Research system · v2.0.0 · validated synthesis

Design the thinking, not just the screen.

A human-centered operating system for comprehension, hierarchy, progressive disclosure, ethical choice architecture, accessibility, and AI-assisted work.

Core decision: make the user’s thinking proportional to the task. Preserve necessary complexity. Remove interface-created interpretation, recall, uncertainty, context loss, and recovery effort.

Understand the model Apply the patterns Audit a release

01 · Orientation

Lower friction without flattening the domain.

Cognitive UX is not a mandate for sparse screens. It is a disciplined way to align information, interaction, and evidence with how people perceive, orient, understand, choose, act, verify, recover, and learn.

Optimize for

Accurate mental models.
Clear next actions and consequences.
Stable orientation and resumability.
Verification and recovery proportional to risk.
Meaningful choice and user autonomy.

Do not optimize blindly for

Minimum clicks or minimum visible content.
Engagement, session duration, or interaction count.
Aesthetic cleanliness that removes cues.
A universal density or novice-only workflow.
An invented “cognitive load score.”

How to read the guidance

Direct support

A standard, official guidance, or empirical source directly supports the stated claim.

Supported synthesis

Multiple sources support a combined operating conclusion.

Transfer inference

Evidence from learning, policy, or another domain is applied to product UX with an explicit boundary.

Recommendation

A proposed house rule or operating default that still requires product-specific validation.

02 · Red team

Guard against overclaiming and cosmetic simplification.

The model is useful only when its boundaries remain visible. These cautions block common misapplications.

Keep

Cognitive Load Theory as an analytical lens.
Working memory as context-sensitive and chunk-dependent.
Progressive disclosure as timing and relevance control.
User testing as the authority for product-specific outcomes.

Reject

Fixed item-count rules derived from memory research.
“One thing per page” as a universal enterprise pattern.
Static cognitive scores presented as human measurement.
Hiding essential content to improve screenshots.

Transfer inference

Learning theory is not interface law.

Instructional evidence informs UX, but domain expertise, time pressure, risk, and interaction structure change the result.

S12 S13

Supported synthesis

Choice overload is conditional.

Option count matters less than relevance, comparability, task difficulty, preference certainty, and decision goals.

S14

Recommendation

Adaptive complexity beats permanent simplification.

Novices need orientation. Experts may need density, batch action, shortcuts, persistent filters, and parallel comparison.

03 · Human model

Support the complete comprehension loop.

Failures propagate. Weak orientation increases interpretation effort. Weak feedback increases uncertainty. Weak recovery suppresses exploration and trust.

Acquire context

01PerceiveNotice the relevant information, state, or control.
02OrientUnderstand location, scope, priority, and relationship to the larger task.

Form meaning

03UnderstandTranslate labels, content, and system state into a usable mental model.
04ChooseCompare relevant options and predict consequences.

Change the system

05ActExecute the intended change through an understandable control.
06VerifyConfirm what was received, changed, generated, or completed.

Preserve continuity

07RecoverCorrect, undo, resume, or safely exit when conditions change.
08LearnBuild a stable schema that reduces effort in future interactions.

S01 S04

04 · Load model

Review cognitive demand as a system of risks.

A sparse screen can be cognitively expensive. A dense screen can remain manageable when its relationships, terminology, state, and priority are stable.

Operational model

Total demand = necessary task complexity + interface interpretation + memory burden + decision uncertainty + context switching + interruption recovery + perceived consequence

Interpretation

Labels, icons, layout, and system states require translation.

Observe: hesitation, misclicks, help opening, inconsistent explanations.

Memory

Users must retain prior values, rules, or location across steps.

Observe: backtracking, repeated entry, notes, copy/paste, abandoned flows.

Decision

Options are irrelevant, overlapping, difficult to compare, or weakly explained.

Observe: oscillation, default dependence, low confidence, immediate reversal.

Orientation

Users cannot locate themselves, active state, parent context, or return path.

Observe: repeated navigation, search loops, lost filters, duplicate work.

Interruption

Notifications, modal surfaces, latency, or context switches disrupt the task.

Observe: resumption delay, partial input loss, restart, missed status changes.

Consequence

Unclear impact or weak recovery makes users anxious or overly cautious.

Observe: avoidance, repeated review, support contact, failed cancellation.

S11 S12 S13

05 · Structure

Use hierarchy to expose the product’s conceptual model.

Visual hierarchy routes attention. Information architecture establishes location. Communication hierarchy establishes meaning. They must describe the same system.

Page anatomy

Identity and purpose

What this surface is and why it exists.

Current state or result

What is true now, including active constraints and status.

Primary decision or action

The next meaningful user outcome and its consequence.

Supporting explanation

Relevant groups, comparison, and guidance.

Evidence and audit

Sources, history, ownership, assumptions, and deeper detail.

Recovery and continuation

Help, undo, return, escalation, and next steps.

