Research

The claim we intend to earn

We have not run a pilot yet. This page says what we are measuring and how, so that when there are results you can judge whether they were measured honestly. It will be replaced with outcomes, not testimonials.

The question

Does explicit instruction in C5 improve middle-school students’ ability to use generative AI critically and independently?

The method

Students complete a realistic scenario before instruction and a structurally matched scenario after. Both are scored against the same four-level rubric by the same evaluator, using the same evidence rules. Growth is reported per competency.

Deterministic behavioral signals are recorded alongside the rubric — did the student supply constraints, ask a follow-up, seek evidence, decide, justify. Those are countable without a model in the loop, which makes them the harder number to argue with.

What gets reported

  • Growth per competency across the five C5 skills.
  • AI deference rate — the share of missions ending on the AI’s recommendation without meaningful challenge. The number we expect to fall.
  • Human override quality — when a student rejects AI advice, the rubric level of their justification.
  • Evidence-seeking frequency — how often a student asks what a claim rests on.

The rubric, in full

Published because a scoring system you can’t inspect is a scoring system you shouldn’t trust.

01

Curiosity

Depth and usefulness of the questions a student asks.
Emerging
Asks only the obvious opening question.
Developing
Uses some follow-up questions.
Proficient
Identifies missing information and explores useful alternatives.
Advanced
Asks questions that reveal assumptions, tradeoffs, or unseen possibilities.
02

Clarity

How well a student communicates objective, context, and constraints.
Emerging
Request is vague or underspecified.
Developing
Includes some context or a goal.
Proficient
Clearly communicates objective, context, and the important constraints.
Advanced
Prioritizes competing constraints and adapts instructions as the problem evolves.
03

Creativity

Breadth and originality of the alternatives a student explores.
Emerging
Accepts the first solution.
Developing
Requests alternatives.
Proficient
Explores multiple meaningfully different approaches.
Advanced
Combines, transforms, or extends ideas into something original.
04

Critique

Ability to find errors, omissions, assumptions, and evidence gaps.
Emerging
Accepts output at face value.
Developing
Notices obvious problems.
Proficient
Identifies assumptions, missing information, weak reasoning, or questionable claims.
Advanced
Systematically tests evidence, counterarguments, and alternative explanations.
05

Choice

Quality and justification of the student’s own decision.
Emerging
Uses the AI recommendation as the final answer, unexplained.
Developing
Makes a choice with limited rationale.
Proficient
Makes an independent choice supported by relevant reasoning.
Advanced
Balances tradeoffs, rejects weak AI advice where warranted, and explains the decision.