Inside the Right Game Five small moves produce a classroom that mostly grades itself.…
Published on designingintelligence.substack.com · Read the piece ↗
Designing Intelligence
Inside the Right Game
A design for AI-era classrooms: five rules that make the most efficient way to cheat indistinguishable from learning the material.
By Jason Prunty · Published 23 Jun 2026designingintelligence.substack.com
This listing exists because its author asked for it, after one analysis run cleared both of the showcase’s bars: Idea Originality of 75+ and model accessibility of L4+. Its published link has not been through the submission queue. The draft text is not republished here.
Headline readings
Idea originality
79/100
Substantially beyond what search and models produced
Model accessibility
Level 5
Not produced within the tested baseline
Idea quality
63/100
Holds together; some steps asserted rather than argued
Search accessibility
Easy
Easy — ordinary searches readily surface this argument
The short answer
The topic and general position are common, but within the sources and model samples examined, the specific synthesis — peer-uptake grading plus rotating roles plus artifact version history closing the gap between faking and learning — does not appear assembled this way elsewhere.
Within the sources searched, AI-driven classroom cheating and institutional responses (proctoring, honor codes, private AI tutors, portfolio/process-transparent assessment like AP Capstone) are well-documented concerns, and the protégé effect is a repeatedly explained psychological mechanism across multiple independent sources. Version-history checking as a signal of authentic student process also has a direct precedent in a classroom-tech tutorial. These are the load-bearing components the draft's mechanism depends on, and none of them individually is rare. What the search results do not show, in any single source, is these components combined into a five-rule system whose explicit claim is that peer-uptake grading makes the cheapest way to fake competence structurally identical to the work of mastering it — framed as resistance to Goodhart's Law. Model-generated samples produce adjacent theses (transparent AI logs, portfolio-based assessment, flipping incentives so struggle becomes credentialed) but stop at 'partial_overlap' with low overlap levels, and do not reach the specific peer-uptake-as-grading-proxy plus rotating-roles plus protégé-effect chain, nor the version-history-as-exposure claim tied to that chain. The deterministic scoring reflects this: synthesis facet is highest (96) while mechanism (63) and evidence (68) trail, consistent with a case where the individual pieces are established but their causal chaining together is comparatively less supported by direct evidence in the draft itself. Several of the highest-rarity supporting and implication claims — the teacher's shift to watching avoidance patterns, the FERPA/credentialing-as-bottleneck argument, and the claim that rotating roles specifically breaks collusion — are flagged as unsupported in the draft as written, meaning their rarity should be read as 'not found elsewhere' rather than as evidence of soundness.
Where the originality sits
Originality by facet
Each facet blends search rarity, model rarity and semantic rarity.
Model-accessibility level, by facet
The cheapest prompt tier at which the baseline reached each facet.
Level 5 means “not produced during this analysis” — a statement about these samples, not about the model’s knowledge.
Every number here describes rarity within the sources searched during one run, and the baseline samples that completed. An absent match is weak evidence, never proof that an idea has no precedent — this page makes no claim that nobody has had this thought before. See the methodology for what these readings can and cannot mean, or the example report for a full analysis read end to end.