PERP — Pedagogical Effectiveness Review Portal

What the field lacks	What PERP provides
No pre-deployment evidence threshold	Structured evaluation before any product receives certification
No learner-model validation	Psychometric review of adaptive claims — IRT, knowledge space theory, LCC
No algorithmic transparency requirement	Developer interviews probing architecture, training provenance, and guardrails
No mandatory equity auditing	Subgroup parity analysis and WCAG accessibility as primary evaluation dimensions
No post-deployment surveillance	Tiered certification (Developing → Adequate → Exemplary) tied to ongoing evidence

Product	Submitted	Status	Score
Loading...

Product	Developer	Submitted	Status	Action

1

Product Information

2

Upload Data

3

Developer Interview

Product Information

Provide identification and instructional design information. All fields marked * are required.

Product Name *

Version *

Developer / Organization *

Learning Domain *

Target Learner Age Range *

- years

AI Role Classification *

Type A — Augmentation Tool

AI assists learner activities; learner remains primary agent throughout

Type B — Intelligent Tutor

AI acts as adaptive instructional agent; provides dialogue-based guidance

Type C — Autonomous Agent

AI makes consequential instructional decisions with minimal human oversight

Stated Learning Objectives *

Theory of Change

Theoretical / Empirical Basis

Upload Materials

Upload product documentation and a representative sample of interaction data. All submissions are encrypted and governed by PERP's Data Governance Policy.

Product Documentation

Design documents, evidence reports, user guides
PDF · DOCX · Markdown accepted

Interaction Data

xAPI logs · chat transcripts · session exports
JSON · CSV · NDJSON · Min. 50 sessions required

Data Requirements

✓Minimum 50 unique learner sessions

○Random sample — not developer-selected

○PII removed or pseudonymized

○Includes both successful and unsuccessful sessions

Developer Interview

A structured AI-mediated interview covering 7 domains across approximately 15 questions. Responses are cross-referenced against submitted documentation.

Protocol

IP-2026.1

Duration

45-60 min

Domains

7 · ~15 questions

Format

AI-mediated chat

Design rationale

Evidence base

AI model & guardrails

Known failure modes

Assessment validity

Equity & access

Data governance

Integrity notice: Responses are cross-referenced against uploaded product documentation. Discrepancies are flagged in the evaluation report. All responses are used solely for evaluation and are not published without explicit consent.

Product	Status	Score	Date
Loading...

Socratic AI Tutor

v2.3.1· TutorLabs Inc.· Type B — Intelligent Tutoring System· Mathematics · Grades 6-8

3.0

Mean Score

1,247

Sessions Analyzed

15 / 15

Interview Complete

✓

Design Quality
Reviewed

Valid through April 2028

Score Profile

Product scores

Cert. threshold (3.0)

Dimension Scores

Dimension Analysis

Flagged Concerns

Learner Equity — Dimension 5

Subgroup analysis limited to one urban district (N=312). No data for rural, low-SES, or ELL learners. Algorithmic equity audit not conducted. Must be addressed before Evidence-Informed tier application.

Evidence Quality — Dimension 6

All cited evaluation studies conducted by or funded by TutorLabs Inc. No independent replication. Effect sizes (d = 0.62-0.71) likely inflated (cf. Cheung & Slavin, 2016). Independent evaluation required for tier upgrade.

Safety & Ethics — Dimension 7 ✓

No auto-fail conditions triggered. FERPA and GDPR compliance confirmed. Human override mechanism present. AI identity disclosed at session start in 100% of sampled sessions.

Recommendations

Expand equity testing to at least 3 demographically distinct populations (rural, low-SES, ELL) before next review cycle.
Commission independent evaluation — minimum: quasi-experimental study with matched control group — to support Evidence-Informed tier.
Raise assessment cognitive level — item bank is predominantly Bloom L1-2. Target L3-4 (application, analysis) for stated algebra objectives.
Document algorithmic bias audit — the adaptive difficulty algorithm has not been tested for differential impact across gender or SES subgroups.
Complete ECD documentation for 2 remaining knowledge components lacking traceable student-model rationale.

The rubric comprises seven dimensions, each independently scored on a 4-point scale (Exemplary · Adequate · Developing · Deficient). All criteria are grounded in peer-reviewed learning sciences literature. No single research group's proprietary framework is privileged. Criteria are revised when new meta-analytic evidence supersedes prior consensus.

Certified Products

Product	Domain	Tier	Score	Issued	Expires

Account

Name

—

Email

—

Role

—

Provider

—

Security

Password

Notifications

Email notifications

Receive updates when your submission status changes

Review reminders

Get reminded about pending interview sessions

Session

Sign out of this device

Rigorous, Evidence‑Based Evaluation of AI in Education

7-Dimension Framework

AI-Assisted Review

Transparent Evidence

Why sign in?

Education Has No FDA for AI

The Problem

The Analogy

The Answer

Welcome back

Create an account

Check your email

Reset your password

Set new password

Admin Overview

All Submissions

Portfolio Scores by Dimension

Certification Tiers

My Submissions

Review Queue

Submit for Review

Product Information

Upload Materials

Data Requirements

Developer Interview

Interview Domains

Flags Raised

My Reports

Evaluation Report

Socratic AI Tutor

Score Profile

Dimension Scores

Dimension Analysis

Flagged Concerns

Recommendations

Evaluation Rubric

Certification Registry

Certified Products

My Profile

Account

Security

Settings

Notifications

Session