The student does the whole inquiry. The AI supports every step, and never writes it.
She does her own real coursework inside a project, turns methods into habits with the thinking cards, and ends up with a report that describes how she thought. Three parts, best read in that order.
From research question to finished draft — support at all six stages
EE, TOK, IA, a research project, a science-fair entry — the form is open. What the project space gives is a route that can actually be finished: six stations carry the structure, and cards arrive as needed. A single course usually calls on three to five of them.
Question
Narrow an interest into an answerable question — object, scope, comparison.
Sources
Move from finding material to making sources argue with each other.
Method
Compare candidate methods; say why this one answers this question.
Data
Work her own evidence — cleaning, analysis, and how certain she is.
Writing
Build the argument, take the counter-case, bound the conclusion.
Managing
Throughout: plan, notice the plan failing, adjust.
Every claim can be traced back to where it started
A post is a retelling; the study behind it is the evidence. With the only three searchable features in an unsourced post, the student found the NASA release herself, then the Nature Sustainability paper behind it.
SIFT
Stop · investigate the source · find better coverage · trace to the original context. It settles whether they said it.
CRAAP / CRRAAB
Currency, relevance, authority, accuracy, purpose. For personal testimony the fifth swaps purpose for bias — the idea of authority bends under load.
Switch the tool with the object
Ride the lateral tool all the way and you stop at “Nature published it, so it's fine”. Once you reach the primary paper you must switch back to vertical and ask whether it holds.
Tracing is only half of it. What got lost between the retelling and the original?
The gap concentrates in three things: the source, the caveats and the frame. Neither swallowing it nor dismissing it requires reading the original. Only wanting to actually know does.
| Kind of gap | What the original says | What the post says |
|---|---|---|
| 1. Uncheckable The source vanishes | NASA Ames release, the Nature Sustainability paper, authors and institutions | No source at all — just “images and text from the internet” |
| 2. A checkable error The denominator is swapped | China is 25% of the global net leaf-area gain. The 42% is forest programmes' share of China's own greening | “Over 40% of the world's greening comes from China” |
| 3. Not fully checkable The caveats vanish | It can't offset tropical vegetation loss; India's gains lean on groundwater and may reverse | Both caveats are simply gone |
| 4. Checkable but misread The frame is displaced | Human activity “also contributes”; climate and CO₂ fertilisation still appear dominant | “The real cause isn't population but tree planting” — a conjunction rewritten as a substitution |
40% and 42% are two points apart — a glance reads them as the same number
But the denominator moved from “China's greening” to “the world's greening”. The number was copied; the meaning doubled. The error is checkable, unfakeable and genuinely deceptive — anyone can say they check data. Catching this is what counts.
The rabbit-hole log
Tracing a source opens side branches. They have a real trigger and real value; what they need is a boundary. The system records the trigger, the side question, its relation to the main one, and why she left.
Leaving in time is also evidence of autonomy
A1 measures who sets the direction, not how far you went. Taking it all on board means someone else set it. Drifting means nobody set it — including her. Neither counts as autonomy.
The student writes the prose. The AI only helps her think it through.
There's no formatting here, and no “submit to school” button. We occupy the thinking layer — her essay lives in her own world, and that file isn't ours.
One continuous conversation
Framing, planning, sources, reading, writing and review are surfaces of the same Imprint, with one agent behind all of them. Changing rooms moves the camera; it carries the whole memory along.
One project spine
Proposal, plan, reading list, outline, activity log — every surface reads and writes one durable state. Solidified dialogue folds into the spine: the thread stays lean and nothing is lost.
There is exactly one red line
The AI may explain a concept, generate search terms, play the opposition, flag a broken draft. It may not set her direction, make her core judgment, or produce submittable prose.
The hardest AI risk to catch is that it agrees with you
A student offers a claim that doesn't stand, and it supplies the reasons. A student hints at a preference, and it shifts to match. Seeing that and holding against it is the hardest and most important move of this era — and it only shows up in tasks long and open enough for the AI to really have the chance to agree.
A course coaches one thinking method until she can use it
Every slice is a deterministic state machine: play the clip, wait for it to end, hand over a card, wait for her answer, then move on. She makes calls inside the course, and what she leaves with is a handful of cards.
A card is a method, packaged
It breaks a thinking method into visible steps. Open one and the AI steps back from answer-giver to midwife: it questions, it does not deliver.
She holds the summons
The trigger is automatic; opening it is her call. Challenging an AI takes no courage, and neither does inviting one. What's left is whether she wants to get it right.
The tap itself is data
Offered by the system: logged as prompted. Summoned with nobody asking: logged as self-initiated. The whole A2/A4 signal lives in that one difference.
