What this page is for
Partners need to see a sequence they could run without buying a product. Students need to see what “done” looks like this week. This page is that sequence. It is not a six-step poster with no classroom consequence, and it is not a laboratory protocol. Sister projects that prototype objects or rerun tests are listed on the about page.
The loop in working order
- Explain. With the interface closed, the student writes what the system is doing in ordinary language. Decorative words are circled. If the sentence only works while a model is completing it, the knowledge is still on loan.
- Map. On paper, the student draws what the idea depends on, what is known, what is borrowed, and which nodes are empty. Empty nodes stay empty. A model is not allowed to fill them invisibly.
- Limit. The student writes one thing the system cannot do even when the output looks fluent, and one action that must not be sent to a model or a public form. Refusal is a rehearsed skill, not a slogan.
- Use. Only then may a tool appear, and only as an object of study: compare two outputs, mark borrowed lines, or test a claim the map already named. The tool is not the assignment.
- Test. Close the interface again. If the spoken account is thinner than the export, the next assignment is a clearer human sentence, not a more impressive tool.
- Teach. A classmate or mentor marks the account: which sentence would be false, which words are still decorative, which human decision remains. The student keeps the pen.
What a reviewer should be able to hold
After a cycle, the folder should contain an explanation card, a knowledge map, a source note, a limitation note, a short revision log, and a teach-back outline. Those artifacts are the operating evidence of HNK. They are not a certificate of AI mastery and not a psychometric score.
Families can review without becoming engineers: ask for the closed-laptop explanation, the limitation, and the sentence that changed after feedback. Schools can mark without software specialists: one concept, one artifact, one responsible-use rule.
Illustrative example (not a student case)
Teaching example only. A student pastes a generated paragraph that says “this chatbot understands your feelings.” The HNK sequence does not ask for a better prompt. It asks the student to rewrite: what the system actually does (next-token prediction on text), what it depends on (training data the student cannot see), what it cannot know (the student’s private state), and who is responsible if the sentence is used as advice. The map marks “training data” as borrowed and “student feelings” as empty. The limitation note forbids sending a classmate’s journal into the tool. The teach-back is the rewritten sentence, not the original fluency.
Where the loop is not enough
This method does not replace subject teaching in biology, history, or mathematics. It does not authorize unsupervised collection of other people’s data. It does not claim that a six-step cycle is a published standard from an outside body. External research on explanation and teach-back may be cited in resources with their own authors; HNK’s public claim is only that this is how its standing training is organized.
To discuss a module, write info@hnk.org. English is welcome. Bring the sentence the student cannot yet explain.