Overview

Decision layer first, evidence and audit available without competing for attention

This system combines cognitive science, cognitive accessibility, information architecture, communication hierarchy, progressive disclosure, interaction design, ethical choice architecture, and product measurement. It is an implementation standard rather than a collection of aesthetic tips.

TL;DR

Design the user’s path through meaning.

Users should not spend effort discovering what a surface is, where they are, what matters, what will happen, whether an action worked, or how to recover. Preserve thinking that belongs to the decision. Remove thinking created by the interface.

Choose a role lens

The four-layer communication architecture

Layer 1

Decision

Purpose, current state, implication, recommended next step, and consequence.

Layer 2

Execution

Controls, workflow, instructions, dependencies, feedback, and recovery.

Layer 3

Evidence

Supporting explanation, comparisons, confidence, trade-offs, and source linkage.

Layer 4

Audit

Provenance, ownership, history, assumptions, open challenges, and test results.

The product objective

Necessary complexityThe domain, decision, or task complexity that cannot be removed without changing the outcome.

Interpretation costEffort spent decoding structure, language, controls, hierarchy, or system behavior.

Memory burdenInformation users must retain because the interface does not keep it visible or recoverable.

Decision uncertaintyEffort spent comparing unclear options or predicting hidden consequences.

Context recoveryEffort required after navigation, interruption, authentication, error, or device change.

Consequence pressureStress created by irreversibility, unclear risk, time pressure, or asymmetric choices.

Operating target Keep necessary complexity visible and well-structured. Reduce the remaining terms. Do not claim to calculate a person’s cognitive load from a static UI score.

Red-team findings

Corrections applied before formalizing the guide

The initial synthesis had a strong doctrine and useful patterns. It also carried several risks that would weaken enterprise adoption if left unresolved.

01

The research output was itself cognitively expensive.

A long linear report asked every reader to process theory, standards, implementation, and governance in the same sequence.

Correction: Use role lenses, layered detail, search, summaries, pattern cards, and an executable audit workflow.

02

Cognitive Load Theory can be over-transferred from learning research to product interaction.

Intrinsic, extraneous, and germane load are useful analytical lenses, but the framework originated in instructional design and should not be treated as a direct interface law.

Correction: Use the model to form hypotheses. Validate actual products through task, comprehension, workload, and recovery evidence.

03

Working-memory findings can become arbitrary item limits.

The “three to five chunks” finding depends on grouping, expertise, task, distraction, and what counts as a chunk.

Correction: Ban universal menu, card, and control count rules. Evaluate uncertain decisions and memory dependencies instead.

04

Progressive disclosure can become aesthetic concealment.

Hiding detail may make a screenshot cleaner while increasing search, recall, comparison cost, or manipulation risk.

Correction: Apply an essential-content test before choosing a disclosure surface.

05

A checklist can create false confidence.

Static conformance does not prove that users understand a concept, predict an outcome, or recover from an interruption.

Correction: Mark unresolved items as “requires user evidence” and treat critical failures as release blockers.

06

Optimizing for novices can penalize experts.

Expert users may require dense comparison, batch action, shortcuts, persistent filters, and simultaneous context.

Correction: Design adaptive complexity through stable defaults, optional expert density, saved views, and role-appropriate paths.

07

Accessibility cannot be reduced to WCAG conformance.

Technically conformant interfaces may still create excessive memory, language, attention, or executive-function demands.

Correction: Combine WCAG 2.2, W3C COGA patterns, semantic contracts, assistive-technology testing, and representative user research.

08

Engagement metrics can reward cognitive friction.

Time on site, interaction count, repeated visits, and notification opens may increase when users are confused or compelled.

Correction: Measure correct completion, confidence, unnecessary actions, recovery cost, and informed reversals.

09

Ethical design requires outcome tests, not intent claims.

A team may not intend manipulation while still producing asymmetric friction, hidden material information, or coercive defaults.

