Mian Zhang

Public Release - 2026-07-14

An AI-assisted answer is not evidence of learning until transfer, error memory, and ownership are recorded.

Artifact: learning_integrity_loop_v1. This note separates a visible answer from a learning claim: the learner must have a recorded attempt, an assistance level, an explanation without AI, a memory of the error, a transfer test, and an accountable feedback owner before the result receives additional authority.

claim_id learner_attempt ai_assistance_level explanation_without_ai error_memory transfer_test feedback_owner evidence_timestamp boundary_update review_status
Evidence Map Registries Boundaries Counterexamples
Learning Integrity Loop v1 public evidence visual.

Artifact

The output can be correct while the learning claim remains unproven.

AI can produce a polished answer, a high score, or a plausible explanation without preserving the learner's attempt, correction path, or ownership of the result. The review object separates answer quality from learning integrity before a school, parent, tutor, or platform treats the result as evidence of capability.

The minimum record is attempt, assistance, explanation, error memory, transfer, and feedback.

Learning Route

The minimum record is attempt, assistance, explanation, error memory, transfer, and feedback.

The protocol does not treat all AI help as cheating. It asks whether the assistance level is visible, whether the learner can explain the result without AI, whether the original error is still available for correction, and whether a transfer test confirms that the skill survives outside the generated answer.

If the tool hides the mistake, the assessment boundary has to move.

Review

If the tool hides the mistake, the assessment boundary has to move.

A useful challenge should identify a missing field, an unsupported causal link, a stale evidence timestamp, or a condition under which the control fails. The boundary update is triggered when AI assistance removes the learner's failure memory or makes the feedback owner unaccountable.

1

claim_id

The learning, assessment, tutoring, homework, or capability claim under review.

2

learner_attempt

The learner's first answer, draft, calculation, explanation, or attempt before AI help.

3

ai_assistance_level

The level of AI contribution: none, hint, rewrite, worked solution, generated answer, or full automation.

4

explanation_without_ai

The learner's ability to explain the result after closing the tool or removing generated text.

5

error_memory

The mistake, failed step, correction, or misconception that remains visible for future learning.

6

transfer_test

A new problem, oral check, variation, or delayed task that tests whether understanding transfers.

7

feedback_owner

The teacher, mentor, parent, reviewer, system, or institution responsible for correction.

8

evidence_timestamp

The date, assessment context, source record, or controversy window that anchors the claim.

9

boundary_update

The scope change required when AI assistance hides effort, error, or actual understanding.

10

review_status

The recorded state: supported, narrowed, pending, or retired.