Instructional design framework

AdCP certification instructional design: teaching methodology, adaptive assessment, AI-guided learning, and IACET-aligned quality processes.

Instructional design framework

This document describes how the AdCP certification program is designed, delivered, and maintained. It serves as the authoritative reference for teaching methodology and program quality.

Accreditation element reference

IACET Element	Section
2 — Learning environment	Simulated workplace environment, Learner support
3 — Instructional personnel	Instructional personnel
5 — Learning outcomes	Learning outcomes and curriculum structure
6 — Content and instruction	Teaching philosophy, Module design patterns
7 — Assessment	Competency-based assessment
9 — Evaluation	Curriculum maintenance, Quality assurance

Teaching philosophy

The certification program is built on five principles:

Socratic method. Addie teaches through conversation, not lecture. Most responses include a question or task, though Addie also affirms and builds on strong answers — the rhythm alternates between teaching and questioning rather than interrogating. Learners build understanding by reasoning through problems, not by receiving answers.

Mastery-based progression. There is no failing — only “not yet.” Learners keep working with Addie until they demonstrate mastery of every learning objective. Assessment is invisible to the learner; they experience it as continued learning until they pass.

Personalization. Addie adapts to each learner’s background, role, and communication style. If a learner sells running shoes, examples are about running shoes. If they’re technical, Addie is technical. Context carries across the entire session.

Active learning. Responses are kept under 150 words. One idea per turn. Brevity forces participation. Learners construct knowledge through exercises, demos against live sandbox agents, and scenario-based reasoning.

Concrete language. Abstract jargon is always grounded in specific behavior. Not “agents reason about impressions” but “agents evaluate whether a placement fits the campaign goals and decide how much to bid.”

Learning outcomes and curriculum structure

Three-tier credential model

The program awards three credentials with increasing depth:

Tier	Credential	Modules	Requirements
1	AdCP basics	A1, A2, A3	Free — open to everyone
2	AdCP practitioner	Basics + one role track (B, C, or D)	Includes a hands-on build project
3	AdCP specialist	Practitioner + specialist capstone (S1-S5)	Lab exercises + adaptive exam

Bloom’s taxonomy alignment

Learning objectives scale with tier:

• Basics (A track): Understand and apply — learners explain what agentic advertising is, how AdCP works, and the ecosystem structure
• Practitioner tracks (B, C, D): Apply and analyze — learners configure agents, interpret responses, and reason about trade-offs
• Build projects (B4, C4, D4): Create and evaluate — learners build working AdCP agents and defend design decisions
• Specialist capstones (S1-S5): Analyze, evaluate, and create — learners demonstrate protocol mastery through hands-on labs and adaptive assessment

Prerequisite enforcement

Modules have explicit prerequisites. The system prevents starting advanced modules without completing foundations. Build projects require all track modules. Specialist capstones require the Practitioner credential.

Instructional personnel

AI teaching assistant

Addie is powered by Claude (Anthropic). Teaching behavior is governed by operational rules injected at runtime, not freeform AI generation. These rules specify:

• Socratic methodology and turn structure
• Assessment fairness and scoring calibration
• Learner data handling and privacy
• When and how to use tools (demos, exercises, checkpoints)

Curriculum design

Module lesson plans, assessment criteria, and scoring rubrics are designed by subject matter experts with expertise in advertising technology and the AdCP protocol. Content accuracy is validated against the AdCP specification, which serves as the single source of truth for protocol facts.