Information-architecture workflow

Inventory content and tasks.
Identify duplication, dependencies, entry points, and failure paths.
Collect user vocabulary.
Use support, search, research, and domain language—not organizational labels.
Discover grouping.
Use open sorting, journey analysis, and object/lifecycle modeling.
Validate findability.
Use tree testing, first-click testing, and representative task scenarios.
Validate comprehension.
Ask users to explain location, consequence, and next action.
Observe production recovery.
Track loops, backtracking, search reformulation, and support escalation.

Deep-link rule

Every important page must stand on its own. Users may arrive through search, notifications, shared links, or AI-generated answers. Preserve product identity, parent context, current location, active state, and a path to related information.

S04 S09

06 · Content hierarchy

Write for scanning first and understanding second.

Readers should not need to consume every word to identify the main claim, implication, action, and evidence boundary.

Front-load meaning

Put the unique, task-relevant words first in titles, headings, navigation labels, buttons, and links.

S05 S24

Use headings as an outline

Headings should accurately describe the following section and preserve logical nesting for visual and assistive navigation.

S03 S04

Make actions specific

Name the object and outcome: “Create access policy,” not “Submit.” State consequence before commitment.

S01

Summarize before detail

Use result → implication → action → explanation → evidence. Preserve deep detail, but do not require it for orientation.

Use examples at the point of uncertainty

Examples work best beside unfamiliar concepts, formats, edge cases, or decisions—not in a detached onboarding tour.

Keep terminology stable

One concept should keep one name across navigation, headings, controls, messages, and documentation.

S03

07 · Progressive disclosure

Control timing without concealing consequences.

Progressive disclosure determines when, where, and under what conditions complexity appears. It is not a visual cleanup technique.

Keep visible

Purpose, current state, and primary action.
Cost, risk, legal or material consequence.
Comparison criteria and active constraints.
Requirements, errors, and recovery.
Information needed for the current decision.

Disclose later when appropriate

Advanced or infrequent configuration.
Conditional fields after the controlling answer.
Background explanation and examples.
Historical, debugging, or raw audit detail.
Supporting evidence after an accurate summary.

Disclosure decision protocol

Select every condition that applies. The recommendation is a starting point, not proof.

Required for the current decisionIncludes cost, risk, consequence, requirements, or comparison criteria. Must be compared with other informationUsers need simultaneous visibility to evaluate differences. Relevant only after a choice or stateThe content has a clear controlling condition. Infrequent or expert-onlySafe defaults exist and core users do not need it routinely. Can be summarized accuratelyThe trigger can predict the hidden content without vague wording. Deep, persistent, or independently linkableThe content needs search, sharing, history, or stable navigation.

Start with inline content

No disclosure conditions are selected. Keep the content visible until its role is clear.

S01 S08 S09

08 · Pattern library

Choose patterns by human problem, not component preference.

A component defines mechanics. A pattern defines when a coordinated set of content, controls, state, and recovery solves a user need.

12 patterns

Orientorientation

Purpose and state header

Expose page identity, scope, current state, and the primary outcome before secondary controls.

Use when: Task pages, dashboards, settings, and deep-linked views.
Avoid when: Marketing copy that delays the actual purpose.

S01 S04 S05

Understandinterpretation

Summary → implication → evidence

Lead with the finding, explain why it matters, then progressively expose supporting detail and provenance.

Use when: Research, analytics, architecture decisions, and incident reviews.
Avoid when: Evidence dumps without an orienting conclusion.

S04 S05 S24

Choosememory

Recognition over recall

Keep prior input, active constraints, history, and relevant options visible at the point of use.

Use when: Multi-step forms, comparison, and operational workflows.
Avoid when: Requiring users to remember values from previous screens.

S01 S22

Choosedecision

Recommended path with alternatives

State the recommended option and rationale while preserving inspectable alternatives and consequences.

Use when: Configuration, product selection, policy choices, and AI recommendations.
Avoid when: Preselection that hides trade-offs or makes alternatives materially harder.

S14 S15

Understandrelevance

Conditional inline disclosure

Reveal dependent fields or guidance immediately after the answer or state that makes them relevant.

Use when: Branching forms and role- or state-dependent configuration.
Avoid when: Detached panels where the cause-and-effect relationship is unclear.

S07 S09

Actdensity

Labeled advanced section

Place infrequent expert controls behind an accurate summary while keeping defaults and current effects visible.

Use when: Expert configuration with safe defaults.
Avoid when: Hiding controls that frequent users require or that alter material outcomes.

S01 S08

Verifyuncertainty

Local action feedback

Show processing, success, failure, and changed state next to the object or control that caused it.

Use when: Inline edits, saves, filters, and background operations.
Avoid when: Generic toasts as the only evidence of a durable change.

S01 S03

Recoverconsequence

Safe recovery path

Provide undo, drafts, review, version history, or explicit confirmation in proportion to consequence.

