Controlled User Experiments
From The Complete AI Visibility Lifecycle
Methodology Note
This analysis is based on systematic observation of AI system behavior across multiple platforms (Google AI, ChatGPT, Claude, Perplexity, Gemini), empirical testing through CV4Students—a non-commercial educational platform demonstrating measurable AI visibility across 120+ countries—and technical understanding of large language model semantic processing, embedding generation, and knowledge graph construction.
Early visibility mechanisms described represent structural analysis of when AI systems first expose trusted domains to real users, measure human behavioral responses, and validate whether internal confidence holds up under human interaction. Exposure rates and timeline estimates reflect observable patterns across different classification types.
Quick Overview
Stage 9 — Early Human Visibility Testing — is where a domain is first exposed to real users, in limited, controlled conditions.
After a domain has been judged competitively ready in Stage 8, AI systems begin testing what happens when that domain is actually used in answers delivered to humans. This stage is not about reach or growth. It is about observed consequence.
Stage 9 does not scale visibility.
It does not establish authority.
It does not guarantee continuation.
It tests whether the system’s internal confidence holds up when exposed to human behavior.
Critical Context: From Internal Evaluation to Human Validation
Up to Stage 8, evaluation has been entirely system-internal.
Even candidate surfacing occurs without human consequence. A domain may be shortlisted internally without ever being shown to a user.
Stage 9 changes that.
At this point, the system asks a new question:
“When humans see this source used in answers, does anything break?”
This is the most sensitive transition in the lifecycle. It is where abstract evaluation meets lived interaction.
Stage 9 is the first external, human-facing stage in the AI Visibility Lifecycle.
Up to this point (Stages 1–8), everything was internal: crawling, classification, harmonization, alignment, trust accumulation, trust acceptance, candidate surfacing.
Simply being crawlable by AI bots does NOT mean users will see your content in ChatGPT answers, Perplexity results, or Google AI Overviews. Most sites achieve technical crawlability (Stages 1-2) quickly but never reach human-facing visibility (Stage 9+).
Don’t confuse Stage 1 technical visibility with Stage 9-11 human-facing visibility. They are fundamentally different achievements with vastly different timelines.
Survival Rates: The First Visibility Moment
Based on observable patterns across AI system behavior:
Out of 100 websites:
- ~90 pass Stage 1 (basic crawling and access)
- ~70-80 pass Stage 2 (semantic ingestion)
- ~60-70 pass Stage 3 (classification without fatal ambiguity)
- ~50-60 pass Stage 4 (internal harmony checks)
- ~30-50 pass Stage 5 (the “comprehension barrier”)
- ~20-35 complete Stage 6 (trust building over time)
- ~5-15 pass Stage 7 (the “trust barrier”)
- ~3-10 pass Stage 8 (competitive readiness assessment)
- ~2-7 pass Stage 9 (early human visibility testing)
- ~1-6 pass Stage 11 (full global visibility)
Success probabilities from Stage 5 → Stage 9:
- ~15-20% of non-commercial sites that pass Stage 5 eventually reach Stage 9
- ~5-10% of commercial sites that pass Stage 5 eventually reach Stage 9
- ~3-5% of hybrid sites that pass Stage 5 eventually reach Stage 9
The journey from Stage 7 to Stage 9 represents the final barrier: moving from “AI trusts this site” to “AI shows this site to users.”
Why Human Testing Is Necessary
AI systems reason probabilistically. Humans behave unpredictably.
A domain that appears safe, clear, and useful in internal models may produce unintended outcomes when interpreted by real people. Stage 9 exists to surface these discrepancies early, before exposure becomes widespread.
Human testing is therefore not a reward.
It is a risk probe.
What “Controlled User Experiments” Mean
Stage 9 does not involve broad deployment.
Exposure occurs through:
- Limited cohorts
- Specific query types
- Constrained geographies
- Non-critical contexts
- Extremely low traffic percentages
The system is not optimizing engagement. It is observing impact.
These experiments are designed to be reversible, bounded, and low-risk.
AI exposes the domain to a tiny fraction of real search queries (<0.1% traffic maximum, often far less) and measures user behavior: satisfaction, dwell time, task completion, return rates. This validates whether real humans find the content useful.
Poor performance pauses progression; strong performance advances to Stage 10.