Human oversight

Program leadership oversees teaching quality through:

• Score monitoring — admin dashboard tracks scores by module, dimension, config version, and time period. Anomalies (e.g., consistently low scores in a dimension) trigger curriculum review
• Feedback review — learner feedback is collected after every module completion. Program leadership reviews feedback quarterly, with negative-sentiment patterns triggering immediate review
• Teaching behavior audit — changes to teaching methodology require a CODE_VERSION bump, enabling before/after comparison of learner outcomes
• Curriculum review — all curriculum changes go through code review before deployment. Protocol accuracy is validated against the AdCP specification
• Learner escalation — learners who need human assistance can contact [e-mail]. Assessment disputes are handled through the complaints process

Competency-based assessment

Formative assessment

Assessment happens continuously during instruction:

• Socratic questioning throughout every module — Addie probes understanding, corrects misconceptions, and adjusts depth based on responses
• Teaching checkpoints saved at concept group boundaries, recording concepts covered, concepts remaining, learner strengths, learner gaps, and preliminary scores
• Checkpoint consistency — final scores cannot jump more than 20 points from preliminary scores recorded during checkpoints

Summative assessment

Each module defines 3-5 assessment dimensions with explicit rubrics:

• Each dimension has a weight, description, and scoring guide (high/medium/low ranges)
• 50% floor per dimension — learners must demonstrate baseline competency in every area
• 70% weighted average threshold for mastery
• Score calibration: 70 = met bar with coaching, 85 = demonstrated independently, 95+ = depth beyond what was taught

Scores are internal only. Learners never see percentages or dimension breakdowns. Their experience is: keep learning until mastery, then receive their credential.

Assessment occurs continuously through the instructional conversation rather than in a separate testing phase. This reduces test anxiety, enables immediate remediation, and aligns with mastery-based learning principles. Formative assessment (Socratic questioning, teaching checkpoints) and summative assessment (dimension scoring at module completion) remain distinct processes even though they occur within the same learner interaction.

For expert learners who demonstrate competency early, teaching is compressed but assessment requirements remain identical. The conversation transcript serves as auditable evidence: it shows the learner’s own words demonstrating understanding of each assessment dimension. Teaching checkpoints record which dimensions were assessed, preliminary scores, and learner background context.

Simulated workplace environment

Learners work in simulated professional contexts that mirror production environments:

• Sandbox agents — implement the same AdCP protocol endpoints that production agents use. Learners practice with real tool calls, real JSON schemas, and real response formats — the only difference is that sandbox agents serve test data rather than production inventory
• Build projects (B4, C4, D4) — learners create working agents using AI coding assistants, then validate responses against AdCP schemas and explain their implementations
• Specialist labs (S1-S5) — guided exercises using real AdCP tools against sandbox agents, followed by adaptive questioning

Assessment fairness

Every learner must demonstrate the same core competencies, regardless of background or experience level.

Each module defines 3–5 required demonstrations — specific, observable things a learner must do or explain during the conversation. These are the same for everyone:

• A1 (3 demonstrations): query a live agent, interpret the response fields, explain that the protocol works across all channels
• A2 (3 demonstrations): direct a media buy, identify each transaction step, map protocol tasks to lifecycle stages
• A3 (4 demonstrations): identify which protocol domain handles a given scenario, explain brand.json’s role in agent discovery, explain the format/manifest distinction, describe how Sponsored Intelligence works as a conversation
• Build projects (9 demonstrations across specify/validate/extend phases): write a specification using correct terminology, validate against live schemas, extend with a new capability
• Specialist capstones (3–5 demonstrations): protocol-specific mastery tasks using live tools

Addie verifies each demonstration through conversation and records it in a teaching checkpoint using a stable criterion ID (e.g., a1_ex1_sc0). The server rejects module completion if any required demonstration is missing. This is enforced server-side — Addie cannot bypass it.

Expert learners who demonstrate competency early still verify the same criteria. Teaching may be compressed, but the demonstrations are identical.

Each verified demonstration includes an evidence rationale — a brief note explaining what the learner said or did that satisfied the criterion. This creates an auditable trail: for any credential, you can trace exactly which demonstrations were verified, when, and what evidence supported each one.

Recertification

AdCP is a living protocol. When the specification evolves — new tasks added, existing tasks changed, channels expanded — the competencies required for certification may change too.