Use when: Destructive, financial, publishing, access, and agentic actions.
Avoid when: Confirmation dialogs for trivial reversible actions or no recovery for high-impact changes.

S01 S03 S20

Recoverorientation

Stable help location

Keep help and escalation mechanisms in a consistent relative location across related pages.

Use when: Multi-page services and enterprise workflows.
Avoid when: Moving support controls based on page layout convenience.

S23

Orientattention

Natural stopping points

Create resumable boundaries with saved state, clear progress, and a stable return path.

Use when: Long workflows, research guides, and mobile tasks subject to interruption.
Avoid when: Endless goal-directed feeds without landmarks.

S01 S09

Verifyoverreliance

Verification surface for AI

Make sources, assumptions, uncertainty, and affected objects easy to inspect before consequential use.

Use when: AI-assisted decisions, summaries, and agent execution.
Avoid when: Generic disclaimers that add reading without making verification easier.

S17 S19 S20

Chooseautonomy

Choice and exit parity

Give comparable choices comparable clarity, prominence, and reasonable effort across the full lifecycle.

Use when: Consent, subscription, notification, privacy, and account flows.
Avoid when: Easy acceptance paired with concealed or obstructed reversal.

S15 S16

09 · Ethical behavior

Help users enact intent without manufacturing it.

Behavioral design is legitimate when it reduces initiation, comparison, and recovery costs. It becomes manipulative when it exploits inattention, urgency, asymmetry, or hidden consequence.

Choice-parity test

Dimension	Accept or enable	Reject, cancel, or disable	Review question
Clarity	Specific label and consequence	Specific label and consequence	Can both choices be understood without interpretation?
Prominence	Visible in the decision context	Visible in the same decision context	Does visual treatment materially steer one option?
Effort	Steps proportional to consequence	Steps proportional to consequence	Is reversal deliberately more difficult?
Timing	Asked when relevant	Available throughout lifecycle	Does timing exploit fatigue or urgency?
Recovery	Confirmation and state	Confirmation and state	Can users verify that their choice took effect?

False minimalism

Removes labels, navigation, state, or help to make a screen appear simple.

Replace withKeep necessary cues visible and reduce decorative competition.

Card soup

Places every idea in an equally weighted container.

Replace withUse boundaries only for real relationships, states, or interaction contracts.

Choice dumping

Transfers prioritization and comparison work to the user.

Replace withFilter irrelevant options, group meaningfully, and explain recommendations.

Hidden essentials

Conceals cost, risk, requirements, or comparison criteria behind disclosure.

Replace withKeep decision-essential information visible at the point of choice.

Nested disclosure trees

Creates unstable location and forces users to remember which branches contain what.

Replace withUse shallower sections, accurate labels, and linkable pages for deep material.

Tour-first onboarding

Explains features before the user has a task context.

Replace withUse contextual guidance at the point of first meaningful use.

Tooltip documentation

Makes required instructions transient and difficult to discover.

Replace withUse persistent helper text or linkable guidance.

Modal cascades

Repeatedly destroys context and traps keyboard or touch navigation.

Replace withUse a dedicated flow, inline expansion, or a contextual panel.

Silent automation

Changes data or state without clear verification or history.

Replace withExpose status, affected objects, provenance, and recovery.

Asymmetric friction

Makes the organization-preferred action easier than rejection, cancellation, or reversal.

Replace withApply choice and exit parity with proportionate safeguards.

Generic AI disclaimers

Adds warning text but does not help users detect or verify mistakes.

Replace withSignal when verification matters and make evidence easy to inspect.

Static cognitive score

Presents a design-review number as if it measured a person’s mental workload.

Replace withUse observable risk classifications plus task and comprehension evidence.

S15 S16

10 · AI and agents

Design for calibrated reliance and inspectable action.

AI adds probabilistic output, uncertain capability, hidden tool use, and delegated action. The interface must lower the effort required to understand, monitor, verify, correct, and stop the system.

Agent interaction contract

Scope

State the goal, known context, constraints, data boundaries, and capability limits.

Plan

Expose the intended steps, assumptions, dependencies, and affected systems before execution.

Authorize

Require proportional human approval for high-impact, external, destructive, or ambiguous actions.

Execute

Show current action, tool, target, progress, and exceptions without flooding the user with raw traces.

Verify

Expose results, sources, changed objects, uncertainty, and what still requires human judgement.

Recover

Support pause, correction, retry, rollback, escalation, and durable audit history.

Overreliance controls

Supported synthesis

Build a realistic mental model

Explain what the system can and cannot do, what kinds of mistakes are plausible, and which context it can access.

Supported synthesis

Signal when verification matters

Use risk, uncertainty, missing evidence, and hard-to-detect error conditions—not generic warnings on every response.

Recommendation

Make verification cheaper than blind trust

Place sources, assumptions, comparisons, and affected objects next to the output. Preserve the user’s ability to inspect before acting.

S17 S18 S19 S20

11 · Review lab

Classify observable risks before release.

