Domain · Education (COE)
Assess the process, not the essay.
When AI can write the assignment, grading the output tells you nothing. CO for Education applies the Mirror Thesis to the classroom: let AI handle execution, and what remains visible is the student's judgment. The deliberation log — how the student directed, corrected, and challenged the AI — becomes the assessment artifact.
Status: emerging application, at pilot stage. Position papers in preparation (AIED / AIES 2026); pilot data targeted for 2027. This page describes the methodology, not field results.
§ · The GenAI assessment crisis
Every other answer fails. One doesn't.
- Ban AI An unwinnable detection arms race — and it's biased against non-native speakers.
- Allow AI, grade the output You can't tell understanding from delegation. The essay looks the same either way.
- Allow AI, add oral exams Doesn't scale — 50+ hours per assignment for a class of 200.
- COE — assess the process The deliberation log reveals the student's thinking. That is the thing worth grading.
§ · The Mirror Thesis, applied
What the machine can't do is the thing to grade.
The Mirror Thesis holds that as AI absorbs execution, it reflects back the uniquely human contribution. In a classroom that contribution is judgment: knowing when the AI is wrong, redirecting it, refusing its overclaims, deciding what matters.
So COE moves the assessment object. Not the essay — the deliberation log that produced it. The log is a natural byproduct of working through a CO setup, so measuring it doesn't add surveillance; it reads what already happened.
Output can be delegated. Judgment, on the record, cannot.
§ · The COE Spectrum
Four levels of student freedom.
How much of the CO setup does the student build? The answer sets what you're actually testing — from working inside a given system to designing one for others.
- 1 · Full Constraint The student works within the instructor's CO setup. Tests working WITH AI
- 2 · Partial Freedom The student designs the agents and the knowledge base. Tests designing FOR AI
- 3 · Full Freedom The student builds the entire CO setup. Tests building AI systems
- 4 · Meta The student designs a setup for someone else to use. Tests teaching through AI systems
§ · What the log reveals
Six competencies you can finally see.
- Critical Thinking identifying when the AI was wrong
- Domain Knowledge redirecting the AI toward better arguments
- Intellectual Honesty challenging the AI's overclaims
- Creative Synthesis making substantive structural decisions
- Ethical Judgment recognizing implications the AI missed
- Contextual Wisdom bringing domain knowledge the AI lacked
§ · Integrity by design
The setup is the medium, not a camera.
The Camera Problem
Trust infrastructure must be the medium through which the work happens — not a camera pointed at the student. The CO setup is the medium; the deliberation log is a byproduct of doing the work, so integrity comes from the design, not from monitoring.
The ironic defense
Faking a convincing deliberation takes more understanding than doing the work honestly. Gaming the assessment is harder than passing it. (Qualified: this holds less for purely procedural tasks and shared templates, which need their own mitigations.)
Cryptographic integrity (EATP)
An unfakeable server-side start (Genesis Record), an explicit envelope of allowed models, tools, and time, and a continuous audit chain. Privacy-preserving by construction: hash → verify → grade → delete content — keep only the hash chain and the grade.
It scales, recursively
Instructors use CO to grade, just as students use CO to produce — Human-on-the-Loop assessment where the instructor calibrates rubrics and handles edge cases while the system prepares summaries and flags anomalies. The approach is designed to hold from 50 students to 500+.
§ · The same methodology, a different domain
One set of principles. Codegen proved it. Education extends it.
COE methodology per the Cognitive Orchestration for Education reference (Hong, 2026). CO is open, CC BY 4.0; specification of record at terrene.foundation. Education delivery is carried out by partners against the open methodology — the Foundation publishes, it does not sell courses.