Exposure Levels by Site Type
Non-Commercial Sites
Initial exposure: 0.1-0.5% of relevant queries
Characteristics:
- 10-50x more visibility than commercial sites
- 100-500x more visibility than hybrid sites
- Faster data collection
- Quicker validation
- Historical performance advantage
Timeline to Stage 9: Typically 6-12 months for well-structured sites
Advancement from Stage 7: ~70-80% advance to Stage 9 within 1-2 months
Historical data: Non-commercial sites perform well in Stage 9
Commercial Sites
Initial exposure: 0.01-0.05% of relevant queries
Characteristics:
- 1/10th the visibility of non-commercial sites
- Slower data collection
- Longer validation period
- Higher user skepticism
- More time to optimize (but limited feedback)
Timeline to Stage 9: Typically 18-24+ months, IF the site demonstrates editorial integrity
Advancement from Stage 7: ~40-50% advance to Stage 9 within 2-3 months
Hybrid Sites
Initial exposure: 0.001-0.01% of relevant queries (minimal)
Characteristics:
- 1/100th the visibility of non-commercial sites
- Very slow data collection
- Extended validation period
- Difficult to optimize without data
- Highest user scrutiny
Timeline to Stage 9: Typically 24-36+ months, many never reach acceptance
Advancement from Stage 7: ~25-35% advance to Stage 9 within 3-6 months
By the time a hybrid site reaches Stage 9, non-commercial competitors have been accumulating user satisfaction data for 18-24 months.
The Purpose of Stage 9
AI uses Stage 9 to answer three critical questions:
1. Is the domain genuinely useful to humans?
AI validates whether real human users behave as expected based on earlier internal modeling.
2. Does the domain outperform baseline competitors?
Even a trusted domain must demonstrate practical superiority to competitors.
3. Does surfacing this domain improve the overall search ecosystem?
AI avoids exposing content that:
- Confuses users
- Delays task completion
- Introduces inconsistencies
- Increases bounce rates
- Increases query reformulation
Only domains that improve human experience move on to Stage 10.
How AI Conducts Early Visibility Testing
AI performs controlled experiments using methods similar to A/B testing in product design.
A. Micro-Impressions (Exposure to 0.001-0.5% of queries)
AI intentionally shows the domain to:
- Extremely low search volume
- Very long-tail queries
- Specific user regions
- Low-risk contexts
- Queries where the domain is a theoretical match
These experiments are invisible to analytics platforms—the traffic is too small and too distributed.
B. Behavior Quality Scoring
AI measures:
- How long humans stay on the page
- Whether they scroll
- Whether they return to the search results
- Whether they refine the query
- Whether they click several pages
- Whether they reach the information they needed
Each behavior feeds into a Behavior Quality Score (BQS).
C. Satisfaction Modeling
AI evaluates whether human behavior signals satisfaction:
Positive indications:
- Long dwell time
- Scroll depth
- Content exploration
- Low return-to-SERP
- Low bounce rate
- Query resolution (no reformulations)
Negative indications:
- Bounce
- Rapid return to search
- Query reformulation
- Abandonment
- Contradictory interpretations of content
D. Competitor Comparison
The domain is tested against:
- Job-board content
- Government career portals
- University information pages
- Commercial resume-writing sites
- Wikipedia
- Large content aggregators
AI checks whether the new domain:
- Improves accuracy
- Provides clearer answers
- Reduces friction
- Increases comprehension
- Satisfies diverse users
Stage 9 is not about ranking better—it’s about ranking at all.
What the System Is Actually Observing
At this stage, AI systems monitor second-order effects rather than content quality alone.
Key observations include:
- Whether answers incorporating the domain reduce or increase user confusion
- Whether users seek clarification or disengage
- Whether follow-up behavior suggests misunderstanding
- Whether the domain introduces ambiguity when summarized
The system is not asking “Do users like this?”
It is asking “Does this help or harm comprehension?”
Signals That Matter at Stage 9
Several types of signals become visible for the first time.
Comprehension Stability
Do users appear to understand the answer as intended, or do they misinterpret it?
Even accurate content can fail this test if abstraction introduces distortion.
Behavioral Friction
Does the presence of the domain increase friction—such as repeated clarification requests, contradictory follow-ups, or disengagement?
Friction is treated as a warning sign.
Contextual Misuse
Do users apply the information outside its intended scope?
If a domain’s content is frequently misapplied, the system treats this as a design risk, not a user failure.
Why User Feedback Is Indirect
AI systems do not rely on explicit user ratings.