This review does not measure a person’s cognitive load. It identifies interface conditions that should be resolved or validated with representative users.

Goal clarity

Can users state what this surface is for and the outcome it supports?

Orientation

Can users identify location, active state, parent context, and a return path?

Priority

Does hierarchy expose the most important information and action without competing signals?

Decision complexity

Are choices relevant, distinct, comparable, and explicit about consequence?

Memory burden

Must users remember information that could remain visible, selectable, or auto-populated?

Terminology

Do labels match user vocabulary and remain consistent across the workflow?

Disclosure integrity

Is all decision-essential information visible when the choice is made?

Feedback

Can users verify processing, completion, changed state, and affected objects?

Recovery

Can users correct, undo, resume, retry, escalate, or safely exit?

Interruption

Is attention requested only when urgency justifies it, and can the task resume?

Accessibility

Do semantics, focus, keyboard, target geometry, reflow, and announcements match visual behavior?

Ethical parity

Do alternatives, refusal, cancellation, and reversal have comparable clarity and reasonable effort?

12 · Measurement

Measure understanding, not visual preference.

Task completion can be accidental. Pair performance with explanation, prediction, confidence, recovery, and resumption evidence.

Performance

Task success and time.
Error count and severity.
Backtracking and repeated navigation.
Abandonment and recovery success.
Unnecessary actions.

Comprehension

Explain what happened.
Predict an action’s consequence.
Identify current state and next step.
Locate supporting evidence.
Resume after interruption.

Subjective workload

Mental effort.
Confidence.
Frustration.
Perceived temporal pressure.
NASA-TLX when research rigor warrants it.

S10

Minimum study matrix

Variable	Minimum contrast	Reason
Expertise	New, occasional, experienced	Familiar schemas change memory and comparison demands.
Attention	Focused and interrupted	Real tasks include notification, authentication, and context switching.
Consequence	Low and high impact	Risk changes verification, hesitation, and recovery needs.
Device	Narrow touch and desktop keyboard	Layout, target geometry, density, and input modality change behavior.
Accessibility	Keyboard, screen reader, zoom, cognitive-accessibility participants where applicable	Conformance checks do not prove usability.
State	Empty, populated, loading, error, restricted	Cognitive failures often appear outside the ideal success state.

13 · Delivery system

Operationalize the guidance as product evidence.

A design principle becomes durable when it has an owner, implementation contract, test method, exception path, and recorded outcome.

Product and content

Define user outcome and consequence.
Separate required from optional information.
Use stable vocabulary and explicit recommendations.
Define ethical success metrics.

Design and research

Model orientation, decision, feedback, and recovery.
Test the smallest credible context.
Include interruption and edge states.
Record comprehension evidence.

Engineering and accessibility

Use semantic native foundations.
Preserve focus, keyboard, state, and deep links.
Support reflow, target size, text spacing, and reduced motion.
Make failures observable and recoverable.

Release evidence

Context
Audience, task, device, risk, authorization, and environment.
Claims
What the design is expected to improve and the comparator.
Implementation
Semantic structure, interaction contract, failure states, and recovery.
Validation
Automated, manual, accessibility, performance, and representative-user evidence.
Decision
Accepted risks, owner, follow-up, and revalidation trigger.

14 · Evidence

Keep authority inspectable.

The guide distinguishes normative standards, empirical research, official practice guidance, supported synthesis, transfer inference, and recommended operating defaults.

Method

Reviewed current W3C accessibility standards and cognitive guidance.
Revisited cognitive-load, working-memory, and choice-overload research.
Compared public-sector and design-system disclosure practices.
Integrated current human-AI and agent-interaction guidance.
Red-teamed transfer assumptions, ethical risks, and measurement claims.
Converted the result into executable patterns and a release review.

Material limitations

No static interface inspection measures an individual’s cognitive workload.
Cognitive Load Theory is transferred from learning research with explicit limits.
Design-system guidance is evidence of intended practice, not universal empirical proof.
AI explanations and warnings can increase or decrease overreliance; validate them in context.
WCAG 3 material is draft direction, not a current conformance standard.

