How ADE Builds Curricula

Executive Summary

We do not just prompt an LLM to create a lesson. ADE’s curriculum generation pipeline is a multi-stage, multi-model architecture. First, an advanced LLM generates lesson content from 65 deterministic parameters; covering pedagogy, 80 brain researched teaching methodologies, and state standards alignment. That output is then passed to a higher-capability model operating as an LLM-as-a-Judge, which evaluates the content against our quality rubric and feeds structured feedback upstream to automatically tune the generation parameters. The refined output then enters a third stage; an even more capable model running Automated Prompt Optimization, which iteratively improves the upstream prompts themselves to maximize output quality at scale. After all three AI stages, every lesson is reviewed word-by-word by credentialed education professionals for coherence, academic rigor, logical sequencing, and adherence to state-specific standards.

Every lesson a student encounters on the American Digital Education platform arrives with four guarantees that administrators can take to a board, an accreditation team, or an audit.

This document explains how those guarantees are built, why they are built the way they are, and what they mean for your students, your teachers, and your district’s academic outcomes.

1. The Problem These Design Choices Solve

Before explaining what ADE does, it is worth naming what it is responding to.

Traditional curriculum materials, whether print or digital, frequently share the same structural weaknesses: passive delivery that does not require students to produce evidence of understanding, assessment that is concentrated at the end of instruction rather than embedded throughout, question sets that test the same narrow cognitive skill repeatedly with only superficial variation, and visual designs that were built for aesthetics rather than cognition.

The result is the illusion of understanding: students appear to follow instruction, complete worksheets, and move on, and then fail to demonstrate mastery on assessments because they never generated the kind of active, varied, retrieval-based engagement that research shows is required for genuine learning.

2. The Research Foundation: Why Our Lessons Look the Way They Do

2.1 Explicit Direct Instruction (Hollingsworth & Ybarra, 2009)

The foundational framework for every ADE lesson is Explicit Direct Instruction (EDI), a methodology developed and documented by Hollingsworth and Ybarra (2009) through implementation in more than 25,000 classrooms across 40+ countries. EDI organizes instruction into seven sequential, non-skippable components that move from teacher modeling to student independence in a carefully scaffolded progression.

EDI is not a worksheet protocol or a pacing guide. It is a theory of how novice learners acquire new conceptual and procedural knowledge: through clear objective-setting, connection to prior knowledge, concept definition with examples and non-examples, modeled skill application, guided practice with immediate feedback, relevance to students’ lives, and closure that consolidates learning. Peer-reviewed studies consistently show significant achievement gains from EDI implementation, with particularly strong effects for English Language Learners and students in Title I schools.

Every ADE lesson follows this seven-component sequence, mapped to 65 ordered content elements. No component can be skipped, reordered, or shortened. The structure is enforced at the platform architecture level, not left to individual lesson authors.

2.2 Cognitive Load Theory (Sweller, van Merriënboer & Paas, 1998)

Human working memory can hold approximately four chunks of novel information at any given moment (Cowan, 2001). Every instructional design choice that adds processing burden, a confusing layout, an irrelevant image, a redundant explanation, steals cognitive capacity that the student needs for actual learning.

ADE’s lesson architecture is built to minimize extraneous cognitive load (load generated by poor design) and maximize germane cognitive load (the productive mental effort of schema construction). Specific applications include:

• Chunking: New information is introduced in manageable sequential steps; students do not encounter a concept before its prerequisites have been established
• Scaffolding: Sentence frames reduce the cognitive burden of formulating verbal responses, so students’ cognitive effort goes to content rather than to language formulation
• Visual design: Content width is constrained to 50-75 characters per line (Dyson & Kipping, 1998), font size is held at 16-18px (W3C WCAG AA), and line height is maintained at 1.5 (Chaparro et al., 2004), all research-backed optima for sustained reading by developing readers
• Image placement: Images appear only at the nine lesson steps where they directly support problem comprehension, never during concept definitions or comprehension checks, where their presence would generate split attention (Chandler & Sweller, 1991)

2.3 Cognitive Theory of Multimedia Learning (Mayer, 2009)

Richard Mayer’s framework, tested across more than 100 experimental comparisons, establishes that students learn best when words and relevant pictures are presented together through both auditory and visual channels. ADE’s primary instructional mechanism is an AI instructor who narrates lesson content while each word on screen highlights in real time as it is spoken.

