20-minute AI-assisted research loop

Hints, not answers.

Use AI like a tutor, not a ghostwriter. In 20 minutes you frame a research question, set one human boundary, check two sources, revise with AI, and end with a reflection that names what AI helped with, what you verified, and what remains yours. Press Start to pace yourself with the 20-minute clock, or click a phase to browse.

🔒

Privacy: The worksheet runs in this browser tab and is saved only to the NOUS OS local backend on this machine. NOUS Guide chat sends only the question and worksheet context when you click Ask, through a server-side Gateway route that keeps keys out of the browser. Software can detect emails, phones, and SSN-style patterns — but it cannot catch your name, school name, or family details, so please don't type them. Use topic area ("CRISPR ethics for biology class"), not identifying context ("Mrs. Smith's class at Lincoln HS").

First-trial run checklist

Use this page with one student or student-adjacent participant. The goal is not to grade the student; it is to see whether the protocol preserves thinking, evidence checks, and responsibility.

BeforePick one topic, keep identities out of the page, and ask a parent/teacher/researcher to observe process only.

DuringStart the timer, let the student click Ask only when ready, and require two source cards before review.

AfterOpen Session Review, export the Markdown packet, and copy it into Obsidian 04 Reviews.

NextChoose exactly one next-run change from review evidence before changing the product again.

NOUS Guide · ask any student question

NOUS Guide is optimized for this project.

NOUS Guide is the learning companion for this sandbox. Ask a research question, source question, or "what should I do next?" It will keep the session inside the NOUS OS rules: hints, boundaries, source checks, and human responsibility.

It will not write your final answer. It turns questions into subquestions, checks, and next moves.

It saves locally. Worksheet snapshots and chat turns are written to the local backend, not browser storage.

It uses the secure NOUS backend route. Gateway keys stay server-side in /api/hermes-student-agent; the browser never sees them.

NOUS Guide

NOUS OS learning guide · hints_not_answers

LLM agent

Hi, I'm NOUS Guide. Ask me your research question. I will help you frame it, set one boundary, choose source checks, and decide your next learning move — but I will not write the final answer for you.

Local backend save: waiting for notes.

Guided worksheet · no prompt engineering required

This is the execution surface. Fill in short notes here, ask NOUS Guide only when you want guided help, or copy the prompt cards into another AI tool. The worksheet is saved only through the local backend so you can recover session notes on this machine without exposing Gateway keys or using browser storage.

Session readiness deterministic checks · student clicks when ready

First pass Write a question and a prior-belief note before asking AI for a learning plan.

Second pass Add a human boundary and at least one complete source card before asking AI to revise.

Review Complete two source cards, revised plan, and all reflection fields before exporting review.

My question

Use topic area only. Do not type your full name, school, teacher, email, or family details.

What I think before asking AI

My boundary

A boundary is one thing AI is not allowed to do for you in this session.

AI plan notes

Sources I checked

Structured source cards

Source 1required for research review

Title or URL Author / institution Date Decision Evidence Uncertainty

Source 2compare against Source 1

Title or URL Author / institution Date Decision Evidence Uncertainty

Research review works best when at least two source cards have author, date, evidence, uncertainty, and a decision.

Revised plan after boundary + sources

Copy prompt to AI · learning plan

Copy prompt to AI · source critique

Copy prompt to AI · revise plan

Local session summary

Saved sessions can be reviewed locally after the backend receives a snapshot.

Open review

20-minute loop · 6 phases

0:00 / 20:00

1 Frame your question

3 min

phase id: intent

You do: Write the research question in your own words and name what you already believe.
AI helps: Suggest clearer subquestions without choosing the thesis.
You keep: goal, curiosity, and initial judgment
Output: one-sentence question + prior-belief note

2 Ask AI for a plan

4 min

phase id: ai_first_pass

You do: Ask for a learning plan, not an answer paragraph.
AI helps: Offer concepts, vocabulary, possible source types, and common traps.
You keep: which path to explore first
Output: three possible subquestions

Example prompt. "Don't answer the question yet. Give me a short plan to research it myself: the key concepts I'll need, two types of sources to check, and two common mistakes to avoid."

3 Set your boundary

3 min

phase id: human_boundary

You do: Choose one boundary: privacy, facts, learning, decision, or values.
AI helps: Restate the boundary and adapt the plan around it.
You keep: values, privacy choices, and assignment constraints
Output: boundary card

What's a boundary? Something you decide AI is not allowed to do for you in this session. Pick one:

privacy — I won't share my name, school, or family details with the AI.
facts — I care most about accuracy and source quality, not persuasion.
learning — I want to understand the topic myself, not just get a finished answer.
decision — the conclusion (what to claim or recommend) stays mine to make.
values — my own judgment about what's right or fair leads; the AI doesn't decide it.

4 Check your sources

5 min

phase id: source_check

You do: Check at least two reviewable sources for author, date, evidence, and uncertainty.
AI helps: Provide a checklist and critique weak evidence.
You keep: verification and source acceptance
Output: source checklist

Floor, not finish. Two reviewable sources is a floor, not a finish. Before moving on, say what a third source would need to show to change your conclusion.

5 Revise with AI

2 min

phase id: ai_second_pass

You do: Ask AI to revise the research plan using the chosen boundary and source notes.
AI helps: Improve next steps, practice questions, and uncertainty notes.
You keep: the final claim and whether evidence is enough
Output: revised learning plan

6 Reflect

3 min

phase id: reflection

You do: Answer the reflection card below before writing any final assignment text.
AI helps: Help compare first-pass vs second-pass behavior.
You keep: responsibility for final work and next learning move
Output: reflection card

Total: 20 minutes · AI only gives hints — it will not write your answer for you. hints_not_answers

Source checklist · for each source, twice

Original source found, not only a summary.
Author or institution is identifiable.
Date is current enough for the topic.
Evidence is separated from opinion.
Write what would change your mind.

Reflection · finish before drafting

Your answers save to the local NOUS OS backend. They are stored on this machine under the sandbox session runtime folder. Copy anything important into your own notes before you leave.

What did AI help with?

What did I verify?

What remains my responsibility?

What would I ask differently next time?

Parent / teacher observation checklist (optional)

The student could explain the research question without AI reading it back.
The student named one source-quality issue or uncertainty.
The student kept their own claim, values, and final responsibility.
The AI was used for hints, critique, or planning, not final-answer writing.

Record process observations only — not student identity or raw prompt text. The trial guide's 8-field safe-observation format is the recommended note shape.