Correction: Audit effect, salience, timing, reversibility, and choice parity.

10

AI interfaces introduce a new layer of uncertainty.

Users must understand what the system knows, what it inferred, what tools it used, and what actions remain reversible.

Correction: Add an agentic UX contract for plans, evidence, uncertainty, execution previews, checkpoints, and audit history.

Human comprehension model

A repeatable path from perception to learning

A usable product supports a recurring sequence. Failures at any stage create downstream uncertainty and rework.

1

Perceive

Notice the relevant information, state, or control.

2

Orient

Understand location, scope, and relationship to the larger task.

3

Understand

Translate labels, content, and system state into meaning.

4

Choose

Compare relevant options and predict consequences.

5

Act

Execute the intended action through an understandable control.

6

Verify

Confirm what the system received, changed, or completed.

7

Recover

Correct, undo, resume, or safely exit when conditions change.

8

Learn

Build a stable mental model that reduces future effort.

Recognition

Keep context visible.

Show prior input, active filters, history, available actions, constraints, and current status instead of requiring recall.

Predictability

Make behavior consistent.

Use stable placement, familiar controls, explicit labels, visible focus, and state changes that do not surprise users.

Recovery

Lower the cost of exploration.

Use drafts, undo, preview, confirmation, version history, and persistent progress to reduce anxiety and suppression of action.

Cognitive variability is normal, not an edge case

Attention, memory, processing speed, language, executive function, and confidence vary across people and within the same person. Stress, fatigue, time pressure, unfamiliarity, device constraints, poor connectivity, and interruption create temporary cognitive limitations.

The standard therefore evaluates products under focused and interrupted conditions, across new and experienced users, with mobile and keyboard interaction, and with representative cognitive-accessibility participation where the product consequence warrants it.

Operating principles

Twelve rules for product, content, design, engineering, and governance

These principles are durable. Component choices remain contextual.

01

Make intent visible.

Every surface communicates its purpose, current state, primary outcome, and next action.

02

Structure by user understanding.

Navigation and grouping reflect user tasks, objects, vocabulary, and lifecycle rather than internal ownership.

03

Put meaning before detail.

Lead with result, implication, action, explanation, then evidence and audit history.

04

Reveal complexity by relevance.

Defer advanced or conditional content, but never hide information required for informed choice.

05

Prefer recognition over recall.

Keep options, status, constraints, prior input, and history visible or immediately retrievable.

06

Keep causality local.

Place controls, consequences, feedback, errors, and recovery near the object they affect.

07

Preserve context.

Maintain state across navigation, interruption, authentication, errors, sessions, and devices.

08

Support reversible exploration.

Use undo, preview, drafts, version history, and non-destructive defaults.

09

Match interruption to urgency.

Use blocking surfaces only when immediate attention is necessary for safety, correctness, or a consequential decision.

10

Preserve autonomy.

Provide transparent defaults, visible consequences, equal clarity, and reasonable friction for comparable alternatives.

11

Design for cognitive variability.

Support different expertise, language, attention, memory, input, vision, motion, and environmental conditions.

12

Prove comprehension.

Do not stop at component conformance. Verify that users understand, predict, act, confirm, and recover.

Structure and hierarchy

Information architecture and page composition as cognitive infrastructure

Visual hierarchy routes attention. Communication hierarchy establishes meaning. Information architecture establishes location. Together they expose the product’s conceptual model.

Recommended page anatomy

Page identity and purposeWhere am I, and why does this page exist?

Current state or key resultWhat is true now, and why does it matter?

Primary decision or actionWhat should I do next, and what will happen?

Main information groupsTask-relevant content organized by relationship.

Secondary actions and detailUseful but not dominant.

Evidence and explanationWhy the result or recommendation is credible.

Metadata, provenance, and historyOwnership, source, method, confidence, and change record.

Recovery and related pathsUndo, support, alternate route, and next destination.