Machine-readable research manifest

{
  "artifact": {
    "title": "Cognitive UX Research System",
    "version": "2.0.0",
    "dateModified": "2026-07-21",
    "format": "single-file HTML",
    "dependencies": [],
    "owner": "Jesse Graupmann",
    "contributor": "OpenAI ChatGPT \u00b7 GPT-5.6 Thinking",
    "derivedFrom": [
      "Cognitive UX Research System v1.2.0",
      "The Communication Guide v2.1.0",
      "Visual Design Research System v1.5.1",
      "Mobile-First Enterprise Design Handbook v3.0",
      "Modern CSS Architecture Explainer v1.2"
    ]
  },
  "methodology": {
    "approach": "standards review, peer-reviewed literature synthesis, product-practice comparison, red-team analysis, and executable pattern formalization",
    "claimTypes": [
      "Direct support",
      "Supported synthesis",
      "Transfer inference",
      "Recommendation"
    ],
    "limitations": [
      "Cognitive Load Theory originated in learning research and is used here as an analytical lens rather than a deterministic UI law.",
      "Static design inspection cannot measure an individual user's cognitive workload.",
      "Product risk, domain expertise, disability, language, interruption, and consequence materially change results.",
      "AI overreliance mitigations require product-specific user research because warnings and explanations can backfire."
    ]
  },
  "counts": {
    "principles": 12,
    "patterns": 12,
    "antiPatterns": 12,
    "sources": 24
  },
  "sources": [
    {
      "id": "S01",
      "title": "Making Content Usable for People with Cognitive and Learning Disabilities",
      "publisher": "W3C WAI",
      "type": "Guidance",
      "status": "Working Group Note",
      "year": "2021",
      "url": "https://www.w3.org/TR/coga-usable/",
      "use": "Cognitive accessibility objectives, user needs, design patterns, and testing questions."
    },
    {
      "id": "S02",
      "title": "Cognitive Accessibility at W3C",
      "publisher": "W3C WAI",
      "type": "Program overview",
      "status": "Current guidance hub",
      "year": "2026",
      "url": "https://www.w3.org/WAI/cognitive/",
      "use": "Readable, predictable, and error-tolerant experience framing."
    },
    {
      "id": "S03",
      "title": "Web Content Accessibility Guidelines 2.2",
      "publisher": "W3C",
      "type": "Standard",
      "status": "W3C Recommendation",
      "year": "2024",
      "url": "https://www.w3.org/TR/WCAG22/",
      "use": "Normative accessibility baseline, including focus, target size, consistent help, redundant entry, and accessible authentication."
    },
    {
      "id": "S04",
      "title": "Page Structure Tutorial",
      "publisher": "W3C WAI",
      "type": "Practice guidance",
      "status": "Updated April 2026",
      "year": "2026",
      "url": "https://www.w3.org/WAI/tutorials/page-structure/",
      "use": "Landmarks, headings, content structure, navigation, and orientation."
    },
    {
      "id": "S05",
      "title": "Writing for Web Accessibility",
      "publisher": "W3C WAI",
      "type": "Practice guidance",
      "status": "Current",
      "year": "2024",
      "url": "https://www.w3.org/WAI/tips/writing/",
      "use": "Meaningful headings, front-loaded information, clear instructions, and understandable links."
    },
    {
      "id": "S06",
      "title": "Requirements for WCAG 3.0",
      "publisher": "W3C",
      "type": "Draft requirements",
      "status": "Draft \u2014 not normative",
      "year": "2026",
      "url": "https://www.w3.org/TR/wcag-3.0-requirements/",
      "use": "Emerging direction toward readable, technology-neutral, user-centered outcomes."
    },
    {
      "id": "S07",
      "title": "Question pages",
      "publisher": "GOV.UK Design System",
      "type": "Design pattern",
      "status": "Production guidance",
      "year": "2026",
      "url": "https://design-system.service.gov.uk/patterns/question-pages/",
      "use": "Focused question flows, labels as headings, and stepwise data collection."
    },
    {
      "id": "S08",
      "title": "Details component",
      "publisher": "GOV.UK Design System",
      "type": "Component guidance",
      "status": "Production guidance",
      "year": "2026",
      "url": "https://design-system.service.gov.uk/components/details/",
      "use": "Choosing between details, accordions, and tabs; keeping important content visible."
    },
    {
      "id": "S09",
      "title": "Information architects and accessible service design",
      "publisher": "UK Government Digital Service",
      "type": "Practice report",
      "status": "Published May 2025",
      "year": "2025",
      "url": "https://accessibility.blog.gov.uk/2025/05/09/unlocking-accessibility-information-architects-share-their-approach-to-digital-accessibility-and-design-with-people-with-disabilities/",
      "use": "Hierarchy, service-flow orientation, progressive disclosure, and participatory design."
    },
    {
      "id": "S10",
      "title": "NASA Task Load Index",
      "publisher": "NASA",
      "type": "Measurement instrument",
      "status": "Established instrument",
      "year": "Current",
      "url": "https://www.nasa.gov/human-systems-integration-division/nasa-task-load-index-tlx/",
      "use": "Subjective workload dimensions; supplemental rather than objective UI scoring."
    },
    {
      "id": "S11",
      "title": "The Magical Number 4 in Short-Term Memory",
      "publisher": "Behavioral and Brain Sciences",
      "type": "Peer-reviewed research",
      "status": "Published",