Same material, different lens, different thing seen
A lens only tells you what to look for in the text; the answer is still yours to reach. So every use has to land on one specific sentence someone else can check.
| Lens | What it interrogates |
|---|---|
| Logic | Does the evidence actually yield the conclusion — hidden premises, causal leaps, overreach. |
| Scientific method | How the data was obtained; whether measurement, sample and error support the claim. |
| Sociology | How power, resources, institutions and social position land differently on different groups. |
| Law | Whether rights, duties, procedure and remedy are actually spelled out. |
| Economics | How a rule shifts choices, incentives, costs and allocation. |
| Ethics | Which values a claim protects or sacrifices — fairness, harm, autonomy, dignity. |
| History | What period conditions the conclusion rests on, and which have since changed. |
| Communication | How headline, wording, framing and absent voices steer the reader. |
| Systems | How a local action produces knock-on effects, delays and feedback loops. |
Read how she thought, straight from the record
Judgment is an aggregate, and a single number won't honestly capture it; a product claiming to score it precisely is overselling. We measure and train the two things underneath it that can actually be observed.
“Analysis” can stop at naming the parts, or reach the relations between them, or reach the assumptions the whole frame rests on.
Six signal families share one denominator: does it start when nobody asked. Readings compare only against her own baseline.
A concession paragraph is two kinds of evidence at once
The depth axis asks how well the check was done. The autonomy axis asks who started it. One action, two readings — by design, not by oversight.
Challenging an AI takes no courage — only a reason
Classroom behaviour is tangled with personality, cultural norms and verbal confidence. The AI setting weakens all three at once — and once courage stops being the barrier, what's left is closer to whether she has a reason and wants to get it right.
“Your Mind Imprint” has nine parts
The first four parts are counted facts. The last five are grounded judgments. Every judgment must point back to a real fragment — the ones that can't are dropped after generation.
Basics
Title, dates, milestones, and counted numbers: AI turns, materials, words, comments, revisions.
Abstract
A paragraph on the route she actually took, plus one line each on materials, writing and AI use.
Timeline
Chat, reading room, map, writing, retrospective — laid out in order.
Materials
Every source: when it arrived, where from, and what job it does in the argument.
Depth D1–D6
Per dimension: a level, a verdict, clickable evidence, one suggestion.
Autonomy A1–A6
Same structure — plus whether each move was self-initiated or prompted.
Prompt lens
3–10 of her actual prompts, with what worked and what to change.
Cards & sub-agents
Which card, at which station, why it was offered, and what changed downstream.
Risk signals
Ghost-writing, missing sources, broken warrants, shifting denominators, tangents — each with advice.
On levels: we don't display an L1–L4 ladder. Readings are shown semantically and by colour. No single session decides a level; trajectory outranks snapshot; and everything is phrased as “observed this cycle”, never as a trait claim.
For the first time, teachers can see how the work took shape
Until now only the result was visible; the process happened out of sight. Now every step is on the record, and a teacher sees how the class is actually working.
The teacher designs the opportunity
How open the task is, whether the material conflicts, whether a counter-case is in the room — the teacher decides all of it.
The teacher reads the evidence
The system gives a reading; the teacher gives the interpretation. Raw chat, source records and draft changes all open up.
The teacher writes the prescription
“Underperforming” has many causes. Some need to be let go of, some need method, some need a different subject. The platform locates it; the teacher decides.
The teacher trains accountability
Explaining a conclusion, answering follow-ups, conceding a limit — that only happens between people.
| Role | Sees | Never sees |
|---|---|---|
| The student | Her own growth report — trajectory, highlights, next step — and her own conversation detail | Internal routing labels; any comparison against others |
| Parents | The cycle's growth narrative and readiness | Tier codes, quadrant labels, any ranking; conversation detail unless she authorises it |
| Teachers | Judgments with evidence links to the raw fragment — and a one-click override | Content detail unrelated to teaching |
| School admins | Class-level aggregates | Any individual's conversation detail |
One process record, mapped onto three official rubrics
On the person's side, the two axes stay constant across systems. On the work's side, we speak only that system's official language and report readiness — which pieces of evidence are still missing. No predicted grades, no conversions, no composite.
International · seven subjects
IGCSE GP, A Level GPR, EPQ, AP Seminar, AP Research, IB TOK, IB EE — every course that tests inquiry and argument rather than recall.
Seven mark schemes; strip the vocabulary and they test one path. That's where D1–D6 comes from.
China's reformed exams
“No thinking, no question.” Open-book reform moved memory out of the exam hall — what's left to differentiate is thinking. You turn pages for evidence, not for answers.
We never predict papers. We do bet on how they're set — on the ruler behind the principles.
Careers
No career matching. Choosing subjects or a major is a judgment act — and subject choice is the first high-stakes one. It shouldn't be the first time judgment gets practised.
The A axis maps what she's drawn to; the D axis maps what she's good at. The overlap is where the conversation belongs.
Walk it yourself first. Then decide whether it belongs in your classroom.
The demo runs on real material: an unsourced post, traced all the way back to NASA and Nature Sustainability. You'll see when a card gets handed over — and what the final report says.