This implements two of Mayer’s strongest findings:

The color system is also research-grounded: yellow highlighting during narration implements Mayer’s Signaling Principle (d = 0.41 across 29 studies); color-coding of lesson components (blue for Learning Objective, green for Examples, red for Non-Examples) implements the schema activation principle from Anderson (1984), allowing students to recognize the type of content they are processing before reading a single word.

2.4 Retrieval Practice and the Testing Effect (Roediger & Butler, 2011)

Research consistently shows that retrieving information from memory, not re-reading it, is the most effective strategy for long-term retention. ADE embeds 27-29 Checks for Understanding (CFUs) across every lesson, interleaved throughout instruction rather than concentrated at the end. Each CFU requires the student to produce evidence of understanding (verbally or by selecting an answer) at the exact moment the concept is freshest and active in working memory.

This is retrieval practice by design. Students cannot advance past a CFU without responding. Incorrect responses trigger targeted re-teaching of the specific misunderstood concept before the student continues. The lesson is not a passive experience; it requires continuous active participation from the first minute to the last.

2.5 Formative Assessment (Black & Wiliam, 1998)

Black and Wiliam’s landmark meta-analysis of 250 studies established that formative assessment, assessment used to adjust instruction in real time rather than to assign grades after the fact, has among the largest effect sizes of any educational intervention, with gains equivalent to moving a student from the 50th to the 85th percentile.

ADE’s CFU system is formative assessment operationalized. Every question gives the system (and the teacher’s analytics dashboard) real-time information about each student’s understanding. When a student answers incorrectly, the lesson provides adaptive re-teaching, not simply “that’s wrong, try again,” but a targeted explanation of the specific concept the student misunderstood. Teachers see aggregated CFU performance data across their class immediately, enabling same-day instructional intervention rather than discovering gaps weeks later.

2.6 Bloom’s Taxonomy (Anderson & Krathwohl, 2001)

Traditional curricula heavily weight Recall and Comprehension questions, the two lowest levels of Bloom’s revised taxonomy, which creates students who can define terms but cannot apply, analyze, evaluate, or create with them. ADE’s 27-29 CFUs per lesson span all six cognitive levels:

The distribution is designed to mirror standardized assessment formats, preparing students for high-stakes tests through authentic cognitive practice rather than test-prep drills.

2.7 Additional Research Anchors

• Variable Reinforcement (Skinner, 1957): Seven rotating confetti patterns on first-attempt-correct responses maintain engagement without reward habituation; celebrations are limited to 2 seconds maximum to preserve flow state (Csikszentmihalyi, 1990)
• Feedback Timing (Kulik & Kulik): Correct-answer feedback arrives within seconds of response, while reasoning is still active in working memory, the window where feedback has the greatest impact
• Dual Coding Theory (Paivio, 1971): Images at worked-example steps provide a second encoding pathway (visual) alongside the verbal narration, creating two retrieval routes in long-term memory
• Self-Regulated Learning: Students control playback speed (0.5x-2.0x) and can pause; the platform saves position down to the word being spoken so students resume precisely where they left off across devices and sessions

3. The 65-Step Lesson Architecture

Every ADE lesson moves through exactly 65 ordered content steps mapped to seven pedagogical components. This structure is not a template; it is the architecture itself, enforced at the platform level. A lesson cannot be published with fewer than 65 steps or with any component missing.

4. The 6 Pedagogical Quality Rules Every Lesson Must Pass

Before any lesson is approved for teacher assignment, it must satisfy six non-negotiable pedagogical rules. These rules are the product of extensive curriculum development experience and are enforced through both automated checks and human editorial review.

5. The Ed.D. Review Process: How Lessons Reach the Classroom

ADE’s quality assurance process treats human editorial review as the most critical gate in the entire production pipeline.