Chunking contract

A valid chunk has one coherent purpose, a descriptive heading, clear internal relationships, separation from unrelated information, and a recognizable role in the larger page.

House rule: use a container only when its boundary communicates relationship, state, interaction, or consequence.

Navigation contract

Use user vocabulary and destination-oriented labels.
Front-load differentiating words.
Keep page titles and navigation labels consistent.
Minimize uncertain decisions rather than clicks.
Preserve orientation for search, notification, and AI deep links.

Discover and validate information architecture

Stage	Methods	Primary question
Discover	Content inventory, support analysis, search logs, open card sorting, journey mapping, vocabulary research	How do users describe and group the domain?
Model	Task, object, lifecycle, status, and decision structures	Which conceptual model best supports the work?
Validate	Closed card sorting, tree testing, first-click testing, search-result testing	Can users predict where information lives?
Operate	Navigation analytics, backtracking, failed search, support contacts, deep-link recovery	Where does the structure break under real use?

Progressive disclosure

Control timing without concealing meaning

Progressive disclosure controls when and where complexity appears. It is not permission to hide content solely to make a screen look cleaner.

Keep visible

Decision-essential

Purpose, current state, costs, risks, comparison criteria, requirements, consequences, errors, and recovery.

Disclose contextually

Relevant later

Conditional fields, examples, brief explanations, object actions, supporting evidence, and just-in-time guidance.

Defer deeply

Advanced or reference

Rare configuration, exhaustive history, debugging detail, raw evidence, edge cases, and expert customization.

Disclosure decision tool

Evaluate the content’s role before selecting a surface. This tool produces a design recommendation, not a compliance result.

Required for the current decision
The user cannot choose safely or correctly without it. Contains material cost, risk, permission, privacy, or legal consequence Must remain visible for comparison
Users need to inspect it alongside alternatives. Explains an error or required correction Relevant only after a prior answer or state Advanced, infrequent, or expert-only The trigger can accurately summarize the hidden content Needs a durable URL, long reading, printing, or independent sharing

Start with visible, structured content

Select the attributes that apply. Essential or material information will remain visible.

Surface selection

Surface	Use when	Main cognitive risk	Contract
Inline detail	Short explanation is closely tied to the current content	Visual density	Keep concise and adjacent
Native disclosure	Independent secondary content can be summarized accurately	Discoverability and repeated opening	Use meaningful summary; preserve state when useful
Tabs	Parallel peer views share the same context	Hidden dependencies and cross-tab comparison	Do not split one dependent workflow across tabs
Drawer or sheet	Contextual tools require background reference	Split attention	Preserve focus, context, and a clear close path
Dialog	A focused decision or interruption must be resolved	Lost page context and focus errors	Use sparingly; label consequence; restore focus
Separate page	Content is deep, durable, linkable, printable, or independently meaningful	Navigation and reorientation	Provide parent context and return path
Multi-step flow	Decisions are ordered, conditional, or consequential	Forced sequencing	Show progress, preserve state, support review and correction
Tooltip	Brief, supplemental, nonessential clarification	Transient and difficult to discover	Never contain required instructions

Pattern library

Reusable solutions organized by the user’s comprehension lifecycle

A component defines mechanics. A pattern defines when several components should be combined to solve a human problem.

Find a pattern

Lifecycle stage

Purpose and state header

Orient · Page-level

Use when: A page represents an object, decision, workflow, or system state.
Contains: Page title, concise purpose, current state, primary consequence, and primary action.
Avoid: Marketing slogans, multiple competing CTAs, or status hidden below the fold.
Verify: A new user can explain the page and next step within seconds.

Context restoration

Orient · Navigation

Use when: Users arrive from search, notifications, shared links, AI answers, or interrupted sessions.
Contains: Parent context, active constraints, current location, saved position, and return path.
Avoid: Generic browser-back dependence or resetting filters silently.
Verify: Users can resume after a realistic interruption without starting over.