      "year": "2001",
      "url": "https://doi.org/10.1017/S0140525X01003922",
      "use": "Working-memory capacity as a variable, chunk-dependent constraint rather than a fixed UI item limit."
    },
    {
      "id": "S12",
      "title": "Cognitive Load During Problem Solving",
      "publisher": "Cognitive Science",
      "type": "Peer-reviewed research",
      "status": "Published",
      "year": "1988",
      "url": "https://doi.org/10.1207/s15516709cog1202_4",
      "use": "Foundational intrinsic and extraneous cognitive-load framing."
    },
    {
      "id": "S13",
      "title": "Element Interactivity and Cognitive Load",
      "publisher": "Educational Psychology Review",
      "type": "Peer-reviewed research",
      "status": "Published",
      "year": "2010",
      "url": "https://doi.org/10.1007/s10648-010-9128-5",
      "use": "Task complexity depends on interacting information elements and user schemas."
    },
    {
      "id": "S14",
      "title": "Choice Overload: A Conceptual Review and Meta-Analysis",
      "publisher": "Journal of Consumer Psychology",
      "type": "Peer-reviewed meta-analysis",
      "status": "Published",
      "year": "2015",
      "url": "https://doi.org/10.1016/j.jcps.2014.08.002",
      "use": "Choice overload is conditional on task difficulty, preference uncertainty, and decision goals."
    },
    {
      "id": "S15",
      "title": "Dark Commercial Patterns",
      "publisher": "OECD",
      "type": "Policy research",
      "status": "Published",
      "year": "2022",
      "url": "https://www.oecd.org/en/publications/dark-commercial-patterns_44f5e846-en.html",
      "use": "Autonomy, coercion, obstruction, manipulation, and consumer detriment in digital choice architecture."
    },
    {
      "id": "S16",
      "title": "Bringing Dark Patterns to Light",
      "publisher": "US Federal Trade Commission",
      "type": "Regulatory report",
      "status": "Published",
      "year": "2022",
      "url": "https://www.ftc.gov/reports/bringing-dark-patterns-light",
      "use": "Disguised ads, difficult cancellation, hidden terms, and unnecessary data-sharing pressure."
    },
    {
      "id": "S17",
      "title": "Explainability and Trust",
      "publisher": "Google People + AI Research",
      "type": "Human-AI design guidance",
      "status": "Living guidebook",
      "year": "Current",
      "url": "https://pair.withgoogle.com/chapter/explainability-trust/",
      "use": "Trust calibration, capability boundaries, uncertainty, and explanation timing."
    },
    {
      "id": "S18",
      "title": "Guidelines for Human-AI Interaction",
      "publisher": "Microsoft Research",
      "type": "Peer-reviewed HCI research",
      "status": "Published",
      "year": "2019",
      "url": "https://www.microsoft.com/en-us/research/publication/guidelines-for-human-ai-interaction/",
      "use": "Eighteen validated interaction guidelines covering expectation-setting, correction, and adaptation."
    },
    {
      "id": "S19",
      "title": "Overreliance on AI: Risk Identification and Mitigation",
      "publisher": "Microsoft",
      "type": "Product safety playbook",
      "status": "Published March 2025",
      "year": "2025",
      "url": "https://learn.microsoft.com/en-us/ai/playbook/technology-guidance/overreliance-on-ai/overreliance-on-ai",
      "use": "Realistic mental models, verification signals, and lowering the effort required to check AI output."
    },
    {
      "id": "S20",
      "title": "Challenges in Human-Agent Communication",
      "publisher": "Microsoft Research",
      "type": "Research paper",
      "status": "Published",
      "year": "2024",
      "url": "https://www.microsoft.com/en-us/research/publication/challenges-in-human-agent-communication/",
      "use": "Goal expression, plan correction, monitoring, feedback, transparency, and user control for agents."
    },
    {
      "id": "S21",
      "title": "Understanding Target Size (Minimum)",
      "publisher": "W3C WAI",
      "type": "Normative explanation",
      "status": "WCAG 2.2 AA",
      "year": "2026",
      "url": "https://www.w3.org/WAI/WCAG22/Understanding/target-size-minimum",
      "use": "Minimum pointer-target geometry and spacing exceptions."
    },
    {
      "id": "S22",
      "title": "Understanding Redundant Entry",
      "publisher": "W3C WAI",
      "type": "Normative explanation",
      "status": "WCAG 2.2 A",
      "year": "2026",
      "url": "https://www.w3.org/WAI/WCAG22/Understanding/redundant-entry.html",
      "use": "Avoiding repeated entry and unnecessary memory demand in multi-step processes."
    },
    {
      "id": "S23",
      "title": "Understanding Consistent Help",
      "publisher": "W3C WAI",
      "type": "Normative explanation",
      "status": "WCAG 2.2 A",
      "year": "2026",
      "url": "https://www.w3.org/WAI/WCAG22/Understanding/consistent-help.html",
      "use": "Stable placement of help mechanisms across page sets."
    },
    {
      "id": "S24",
      "title": "Plain language",
      "publisher": "Office for National Statistics Service Manual",
      "type": "Content practice",
      "status": "Current guidance",
      "year": "2026",
      "url": "https://service-manual.ons.gov.uk/content/writing-for-users/plain-language",
      "use": "Front-loaded headings, clear vocabulary, and concise web writing."
    }
  ],
  "patterns": [
    {
      "id": "purpose-state",
      "title": "Purpose and state header",
      "phase": "orient",
      "risk": "orientation",