6. Curriculum Built for Every State: The Standards Alignment System

Scale and Coverage

ADE’s standards infrastructure is the most comprehensive K-12 standards system ever built into a lesson-delivery platform:

Crosswalk reference coverage: 99.96% (44,650 of 44,670 standards have at least one verified mapping to a corresponding national framework standard).

How Standards Are Organized

Every standard in the database carries:

• State or framework code (e.g., CCSS, TX, FL, NGSS, C3)
• Grade level and grade band (Elementary K-5, Middle 6-8, High School 9-12)
• Depth of Knowledge (DOK) level 1-4 (Webb’s framework)
• Rigor calibration: below grade, on grade, above grade, or honors
• Hierarchy level 1-5 (domain, cluster, standard, sub-standard, component)
• Assessment item types that the standard calls for

Custom Framework States

Not all states use national frameworks. ADE maintains dedicated standards records for states that have built their own frameworks:

Every state’s lessons are built directly from that state’s own standards. This is true for all 50 states and DC, whether the state adopted a national framework or developed an independent one. Lessons are never generated from a national framework and adapted. Crosswalk references exist solely as a traceability layer, allowing administrators to see which national framework standard a given state-specific lesson maps back to for reporting and compliance purposes.

The 10-Step Lesson Generation Workflow

When a lesson is created for a teacher in any state, it passes through a rigorous 10-step process:

1. Parse the Request – Extract grade level, subject, state code, and topic
2. Query State Standards – Retrieve the specific state-level standards for the topic (e.g., TEKS for Texas, BEST for Florida, CCSS for California)
3. Resolve National Framework Reference – Identify the corresponding national framework standard for traceability and reporting purposes
4. Build the AI Prompt – Construct a structured prompt embedding the state-specific standard, DOK level, and all 65 EDI steps
5. Generate Draft Content – Gemini 2.5 Pro returns a complete 65-element draft built against the state standard
6. Automated Validation – Check all quality gates (see Section 5, Stage 2)
7. Tag Lesson with State Standard – Mark the primary state-level standard the lesson was authored against
8. Attach National Framework Reference – Record the corresponding national framework standard (e.g., CCSS, NGSS, C3) as a reference for reporting and compliance traceability
9. Create Assessment Profile – Set DOK-matched question types and rigor level based on the state standard
10. State-Specific Display – Configure the lesson to display the state’s own standards codes to teachers and administrators

How State-Specific Lessons Work in Practice

Lesson topic: Adding fractions with unlike denominators, Grade 5

Texas teacher receives a lesson built on: TEKS 5.3A
California teacher receives a lesson built on: CCSS.MATH.5.NF.A.1
Florida teacher receives a lesson built on: BEST MA.5.FR.2.1
New York teacher receives a lesson built on: NY-5.NF.1

Each lesson is authored directly from the state’s own standard. Crosswalk references let administrators trace any lesson back to its national framework equivalent for reporting and compliance.

Audit-Defensible Alignment

Every alignment claim is documented with:

• Alignment type: direct, partial, or supporting
• Confidence level: 15-100 (automating only references ≥ 85)
• Verification status: system-generated or human-verified
• Date and method of verification

This creates a complete evidentiary chain from lesson content to state-level standard to national framework reference, exactly what a Title I audit, an ESSA compliance review, or an accreditation visit requires.

Depth of Knowledge Across Subjects

Standards are distributed across DOK levels reflecting the nature of each subject’s learning demands. ADE lessons are calibrated to match these distributions exactly:

7. Deterministic AI: Why ADE Does Not Use Generative AI With Students

This is a distinction that every administrator should understand before presenting ADE to a school board, a parent organization, or a state compliance auditor.