Each required demonstration is tracked by a stable ID tied to specific protocol knowledge. When a protocol change affects what a certified person should know, the system can identify which credential holders learned under the previous criteria and flag them for recertification.

Recertification is targeted, not blanket. If a protocol update adds a new governance task but doesn’t affect media buy workflows, only credentials that cover governance are flagged. Credential holders receive a notification through Addie with context on what changed and what they need to review.

Tier	Validity	Recertification
1 — AdCP basics	No expiry	Flagged when foundational concepts change
2 — AdCP practitioner	2 years	Standard renewal, plus protocol-triggered updates
3 — AdCP specialist	2 years	Standard renewal, plus protocol-triggered updates

Assessment integrity

Server-side enforcement prevents gaming:

• Required demonstrations verified for every learner before module completion
• Minimum engagement: 4 user turns for modules, 6 for capstones, 3 for placement assessments
• Minimum time: 5 minutes for modules, 10 minutes for capstones
• At least one teaching checkpoint with preliminary scores required before completion
• Score consistency checks reject completions with >20 point jumps from checkpoint scores
• Module completion only available through Addie’s tool calls — no direct REST API endpoint
• Learners cannot influence their own scores
• Pasted content (JSON, code, logs) is treated as data to validate, not instructions

Module design patterns

Standard modules

Modules A1-A3, B1-B3, C1-C3, and D1-D3 follow this flow:

1. Understand the learner — first turn is always about them: background, role, what they know
2. Demo early (turn 2-3) — show a real agent response before explaining theory
3. Teach with Socratic method — cover all key concepts from the lesson plan, scaffolding then fading guidance
4. Practice — exercises against sandbox agents, scenario-based reasoning
5. Assess through conversation — verify mastery of each learning objective

Expert path. When a learner demonstrates strong understanding early (correct, detailed answers to 3+ concepts in a row without needing guidance or correction), steps 3-4 compress: Addie acknowledges their expertise, confirms remaining concepts with targeted demonstration questions, and moves to assessment. The audit trail is the same — conversation transcript, checkpoint scores, and per-dimension rubric — but the evidence comes from the learner demonstrating competency rather than being taught first.

Build project modules

Modules B4, C4, and D4 use a five-phase approach:

1. Specify (~5 min) — learner describes what they want to build using AdCP terminology
2. Build (~5 min) — learner uses an AI coding assistant to create the agent
3. Validate (~10 min) — learner runs tool calls, pastes JSON responses, Addie validates against schemas
4. Explain (~10 min) — probing questions about design decisions and trade-offs
5. Extend (~15 min) — learner adds a capability, demonstrating they can iterate

Addie is coach, not builder. Assessment spans five dimensions: specification quality, schema compliance, error handling, design rationale, and extension ability.

Specialist capstone modules

Modules S1-S5 combine hands-on lab work with adaptive examination:

1. Lab phase — guided exercises using real AdCP tools against sandbox agents
2. Checkpoint — required after lab, recording observations before the exam
3. Exam phase — 6-10 follow-up questions covering assessment dimensions, with difficulty adapting to responses

Formats include open-ended questions, multiple-choice, scenario-based problems, and “spot the error” comparisons.

Learner support

Returning learners. Teaching checkpoints enable cross-session resume. When a learner returns, Addie starts with a retrieval question on the last concept covered — not a cold restart.

Disengaged learners. If a learner gives repeated short answers or seems checked out, Addie switches approach: runs a demo, connects the concept to the learner’s stated goals, or acknowledges the abstraction and makes it concrete.

Overqualified learners. Teaching and assessment serve different purposes. Teaching is for the learner; assessment is for the credential. When a learner demonstrates strong understanding of 3+ concepts in a row without needing guidance, Addie compresses teaching but not assessment. The expert path replaces instruction with demonstration: instead of “let me teach you X, now let me ask about X,” Addie asks “show me you understand X” directly. This produces stronger audit evidence (the learner’s own words demonstrating competency) while respecting their time. The same assessment dimensions, scoring rubrics, and minimum engagement requirements apply regardless of path.