Summary → explanation → evidence

Understand · Communication

Use when: Content supports a decision, recommendation, analysis, or technical explanation.
Contains: Result, implication, action, rationale, evidence, confidence, and provenance.
Avoid: Beginning with methodology or making evidence compete with the conclusion.
Verify: Executives and implementers can enter at different layers without losing traceability.

Persistent labels and requirements

Understand · Forms

Use when: Users enter, select, or review information.
Contains: Visible label, requirement before input, format example where necessary, and reason for unusual requests.
Avoid: Placeholder-only labels or requirements revealed only after submission.
Verify: Labels remain understandable after values are entered and at 200% zoom.

Comparable options

Choose · Decision support

Use when: Users must select among plans, settings, paths, permissions, or strategies.
Contains: Meaningful distinctions, relevant attributes, recommendation rationale, and visible consequences.
Avoid: Choice dumping, false precision, or a visually dominant organizational preference without disclosure.
Verify: Users can explain why an option fits their goal.

Choice parity

Choose · Ethical interaction

Use when: Options have comparable consequences, including consent, notification, subscription, and privacy choices.
Contains: Equivalent clarity, salience, understandable language, and reasonable friction.
Avoid: Confirmshaming, faded rejection, repeated prompting, or difficult cancellation.
Verify: The preferred business outcome is not easier only because alternatives are concealed or obstructed.

Specific action language

Act · Controls

Use when: A control changes, creates, sends, deletes, schedules, purchases, or publishes something.
Contains: Verb plus object, with consequence text for high-risk actions.
Avoid: Submit, continue, okay, or generic icons when the result is not obvious.
Verify: Users can predict the immediate result before activation.

Conditional form path

Act · Workflow

Use when: Later questions genuinely depend on earlier answers.
Contains: Relevant questions only, preserved progress, visible requirements, and server-compatible progression.
Avoid: Large hidden branches, nested reveals, or JavaScript-only completion.
Verify: Keyboard and screen-reader users receive the new context and can recover from errors.

Consequential confirmation

Verify · Completion

Use when: An action creates an external, financial, legal, operational, or irreversible effect.
Contains: What changed, when, affected object, reference, next step, and recovery path.
Avoid: “Success” without a durable result or relying only on a disappearing toast.
Verify: Users can state what happened and locate the result later.

Right-sized feedback

Verify · System response

Use when: The system receives input, processes work, changes state, or requires intervention.
Contains: Local validation, adjacent state, persistent progress, contextual confirmation, or alert according to urgency.
Avoid: Using toasts, dialogs, alerts, and notifications interchangeably.
Verify: Feedback remains perceivable and does not interrupt unrelated work.

Error summary + local correction

Recover · Forms

Use when: A page or step contains one or more correctable errors.
Contains: Summary, linked error locations, local messages, preserved input, plain explanation, and correction guidance.
Avoid: Clearing fields, color-only errors, or generic “invalid input.”
Verify: Focus, title, and announcements make the problem discoverable without creating a trap.

Reversible destructive action

Recover · Safety

Use when: Users delete, overwrite, revoke, publish, or make a high-consequence change.
Contains: Preview, explicit object name, undo or recovery window, and audit history where possible.
Avoid: Routine confirmation fatigue for reversible low-risk actions.
Verify: Safeguards match consequence rather than organizational anxiety.

Expert density mode

Understand · Adaptive complexity

Use when: Experienced users need parallel comparison, batch work, or operational monitoring.
Contains: Stable compact layout, shortcuts, persistent filters, saved views, and visible state.
Avoid: Personalization that silently relocates controls or hides capabilities.
Verify: Density improves throughput without increasing errors or losing touch support.

Natural stopping points

Orient · Attention

Use when: Content supports retrieval, comparison, review, or completion rather than open-ended consumption.
Contains: Pagination, load-more boundaries, result count, filters, saved position, and clear completion.
Avoid: Infinite scroll that removes landmarks or prevents reliable return.
Verify: Users can stop, resume, cite, and revisit a specific location.