The Critical Difference

How Determinism Is Engineered

Every element of a student’s lesson experience is pre-built, reviewed, and fixed before delivery:

• All lesson narration is text-to-speech audio generated in advance, cached as fixed audio files, and served identically to every student
• All images are pre-generated, vetted by reviewers, and stored; nothing is generated during a student session
• Multiple-choice questions have predefined answer keys, predefined correct-answer feedback, and predefined reteaching feedback for each incorrect option
• Verbal-response questions are scored against fixed rubrics with four confidence bands (95-100 correct, 85-94 strong, 70-84 developing, below 70 needs reteaching) using pattern-matching against pre-defined correct answers, not real-time AI generation
• Sentence frames guide students toward structured responses, ensuring consistency in both what students produce and how the system evaluates it

What AI Does in ADE, and Where It Stops

ADE uses AI in two contexts:

1. Lesson authoring: Gemini assists curriculum authors in drafting lesson content against the 65-step EDI template. This is the same relationship as using a word processor to write a textbook. The tool assists the author; the human editorial team produces the final work.

2. The “Ask a Question” and “I Don’t Understand” features (described in Sections 9 and 10): These are live AI features, but they are bounded by strict guardrails, operate with full lesson context, are never used to score formal assessments, and maintain a complete audit log of every student-AI interaction available for teacher review.

In both cases, students never interact with unconstrained, open-ended generative AI. The system is designed, not hoped, to be appropriate for every student every time.

8. Interactive Student Questions: Supporting Curiosity While Protecting Integrity

ADE’s “Ask a Question” feature allows students to pause their lesson at any point, voice a question, and receive an immediate, age-appropriate, educational response. It is designed to replicate what good teachers do when a student raises a hand: redirect curiosity toward understanding without giving away answers.

What the System Does

1. Student activates the feature via the sidebar
2. Lesson pauses; student speaks their question into their microphone
3. The AI analyzes the question against the full lesson context (learning objective, definitions, examples, current step)
4. An educational response is provided in 2-3 sentences, focused on teaching the underlying concept
5. Student continues the lesson

What the System Refuses to Do

• Off-topic questions receive a redirect: “I’m sorry, that’s not something we are covering at this time. Either rephrase your question or hit the continue button so we can keep going.”
• Direct answer-seeking on assessment questions triggers concept teaching, not answer delivery: “It looks like that’s a specific question from this lesson. Let me help you understand how to identify ratios…”
• Questions outside lesson scope are politely declined and redirected to the lesson content

What Teachers and Administrators Can See

A complete audit trail of every student question and AI response is available in the teacher and administrator dashboards. Teachers can identify which concepts generate the most questions across their class, use that data to inform reteaching decisions, and present it as evidence of student engagement in SST or parent communication.

9. The “I Don’t Understand” Feature: Adaptive Support at the Moment of Confusion

Separate from the “Ask a Question” feature, the “I Don’t Understand” button allows students to request re-teaching of the specific step they are on, without needing to formulate a question. It is designed for the student who knows they are confused but does not yet know what to ask.

How It Works

1. Student clicks the “I Don’t Understand” button during any lesson step
2. The lesson pauses at the exact point of confusion
3. The system generates a contextual explanation using the lesson’s current step, the surrounding 2-3 steps, the learning objective, the student’s grade level, and the subject’s standards alignment
4. The explanation is delivered as audio with word-by-word highlighting, the same modality as the lesson itself
5. The student chooses: Tell Me More (deeper explanation), Ask a Question (voice a specific question), or Continue Lesson (return to exactly where they left off)

The “Tell Me More” option supports up to three or four progressive deepening levels before the system gently suggests that formulating a specific question might be more helpful.

What Teachers Gain

Every “I Don’t Understand” activation is logged and visible in the teacher dashboard. Aggregated data shows:

• Which lesson steps generate the most help requests across the class
• How deeply individual students are exploring explanations
• Correlation between help-seeking behavior and final mastery scores
• Which content may benefit from revision or additional classroom instruction

10. The Science of Image Placement: Why Images Appear at Exactly 9 Steps

Images in ADE lessons appear at exactly nine specific lesson steps per lesson, no more, no less. This is not an artistic choice; it is a direct application of Cognitive Load Theory and the Cognitive Theory of Multimedia Learning.