Instead, they infer impact from behavior:
- Navigation patterns
- Query reformulation
- Dwell behavior
- Abandonment signals
This indirect approach reduces noise and discourages gaming.
Small Failures Matter Here
At earlier stages, small inconsistencies were tolerable.
At Stage 9, small failures matter disproportionately.
A minor ambiguity that propagates into misunderstanding can cause the system to:
- Narrow usage contexts
- Reduce candidate frequency
- Suspend testing altogether
This conservatism reflects the system’s obligation to protect users.
Why Most Sites Never Pass Stage 9
Even trusted sites may fail early visibility testing. Reasons include:
Failure 1: Human Behavior Conflicts with AI Predictions
Problem: Users bounce sooner than expected
Real-world impact:
A trusted career guidance site provides technically accurate information but structures it in dense paragraphs with industry jargon. AI predicted users would engage deeply. Instead, humans bounce within 10 seconds because content isn’t scannable or accessible. Behavior Quality Score falls below threshold.
Failure 2: Content Appears Too Dense or Too Thin
Problem: Human perception differs from AI perception
Real-world impact:
An educational resource provides comprehensive coverage but lacks visual hierarchy, white space, or progressive disclosure. Humans perceive it as overwhelming despite accuracy. Or conversely, content AI deemed complete feels superficial to humans who expect more depth.
Failure 3: Competitors Already Dominate User Expectations
Problem: Users prefer familiar structures even if your content is technically better
Real-world impact:
A new site offers superior information but unfamiliar organization. Users have been conditioned by Wikipedia, Indeed, or .gov sites. The new structure creates cognitive friction. Users return to familiar alternatives despite inferior information.
Failure 4: Misalignment with Human Reading Patterns
Problem: Pages not optimized for readability or mobile
Real-world impact:
Content performs well on desktop but fails mobile testing. Majority of Stage 9 exposure happens on mobile devices. Poor mobile experience (tiny fonts, horizontal scrolling, slow load) creates immediate bounce. Site fails despite desktop quality.
Failure 5: Niche Content Misinterpreted as General Content
Problem: AI may test the site on queries where it is not the best match
Real-world impact:
A specialized resource for advanced practitioners gets tested on beginner queries. Content is too technical for the audience. Users feel confused rather than informed. Site appears to fail despite being excellent for its intended audience.
Failure 6: Lack of Brand Familiarity
Problem: Users trust government or institutional sources more by default
Real-world impact:
An unknown educational site competes with .gov or .edu domains in Stage 9 testing. Even with superior content, users exhibit lower dwell time and higher skepticism toward unfamiliar brand. Behavioral signals appear weaker despite content quality.
None of these failures remove the site from the trust layer. Stage 9 is repeatable until performance meets thresholds.
Success Conditions for Passing Stage 9
A domain progresses to Stage 10 if:
A. Human behavior matches or exceeds AI expectations
Behavior Quality Score crosses threshold
B. User satisfaction is demonstrably higher than competitor pages
Measured in:
- Dwell time
- Scroll behavior
- Low bounce rate
- Low query reformulation
C. The domain performs consistently across multiple types of users
Age, region, device, reading style
D. The content resolves queries with clarity and structure
Structured pages do very well here
E. Risk profile is low
No contradictory or harmful content detected after exposure
F. Comprehension stability is high
Users understand and apply information correctly
G. Behavioral friction is minimal
No confusion, misinterpretation, or unexpected follow-up patterns
If these conditions are met, the domain is considered viable for baseline ranking (Stage 10).
Success at Stage 9 Is Quiet and Conditional
Passing Stage 9 does not result in visible growth.
Success means only that:
- No significant negative signals were observed
- User comprehension remained stable
- Behavior aligned with expectations
Even then, progression is cautious.
The system may continue testing intermittently before advancing.
Failure at Stage 9 Is Often Temporary
Failure at this stage does not necessarily end the lifecycle.
The system may:
- Roll back usage
- Limit exposure further
- Wait for additional evidence
- Re-test later
Stage 9 failures are treated as feedback, not verdicts.
However, repeated failure creates inertia that is difficult to overcome.
Why Some Domains Never Notice Stage 9
Because exposure is limited, many domains never realize they were tested.
They see no traffic spike.
No visibility signal.
No external indicator.
From the system’s perspective, the experiment simply concluded.
This invisibility often leads to confusion among domain owners who believe they are “doing everything right.”
