a gate, not one more rubric row: it authenticates that the work is the team’s. Possible outcomes: Passed · In doubt (a second panel re-examines) · Not passed.

Questions from the panel bank:

Which AI recommendation did you reject most firmly, and what did you do instead?
Show me something you abandoned. Why?
Which assumption, if false, would destroy your solution?
Explain this without using your slide’s vocabulary.
Who is worse off if this works exactly as you designed it?

Judges’ sheet — a profile, never a score

One level per row. The rows are never summed or averaged — the result is the whole profile. The weight indicates the judges’ emphasis of attention, not a coefficient. Process 60 · Product 40.

Team: ______________________ Brief: ____ Defense gate: ☐ Passed ☐ In doubt ☐ Not passed

Dimension (emphasis)	Not achieved	Developing	Achieved	Mastered
Problem framing and human insight Process · 12Did they state the problem BEFORE consulting the AI, and hold that framing under the model’s pressure?	☐	☐	☐	☐
Creative divergence and originality Process · 15Were there three genuinely different approaches, and was the final concept CHOSEN from a space of possibilities — not accepted from the first result?	☐	☐	☐	☐
Human-AI direction Process · 15It does not matter what percentage the AI did — what matters is who exercised the judgment. What did they reject, verify, transform?Adoption sub-anchor (ALC-RB-03, D22): what was declared at the Process Gate is scored here, and ONLY the deliberateness of the adoption — never WHICH framework the team chose, never how faithfully it followed it, never how well it narrated it.	☐	☐	☐	☐
Iteration and adaptation Process · 12Did they deliberately abandon a path, with reasons? Did the perturbation produce a reasoned adaptation?	☐	☐	☐	☐
Collaboration and transparency Process · 6Does each member hold a defensible part? Is the attribution (to people and to tools) honest, and made as they worked?	☐	☐	☐	☐
Demonstrated value of the artifact Product · 15Was it shown doing what it says it does — or was it a presentation about something that does not exist?	☐	☐	☐	☐
Fit to the local context Product · 10Could this solution only have been designed by people who know this context — or would it read the same in any country?	☐	☐	☐	☐
Communication and quality of defense Product · 10Can they explain it without the slide’s vocabulary? Do unexpected questions get answers or reformulations?	☐	☐	☐	☐
Ethical judgment in consequential choices Product · 5The dilemma is INSIDE the brief. Did the solution visibly change because of the position they took — or was ethics a paragraph glued on at the end?	☐	☐	☐	☐

Negative rules for the judges — binding: no points for the number of AI tools; no prompt-engineering category; no prize for the newest model; no assuming more AI is better than less; no authorship percentages or numeric provenance formulas.

Debrief — back in the organizers’ role

At what moment did you feel you were learning the learning object — and at which were you only producing an artifact?
Did the perturbation hurt? Where? (That is where the real design was.)
Which defense question could you not answer — and what does that say about the process?
How would a student cheat in this format — and what would the gate detect?
What would you remove or change in the format before putting it in front of real students?
In your own words: what was today’s learning object — the artifact, or the other thing?

Meet the aiLearning Challenge Read the full announcement

The aiLearning Challenge is an initiative of aiLearning.global, CEMI.ai and CEMI Labs — created by Carlos Miranda Levy, applying the Smoother learning methodology. This rehearsal derives its mechanics from the Challenge canon.