Anti-patterns

Cognitive and autonomy failures to detect before release

Define urgency tiers, consolidate routine updates, provide controls, and match interruption to the required response time.

AI and agentic UX

Make probabilistic reasoning and tool execution understandable

AI products add capability uncertainty, source uncertainty, and execution uncertainty. Conversational fluency must not conceal those differences.

Before execution

Make the plan inspectable.

State the interpreted goal, assumptions, information needed, tools or data sources, material side effects, and approval boundaries.

During execution

Expose meaningful progress.

Show current activity, dependencies, waiting states, partial results, and recoverable failures without narrating low-level noise.

After execution

Separate result from evidence.

Show outcome, changes made, uncertainty, sources, rejected alternatives, audit history, and undo or correction paths.

Agentic interaction contract

Risk	Required product behavior	Release evidence
Capability ambiguity	State what the system can and cannot do in the current context.	Users accurately predict whether the system will answer, recommend, draft, or execute.
Hidden inference	Separate user-provided facts, retrieved facts, assumptions, and generated conclusions.	Users can identify the basis of consequential recommendations.
Tool execution	Preview material actions and show tool status, scope, target, and result.	Users can detect and stop an incorrect action before commitment.
Uncertainty	Communicate bounded uncertainty and missing evidence without false precision.	Confidence wording calibrates appropriately in comprehension tests.
Long responses	Use conclusion-first hierarchy with optional evidence and raw detail.	Users can find the answer and source without reading the full transcript.
State loss	Maintain a visible task state, accepted decisions, pending questions, and completed actions.	Users can resume after interruption or model handoff.
Automation bias	Offer meaningful review, alternatives, and human checkpoints for high-consequence decisions.	Testing includes deliberate model error and user correction.
Accountability	Record human owner, AI contribution, source provenance, model/tool activity, and accepted changes.	Audit artifact can reconstruct the consequential workflow.

Cognitive Friction Review

Observable risk review rather than an invented cognitive-load score

Classify each dimension. Any critical failure blocks release. “Requires user evidence” is a valid and important result.

Goal clarity

Can users state what this surface is for and what successful completion means?

Orientation

Can users identify where they are, what is active, and how to return or resume?

Priority and hierarchy

Does visual emphasis match task importance rather than decoration or business preference?

Decision complexity

Are choices relevant, distinct, comparable, and supported by visible consequences?

Memory burden

Must users remember information, state, or constraints that could remain visible?

Language and terminology

Do labels use user vocabulary, concrete nouns and verbs, and understandable sentence structure?

Disclosure integrity

Is decision-essential information visible at the point of choice?

Feedback and system status

Can users verify what was received, changed, completed, delayed, or failed?

Recovery and reversibility

Can users correct, undo, resume, or safely exit without losing unrelated work?

Interruption and attention

Does the surface request attention only when urgency and consequence justify it?

Accessibility contract

Do semantics, keyboard behavior, focus, names, state, announcements, zoom, contrast, and motion match the visual experience?

Ethical parity

Are alternatives presented with comparable clarity, timing, salience, and reasonable friction?

Measurement

Measure understanding, performance, workload, recovery, and ethical outcomes

A fast task can still be misunderstood. A visually preferred design can still be harder to operate. Use converging evidence.

Performance

What happened?

Task success, time, errors, unnecessary actions, backtracking, abandonment, and recovery.

Comprehension

What did users understand?

Prediction, explanation, consequence recognition, source location, and mental-model accuracy.

Workload

How difficult did it feel?

Mental effort, temporal demand, confidence, frustration, and NASA-TLX where appropriate.

Ethics

Was choice autonomous?

Reversal, cancellation, repeated prompting, consent comprehension, and friction parity.

Comprehension tasks

Explain the page’s purpose.
Predict the result of the primary action.
Identify the recommended next step.
Explain why one option differs from another.