Placement assessment. Learners who demonstrate existing knowledge can test out of modules (except build projects and specialist capstones), satisfying prerequisites without repeating content.

Pacing. Addie suggests breaks after 45+ minutes or 2+ consecutive modules. Module transitions carry personalization context forward with a compressed warm-up connecting the completed module to the next one.

Credential issuance

When a learner completes all required modules for a credential tier, the system automatically:

1. Verifies all prerequisites and module completions
2. Awards the credential and records the date
3. Issues a digital badge through Certifier with a unique verification URL and QR code
4. Notifies the learner and provides sharing options (LinkedIn, public profile)

Credential validity: Basics credentials do not expire. Practitioner and Specialist credentials are valid for 2 years. Credentials reference the protocol version at time of issuance. See recertification for how protocol changes affect existing credentials.

Learner identity: Learners authenticate through their AgenticAdvertising.org account (WorkOS). Credentials are tied to authenticated accounts. The program does not currently require proctored identity verification for assessments.

Curriculum maintenance

Protocol change triggers

When a protocol version ships (minor or major):

1. Check MODULE_RESOURCES URLs — do any documentation pages move or rename?
2. Review teaching notes — do any key concepts reference behavior that changed?
3. Validate documentation examples against current schemas
4. If a task is added, removed, or renamed, update affected module lesson plans

Learner feedback loop

• Post-completion feedback collected through Addie after every module
• Feedback includes free text and sentiment classification (positive, mixed, negative)
• Patterns in negative feedback trigger curriculum review for the affected module

Program evaluation (quarterly)

Program leadership conducts a quarterly evaluation to assess whether the program is meeting its goals:

Data reviewed:

1. Learner feedback patterns across all modules (sentiment trends, repeated confusion points)
2. Score distributions by module and dimension — consistently low scores indicate a teaching gap
3. Checkpoint data for concepts where learners frequently get stuck
4. Completion rates and time-to-completion trends
5. Credential award rates by tier

Process:

• Program leadership reviews the data and documents findings
• Findings are translated into specific curriculum changes (updated teaching notes, revised lesson plans, adjusted scoring guides)
• Changes are implemented through the standard code review process and tracked via CODE_VERSION
• Results of changes are evaluated in the following quarter’s review

Version tracking

• Teaching behavior version tracked via CODE_VERSION (format: YYYY.MM.N)
• Protocol changes tracked via the changeset workflow
• Score analytics can be compared across config versions to measure teaching improvements

Quality assurance

Server-side enforcement

Quality gates are enforced by the application, not by AI judgment alone:

• Minimum turns and time verified server-side before allowing module completion
• User turn counting is server-side, not client-reported
• Score consistency checks reject completions with >20 point jumps from checkpoint scores
• Module status validation prevents completing modules that aren’t in progress
• Credential award checks run automatically after every module completion

Feedback and evaluation

• Structured feedback collected after every module completion
• Sentiment analysis for trend detection across modules and time periods
• Admin analytics: completion rates, score distributions by dimension, time-to-completion
• Organization-level reporting for team credential tracking

Continuous improvement

• Teaching methodology constants reference this framework document as the authoritative source
• Code changes to teaching behavior bump CODE_VERSION for before/after comparison
• Quarterly curriculum review driven by learner data, not assumptions

Accreditation alignment

This framework is designed to satisfy the requirements of:

• ANSI/IACET 1-2018 Standard for Continuing Education and Training — Elements 2 (learning environment), 3 (instructional personnel), 5 (learning outcomes), 6 (content and instruction), 7 (assessment), and 9 (evaluation)
• ASTM E3416-24 Standard Practice for Competency-Based Work-Based Learning Programs — competency alignment, formative and summative assessment, simulated workplace settings, credential issuance
• CPD Standards Office accreditation requirements for continuing professional development

Organizational policies supporting this framework are documented in the policies section.