Where Images Appear (and Why)

Where Images Are Intentionally Absent (and Why)

• Learning Objective: Verbal encoding must have full access to working memory; no visual processing competition
• Concept Definitions: A definition is a linguistic task; images during this phase generate split attention even when topically related
• Checking for Understanding: Assessment requires effortful retrieval from memory; images would either provide unwanted retrieval cues or add extraneous load
• Closure and Relevance: Metacognitive consolidation is best supported by visual simplicity, not stimulation

11. The Four-Tier Assessment System

Curriculum without assessment is a hypothesis. ADE’s four-tier assessment architecture converts instructional design into measurable evidence of student learning, from individual concept checks within a single lesson to pre-to-post growth across an entire school year.

Every assessment in the system is generated by ADE’s super-admin team and reviewed by Ed.D.-credentialed educators before publication, the same authoring pipeline that governs lessons. Teachers, school admins, and district admins assign published assessments to students; they cannot create or modify assessment content. This design ensures that every student taking a given quiz sees the same validated, standards-aligned questions.

Tier 1: Embedded Lesson CFUs (27-29 Per Lesson)

Tier 2: Lesson Quiz (1 Per Lesson, 10 Questions)

After completing a lesson, a teacher can assign the corresponding lesson quiz, a standalone assessment that tests whether the student retained and can transfer what the lesson taught, outside the lesson’s context.

Tier 3: Unit Test (1 Per Unit, 20 Questions)

A unit test synthesizes learning across a group of related lessons. ADE’s courses are organized into units (groups of lessons sharing a common domain or standard cluster). Each unit has exactly one associated unit test.

Tier 4: Course Benchmark Assessments (Pre-Test, Mid-Test, Post-Test)

Three benchmark assessments are generated for every course: one administered at the beginning of the year (before any lessons), one at midyear, and one at the end of the year. These are the mechanism that converts ADE from a curriculum platform into an evidence generator.

The parallel-form design is the critical feature: because all three instruments cover identical standards at identical rigor, a score increase from pre to post cannot be explained by “the post-test was easier”; it can only reflect genuine learning gains.

Assessment Authoring and Quality Assurance

The same quality pipeline that governs lessons governs assessments:

1. AI-generated draft: Questions generated against lesson/unit/course standards, existing CFU questions supplied as negative examples to prevent duplication, grade-appropriate vocabulary, BIAS audit (no cultural/socioeconomic/gender bias), each question tagged to a specific standard and Bloom’s level
2. Automated validation: Format, non-duplication, standards tagging, and DOK calibration checks
3. Ed.D. editorial review: Reviewers assess standards alignment accuracy, rigor appropriateness, grade-level calibration, and absence of construct-irrelevant variance
4. Lock and publish: Finalized assessments published to the platform; only super-admin role can modify published assessment content

Scoring Mechanics

All assessments are fully auto-graded:

• Multiple-choice: Binary correct/incorrect at submission
• Verbal responses: Evaluated by AI meaning-match against the pre-defined correct answer (≥70% semantic similarity threshold = correct); same deterministic evaluation infrastructure used in lessons
• Mastery score: Calculated as first-attempt-correct ÷ total questions × 100; tracks performance quality, not just eventual correctness across retries
• Standards mastery: Per-standard accuracy aggregated across all assessment sources; a student’s mastery level for any given standard is never downgraded, only the highest achieved level is retained

Role-Based Access

All access is enforced by row-level security; a teacher cannot view another teacher’s class results, and a student cannot view another student’s scores.

12. Measuring Growth and Producing Efficacy Evidence

The benchmark system is designed explicitly to produce the kind of evidence that satisfies ESSA requirements and enables district leaders to answer board questions with data, not anecdotes.

Pre-to-Post Growth Reports

When a student completes the end-of-year benchmark, the platform automatically calculates:

• Individual growth: “This student scored 42% on the pre-test and 78% on the post-test, a 36-point gain across 14 of 18 assessed standards.”
• Class growth: “This class showed an average gain of 31 percentage points; strongest growth in Standards X, Y, Z; weakest in Standards A, B, C.”
• District growth: “Across all students using ADE this year, the average pre-to-post effect size was d = 0.65, with statistically significant gains in 94% of assessed standards.”