      "summary": "Expose page identity, scope, current state, and the primary outcome before secondary controls.",
      "use": "Task pages, dashboards, settings, and deep-linked views.",
      "avoid": "Marketing copy that delays the actual purpose.",
      "refs": [
        "S01",
        "S04",
        "S05"
      ]
    },
    {
      "id": "summary-evidence",
      "title": "Summary \u2192 implication \u2192 evidence",
      "phase": "understand",
      "risk": "interpretation",
      "summary": "Lead with the finding, explain why it matters, then progressively expose supporting detail and provenance.",
      "use": "Research, analytics, architecture decisions, and incident reviews.",
      "avoid": "Evidence dumps without an orienting conclusion.",
      "refs": [
        "S04",
        "S05",
        "S24"
      ]
    },
    {
      "id": "recognition",
      "title": "Recognition over recall",
      "phase": "choose",
      "risk": "memory",
      "summary": "Keep prior input, active constraints, history, and relevant options visible at the point of use.",
      "use": "Multi-step forms, comparison, and operational workflows.",
      "avoid": "Requiring users to remember values from previous screens.",
      "refs": [
        "S01",
        "S22"
      ]
    },
    {
      "id": "recommended-path",
      "title": "Recommended path with alternatives",
      "phase": "choose",
      "risk": "decision",
      "summary": "State the recommended option and rationale while preserving inspectable alternatives and consequences.",
      "use": "Configuration, product selection, policy choices, and AI recommendations.",
      "avoid": "Preselection that hides trade-offs or makes alternatives materially harder.",
      "refs": [
        "S14",
        "S15"
      ]
    },
    {
      "id": "conditional-inline",
      "title": "Conditional inline disclosure",
      "phase": "understand",
      "risk": "relevance",
      "summary": "Reveal dependent fields or guidance immediately after the answer or state that makes them relevant.",
      "use": "Branching forms and role- or state-dependent configuration.",
      "avoid": "Detached panels where the cause-and-effect relationship is unclear.",
      "refs": [
        "S07",
        "S09"
      ]
    },
    {
      "id": "advanced-section",
      "title": "Labeled advanced section",
      "phase": "act",
      "risk": "density",
      "summary": "Place infrequent expert controls behind an accurate summary while keeping defaults and current effects visible.",
      "use": "Expert configuration with safe defaults.",
      "avoid": "Hiding controls that frequent users require or that alter material outcomes.",
      "refs": [
        "S01",
        "S08"
      ]
    },
    {
      "id": "local-feedback",
      "title": "Local action feedback",
      "phase": "verify",
      "risk": "uncertainty",
      "summary": "Show processing, success, failure, and changed state next to the object or control that caused it.",
      "use": "Inline edits, saves, filters, and background operations.",
      "avoid": "Generic toasts as the only evidence of a durable change.",
      "refs": [
        "S01",
        "S03"
      ]
    },
    {
      "id": "safe-recovery",
      "title": "Safe recovery path",
      "phase": "recover",
      "risk": "consequence",
      "summary": "Provide undo, drafts, review, version history, or explicit confirmation in proportion to consequence.",
      "use": "Destructive, financial, publishing, access, and agentic actions.",
      "avoid": "Confirmation dialogs for trivial reversible actions or no recovery for high-impact changes.",
      "refs": [
        "S01",
        "S03",
        "S20"
      ]
    },
    {
      "id": "stable-help",
      "title": "Stable help location",
      "phase": "recover",
      "risk": "orientation",
      "summary": "Keep help and escalation mechanisms in a consistent relative location across related pages.",
      "use": "Multi-page services and enterprise workflows.",
      "avoid": "Moving support controls based on page layout convenience.",
      "refs": [
        "S23"
      ]
    },
    {
      "id": "natural-stops",
      "title": "Natural stopping points",
      "phase": "orient",
      "risk": "attention",
      "summary": "Create resumable boundaries with saved state, clear progress, and a stable return path.",
      "use": "Long workflows, research guides, and mobile tasks subject to interruption.",
      "avoid": "Endless goal-directed feeds without landmarks.",
      "refs": [
        "S01",
        "S09"
      ]
    },
    {
      "id": "verification-surface",
      "title": "Verification surface for AI",
      "phase": "verify",
      "risk": "overreliance",
      "summary": "Make sources, assumptions, uncertainty, and affected objects easy to inspect before consequential use.",
      "use": "AI-assisted decisions, summaries, and agent execution.",
      "avoid": "Generic disclaimers that add reading without making verification easier.",
      "refs": [
        "S17",
        "S19",
        "S20"
      ]
    },
    {
      "id": "choice-parity",
      "title": "Choice and exit parity",
      "phase": "choose",
      "risk": "autonomy",
      "summary": "Give comparable choices comparable clarity, prominence, and reasonable effort across the full lifecycle.",
      "use": "Consent, subscription, notification, privacy, and account flows.",
      "avoid": "Easy acceptance paired with concealed or obstructed reversal.",
      "refs": [
        "S15",
        "S16"
      ]
    }
  ]
}