Locate supporting evidence.
Correct a deliberate error.
Resume after interruption.
Return to a previously viewed object or state.

Minimum testing matrix

Variable	Coverage
Expertise	New, occasional, and experienced users
Device and input	Narrow touch, desktop keyboard, zoomed layout, assistive technology
Attention	Focused and realistically interrupted conditions
Consequence	Low-risk routine tasks and high-risk decisions
Content state	Empty, typical, dense, loading, error, restricted, and stale
Language	Primary language, translated content, unfamiliar terminology, and plain-language review
Connectivity	Normal, slow, interrupted, and resumed
Cognitive accessibility	Representative participation when risk, scale, or public impact warrants it

Measurement caveats

Do not treat time on task as an independent quality metric. Faster can mean efficient, rushed, or misunderstood.
Do not use NASA-TLX as a diagnostic of interface quality. It measures subjective workload in a task context.
Do not infer cognition directly from clicks, dwell time, eye tracking, or biometrics without a validated research design.
Do not aggregate away severe accessibility or autonomy failures behind a favorable average.

Delivery system

Turn research into components, patterns, acceptance criteria, tests, and governance

The standard becomes effective only when it changes product decisions and release evidence.

Component level

Behavior contract

Native foundation, semantics, keyboard behavior, focus, state, motion, feedback, failure modes, and accessibility-tree expectations.

Pattern level

Usage contract

User need, when to use, when not to use, essential content, alternatives, ethical risks, and test scenarios.

Product level

Evidence contract

Friction review, task testing, comprehension evidence, accessibility validation, telemetry, ownership, and decision record.

Definition of done

Discipline	Required evidence
Product	User need, success outcome, consequence level, and ethical guardrail metrics
Content	Purpose, terminology, front-loaded meaning, decision-essential content, and localization review
Design	Hierarchy, orientation, disclosure rationale, state coverage, recovery, responsive behavior, and choice parity
Engineering	Semantic HTML, progressive enhancement, keyboard/focus behavior, live state, error handling, persistence, and failure recovery
Accessibility	WCAG 2.2 evaluation, accessibility-tree inspection, assistive-technology testing, zoom, contrast, motion, and cognitive-accessibility considerations
Research	Task, comprehension, interruption, confidence, workload, and representative-user evidence proportional to risk
Governance	Ownership, source provenance, AI contribution, assumptions, exceptions, open challenges, and review date

Machine-readable guidance object

{
  "guidanceId": "cognitive.progressive-disclosure.essential-content",
  "userNeed": "Understand consequences before making a choice",
  "cognitiveRisks": ["memory-burden", "decision-uncertainty", "hidden-dependency"],
  "principle": "Reveal complexity by relevance without hiding essential content",
  "whenToUse": ["advanced settings", "conditional sections", "supporting evidence"],
  "whenNotToUse": ["price", "risk", "required instructions", "material consequences"],
  "accessibilityContract": {
    "nativeFoundation": "details/summary where appropriate",
    "keyboard": true,
    "focusBehavior": "preserve",
    "stateExposed": true
  },
  "ethicalChecks": ["choice-symmetry", "consequence-visibility", "friction-parity"],
  "validation": ["task-success", "comprehension", "mental-effort", "keyboard-test"],
  "evidenceConfidence": "high",
  "owner": "Design System",
  "humanOwner": "Jesse Graupmann",
  "aiContribution": "Research synthesis and initial formalization",
  "lastReviewed": "2026-07-17"
}

Recommended rollout sequence

Add the twelve principles and friction dimensions to the canonical handbook.
Annotate high-use components with cognitive, accessibility, and ethical behavior contracts.
Create templates for forms, settings, dashboards, documentation, notifications, and AI actions.
Pilot the review on one consumer flow and one expert enterprise workflow.
Calibrate release blockers against observed failures rather than opinion.
Add automated checks only where a machine can evaluate the requirement reliably.
Record exceptions, evidence, and unresolved research gaps in the provenance graph.