ESSA Evidence Pathway

The pre/post benchmark design positions ADE to support ESSA evidence requirements across two tiers:

• ESSA Tier 4 (achievable immediately): Requires a logic model based on empirical research and ongoing efforts to examine impact. ADE’s research foundation (detailed in Sections 2 and 3) satisfies the logic model; the benchmark system satisfies the “ongoing examination” requirement.
• ESSA Tier 3 (achievable within one to two years of deployment): Requires at least one well-designed correlational study showing statistically significant positive effects. The benchmark data this system automatically generates, across a sufficient sample of students, is that study data. Districts do not need to commission external research; they need to deploy the pre-test at the start of the year and the post-test at the end.

Teacher-Level Analytics

Teachers have access to analytics at every assessment tier without waiting for district reports:

• Lesson CFU data: Which concepts generated the most errors, which students needed re-teaching, per-step accuracy rates
• Lesson quiz performance: Per-student scores, class averages, most-missed questions, per-standard breakdown
• Unit test results: Cross-lesson synthesis data; class comparison to school average (when multiple teachers teach the same course)
• Benchmark growth: Per-student pre-to-mid-to-post trend lines; class-level aggregate growth
• Standards mastery grid: A unified view combining CFU, quiz, unit test, and benchmark data into a single student-by-standard mastery map, the most complete picture of student learning available in any platform

All teacher analytics are exportable as CSV for Professional Learning Community meeting protocols. A teacher can walk into a Wednesday PLC with a single-page report showing class performance on every standard addressed in the past unit.

Board-Ready Efficacy Documentation

The district efficacy report generator produces a formatted document containing:

• Mean pre and post scores with confidence intervals
• Mean growth in percentage points
• Cohen’s d effect size per course and district-wide
• Percentage of students showing positive growth
• Per-standard detail: which standards showed the strongest gains, which need additional attention
• Sample sizes for statistical validity

This document is exportable as PDF and designed for board presentations, grant applications, and state reporting.

13. What This Means for Your District: Talking Points for Leaders

For Board Presentations

ADE delivers consistent, research-grounded, standards-verified instruction to every student, regardless of class size, substitute teacher availability, or time of day. The instructional quality a student receives on a Friday afternoon is identical to what they receive on a Monday morning with a veteran teacher. This consistency is especially significant for Title I schools and underserved communities, where achievement gaps often reflect opportunity gaps in instructional quality.

For Parent Communities

Students who use ADE are engaged in active, structured learning, not passive video watching or worksheet completion. Every lesson requires students to speak, reason, and demonstrate understanding at multiple points. Parents can be told accurately: “Our platform was designed by educators with doctoral-level expertise and is reviewed by Ed.D. credentialed staff before any student sees it. It never exposes students to open-ended generative AI.”

For Accreditation and Compliance

ADE provides complete, traceable documentation of standards alignment for every lesson every student completes. The audit trail includes the specific standard addressed, the alignment type and confidence level, the DOK level, and the student’s performance on each of the 27-29 CFU questions, all retrievable through the administrator dashboard.

For Curriculum Coordinators

ADE does not replace curriculum coordinators; it gives them a research-grounded, auditable baseline. Teachers can preview every lesson before assigning it. Coordinators can verify standards alignment through the Standards Browser. Teachers can flag alignment concerns directly from the lesson interface, and those flags enter a review queue monitored by ADE’s standards team.

14. Frequently Asked Questions

Appendix A: Research Bibliography

The following researchers and publications are cited throughout this guide. All ADE design decisions are traceable to peer-reviewed sources in this list.

Appendix B: Standards Coverage Snapshot

All 50 states + DC are represented. Every state receives lessons built directly from its own standards. Mirror states (those that adopted national frameworks with minor branding changes) receive direct reference mappings at confidence 90-100. Custom framework states (TX, FL, CA, NY, VA, and others) receive dedicated standards records and lessons authored against their own frameworks, with crosswalk references back to national standards for traceability.

Appendix C: Assessment System at a Glance