S01Guidance

Making Content Usable for People with Cognitive and Learning Disabilities

W3C WAI · 2021 · Working Group Note

Cognitive accessibility objectives, user needs, design patterns, and testing questions.

Open source

S02Program overview

Cognitive Accessibility at W3C

W3C WAI · 2026 · Current guidance hub

Readable, predictable, and error-tolerant experience framing.

Open source

S03Standard

Web Content Accessibility Guidelines 2.2

W3C · 2024 · W3C Recommendation

Normative accessibility baseline, including focus, target size, consistent help, redundant entry, and accessible authentication.

Open source

S04Practice guidance

Page Structure Tutorial

W3C WAI · 2026 · Updated April 2026

Landmarks, headings, content structure, navigation, and orientation.

Open source

S05Practice guidance

Writing for Web Accessibility

W3C WAI · 2024 · Current

Meaningful headings, front-loaded information, clear instructions, and understandable links.

Open source

S06Draft requirements

Requirements for WCAG 3.0

W3C · 2026 · Draft — not normative

Emerging direction toward readable, technology-neutral, user-centered outcomes.

Open source

S07Design pattern

Question pages

GOV.UK Design System · 2026 · Production guidance

Focused question flows, labels as headings, and stepwise data collection.

Open source

S08Component guidance

Details component

GOV.UK Design System · 2026 · Production guidance

Choosing between details, accordions, and tabs; keeping important content visible.

Open source

S09Practice report

Information architects and accessible service design

UK Government Digital Service · 2025 · Published May 2025

Hierarchy, service-flow orientation, progressive disclosure, and participatory design.

Open source

S10Measurement instrument

NASA Task Load Index

NASA · Current · Established instrument

Subjective workload dimensions; supplemental rather than objective UI scoring.

Open source

S11Peer-reviewed research

The Magical Number 4 in Short-Term Memory

Behavioral and Brain Sciences · 2001 · Published

Working-memory capacity as a variable, chunk-dependent constraint rather than a fixed UI item limit.

Open source

S12Peer-reviewed research

Cognitive Load During Problem Solving

Cognitive Science · 1988 · Published

Foundational intrinsic and extraneous cognitive-load framing.

Open source

S13Peer-reviewed research

Element Interactivity and Cognitive Load

Educational Psychology Review · 2010 · Published

Task complexity depends on interacting information elements and user schemas.

Open source

S14Peer-reviewed meta-analysis

Choice Overload: A Conceptual Review and Meta-Analysis

Journal of Consumer Psychology · 2015 · Published

Choice overload is conditional on task difficulty, preference uncertainty, and decision goals.

Open source

S15Policy research

Dark Commercial Patterns

OECD · 2022 · Published

Autonomy, coercion, obstruction, manipulation, and consumer detriment in digital choice architecture.

Open source

S16Regulatory report

Bringing Dark Patterns to Light

US Federal Trade Commission · 2022 · Published

Disguised ads, difficult cancellation, hidden terms, and unnecessary data-sharing pressure.

Open source

S17Human-AI design guidance

Explainability and Trust

Google People + AI Research · Current · Living guidebook

Trust calibration, capability boundaries, uncertainty, and explanation timing.

Open source

S18Peer-reviewed HCI research

Guidelines for Human-AI Interaction

Microsoft Research · 2019 · Published

Eighteen validated interaction guidelines covering expectation-setting, correction, and adaptation.

Open source

S19Product safety playbook

Overreliance on AI: Risk Identification and Mitigation

Microsoft · 2025 · Published March 2025

Realistic mental models, verification signals, and lowering the effort required to check AI output.

Open source

S20Research paper

Challenges in Human-Agent Communication

Microsoft Research · 2024 · Published

Goal expression, plan correction, monitoring, feedback, transparency, and user control for agents.

Open source

S21Normative explanation

Understanding Target Size (Minimum)

W3C WAI · 2026 · WCAG 2.2 AA

Minimum pointer-target geometry and spacing exceptions.

Open source

S22Normative explanation

Understanding Redundant Entry

W3C WAI · 2026 · WCAG 2.2 A

Avoiding repeated entry and unnecessary memory demand in multi-step processes.

Open source

S23Normative explanation

Understanding Consistent Help

W3C WAI · 2026 · WCAG 2.2 A

Stable placement of help mechanisms across page sets.

Open source

S24Content practice

Plain language

Office for National Statistics Service Manual · 2026 · Current guidance

Front-loaded headings, clear vocabulary, and concise web writing.

Open source