Evidence and provenance

Primary standards, research, design systems, and regulatory guidance

Sources are classified by role. Standards and official guidance establish requirements or tested practice. Foundational research provides theory. UX guidance provides applied heuristics that still require product validation.

Research limitations

Cognitive-load constructs are difficult to isolate in real product use.
Much foundational research originates in instructional or laboratory contexts.
Design-system guidance reflects tested practice but is not universally transferable.
Cultural, language, age, disability, expertise, and consequence alter what is understandable.
Pattern effectiveness must be validated against the product’s users and task.

Provenance policy

Preserve original authors and source institutions.
Separate source claims from synthesis and inference.
Record human ownership and AI contribution.
Do not upgrade confidence without evidence.
Retain contradictions and open challenges.

Open regulation

Research graph and authorship model

{
  "@context": {
    "schema": "https://schema.org/",
    "prov": "http://www.w3.org/ns/prov#"
  },
  "@graph": [
    {
      "@id": "urn:cognitive-ux:agent:jesse-graupmann",
      "@type": ["prov:Person", "schema:Person"],
      "schema:name": "Jesse Graupmann",
      "prov:hadRole": ["research director", "author", "editor", "accountable owner"]
    },
    {
      "@id": "urn:cognitive-ux:agent:openai-gpt-5-6-thinking",
      "@type": ["prov:SoftwareAgent", "schema:SoftwareApplication"],
      "schema:name": "OpenAI ChatGPT",
      "schema:softwareVersion": "GPT-5.6 Thinking",
      "prov:actedOnBehalfOf": "urn:cognitive-ux:agent:jesse-graupmann"
    },
    {
      "@id": "urn:cognitive-ux:dataset:v1-0-0",
      "@type": ["prov:Entity", "schema:Dataset"],
      "schema:name": "Cognitive UX Research System",
      "schema:version": "1.1.0",
      "schema:dateModified": "2026-07-17",
      "prov:wasAttributedTo": "urn:cognitive-ux:agent:jesse-graupmann"
    }
  ]
}

Make complex products easier to understand.

Design the user’s path through meaning.

The four-layer communication architecture

Decision

Execution

Evidence

Audit

The product objective

The research output was itself cognitively expensive.

Cognitive Load Theory can be over-transferred from learning research to product interaction.

Working-memory findings can become arbitrary item limits.

Progressive disclosure can become aesthetic concealment.

A checklist can create false confidence.

Optimizing for novices can penalize experts.

Accessibility cannot be reduced to WCAG conformance.

Engagement metrics can reward cognitive friction.

Ethical design requires outcome tests, not intent claims.

AI interfaces introduce a new layer of uncertainty.

Perceive

Orient

Understand

Choose

Act

Verify

Recover

Learn

Keep context visible.

Make behavior consistent.

Lower the cost of exploration.

Make intent visible.

Structure by user understanding.

Put meaning before detail.

Reveal complexity by relevance.

Prefer recognition over recall.

Keep causality local.

Preserve context.

Support reversible exploration.

Match interruption to urgency.

Preserve autonomy.

Design for cognitive variability.

Prove comprehension.

Recommended page anatomy

Chunking contract

Navigation contract

Discover and validate information architecture

Decision-essential

Relevant later

Advanced or reference

Disclosure decision tool

Start with visible, structured content

Surface selection

Failure

Replacement

Failure

Replacement

Failure

Replacement

Failure

Replacement

Failure

Replacement

Failure

Replacement

Failure

Replacement

Failure

Replacement

Make the plan inspectable.

Expose meaningful progress.

Separate result from evidence.

Agentic interaction contract

Goal clarity

Orientation

Priority and hierarchy

Decision complexity

Memory burden

Language and terminology

Disclosure integrity

Feedback and system status

Recovery and reversibility