PROOF OVER PROMPT FRAMEWORK

Proof over Prompt™

AI-Resistant Curriculum Design
What BMW's quality systems taught us about AI-resistant assessment design in education.

The manufacturing sector solved the quality inspection problem decades ago: instead of detecting defects after production, design processes where defects cannot occur. Proof over Prompt transfers this philosophy — specifically FMEA and poka-yoke — to K-12 curriculum design. The result: assessment pipelines where AI submission is structurally impossible, without requiring detection tools.

Detection Is a Dead End

Most schools fight AI-generated student work with detection tools — AI checkers, plagiarism scanners, statistical analysis. This is the equivalent of quality inspection at the end of a production line: expensive, unreliable, and always one step behind.

Manufacturing abandoned this model decades ago. Education hasn't caught up yet.

Don't inspect after failure.
Design to prevent failure.

From Manufacturing to Education

Principle 1: LMEA (Learning Mode and Effects Analysis) 📋

Educational adaptation of FMEA. Systematically identifies failure points where AI-generated submissions could bypass learning objectives. Every assessment mapped, every vulnerability addressed.

Principle 2: Poka-yoke Checkpoints ✋

Assessment structures require process artifacts — drafts, reasoning logs, self-corrections, live defense — that are impossible to produce retroactively. Error-proofing by design.

Principle 3: Clean Pipe Principle 🔧

A correctly designed assessment pipeline structurally excludes AI-generated work without requiring detection tools. "Kirli borudan temiz su akmaz."

Principle 4: Evidence by Design 📐

Three-layer evidence taxonomy: Process (how it was made), Product (what was made), Communication Strategy (how it's defended).

Principle 5: Red Line (Bloom Level 3+) 🔴

Below Bloom L3: AI-assisted tasks acceptable. Above Bloom L3: AI-supervised zone — human proof required at every checkpoint.

Deployed, Not Theoretical

PoP has been deployed across 43 production-locked curriculum files covering a major ELT publisher's secondary English series (CEFR A2-B1) at a K-8 school in Istanbul. Each file embeds:

4

Canonical Checkpoints

2

CEFR-variant Student Pathways

60/40

Process/Product Weighting

3

Stakeholder Support Layers

Iterative deployment across three textbook units demonstrates measurable quality convergence in assessment design consistency:

Quality Convergence (U1 → U2 → U3)

U1Critical failures → Gold Standard rebuild
U23 fix rounds → Structural contracts established
U3Near-clean first pass → Zero-defect approaching

Direction: → Each iteration gets cleaner. The system learns.

Research & Publication

PoP is not just a practitioner tool — it's an academic contribution to the intersection of quality engineering and educational assessment.

✅ EC-TEL 2026 abstract submitted (Springer LNCS, Valencia)

Track: Industry & Practitioner Reports

⬜ EC-TEL 2026 full paper — April 3, 2026

⬜ Article A: Quality Assurance in Education (Emerald) — 2026

⬜ Article B: Assessment in Education (T&F) — 2027

⬜ EC-TEL 2026 (September 14-18, 2026, Valencia, Spain)

The PoP Dictionary

N=N Principle

N students = N learning paths. No one-size-fits-all.

LPA

Latent Potential Actualization — every student has untapped capacity.

AI Swarm

5-agent architecture (Input, Scaffold, Practice, Monitor, Motivation).

Zero-Defect Pedagogy

Prevention by design, not detection after the fact.

Teacher-as-QA

The teacher is the quality engineer, not the content deliverer.

Experience PoP in 15 Minutes

← Back to Home

Part of the Proof over Prompt™ ecosystem

Also: EviScope AI | ProofBridge (coming soon)