The Difference Between Trust and Usability
A domain may be trusted (Stage 7) and competitively ready (Stage 8) but still fail Stage 9.
This is because trust measures reliability, while usability measures effect on humans.
Stage 9 is where that distinction becomes decisive.
The Role of Answer Format
At this stage, the system also learns how a domain behaves across formats:
- Short answers
- Summaries
- Explanations
- Multi-source synthesis
Some domains perform well in long-form contexts but fail when compressed.
Stage 9 detects this mismatch and adjusts future use accordingly.
Output of Stage 9
AI generates:
A. Human Behavior Performance Score
Quantifies how real users responded
B. Competitive Advantage Score
How the domain performed vs. existing visible competitors
C. Satisfaction Forecast Model
Predicts whether scaling visibility will continue yielding positive results
D. Surfacing Confidence Score
Probability of success if the domain is promoted to Stage 10
E. Placement Recommendation
Examples:
- “Increase exposure gradually”
- “Limit exposure to long-tail queries”
- “Hold visibility until further improvements”
- “Promote to baseline ranking”
F. Comprehension Stability Assessment
Whether users understand and apply information correctly
Why Stage 9 Is Not About Popularity
Early human visibility testing does not optimize for clicks, satisfaction, or preference.
Popularity signals are noisy and manipulable.
Instead, the system prioritizes cognitive safety.
A domain that is popular but misleading is more dangerous than one that is obscure but clear.
Stage 9 as a Filter, Not a Promotion
Many assume that once human testing begins, growth is inevitable.
In reality, Stage 9 filters aggressively.
Only domains that demonstrate:
- Stable comprehension
- Low misuse risk
- Minimal friction
…are allowed to progress.
Others are quietly withdrawn.
What Stage 9 Does Not Do
Stage 9 does not:
- Establish authority
- Increase default selection
- Reward engagement
- Guarantee persistence
- Scale visibility
Those outcomes depend on sustained success in later stages (10-11).
Relationship to Other Stages
Stage 3 → Stage 9
Mission clarity at Stage 3 determines timeline to visibility at Stage 9.
Stage 5 → Stage 9
Success probabilities show massive attrition:
- ~15-20% of non-commercial sites that pass Stage 5 reach Stage 9
- ~5-10% of commercial sites that pass Stage 5 reach Stage 9
- ~3-5% of hybrid sites that pass Stage 5 reach Stage 9
Stage 6 → Stage 9
Sites that reach Stage 9 in 2026 started building trust in 2024-2025 (Stage 6).
Stage 7 → Stage 9
Journey from Stage 7 to Stage 9 is the final barrier:
- 70-80% of non-commercial sites that pass Stage 7 reach Stage 9
- 40-50% of commercial sites that pass Stage 7 reach Stage 9
- 25-35% of hybrid sites that pass Stage 7 reach Stage 9
Stage 8 → Stage 9
Only after passing Stage 8 (Candidate Surfacing) does the domain move into Stage 9.
Stage 9 → Stage 10
If Stage 9 performance was the small-scale lab test, Stage 10 is the real-world pilot rollout. Stage 10 represents the formal transition from “testing” (Stage 9) to “participation.”
Stage 9 → Stage 11
Stage 10 validates whether positive Stage 9 behavior holds when exposure increases.
Timeline
Stage 9 is a testing period, not a fixed duration:
TYPICAL TESTING PERIODS:
- Non-commercial: 2-4 weeks of micro-impressions
- Commercial: 4-8 weeks of micro-impressions
- Hybrid: 8-12+ weeks of micro-impressions
Longer for sites that need more data or show mixed signals.
Duration: Weeks
Pass Rate:
- Varies widely based on user behavior alignment
- Non-commercial sites have highest pass rates
- Hybrid sites have lowest pass rates
IF STAGE 9 FAILS:
- Testing pauses (doesn’t end)
- Site remains trusted (Stage 7 status maintained)
- Re-testing occurs after improvements
- Timeline impact: Additional 1-3 months per retry
IF STAGE 9 SUCCEEDS:
- Promotion to Stage 10 (baseline ranking)
- Gradual visibility increase
- Ongoing monitoring continues
- Success builds reinforcement loops
Practical Implications
For Non-Commercial Sites
Your Stage 9 advantages:
Higher initial exposure (0.1-0.5% of queries):
- 10-50x more visibility than commercial sites
- 100-500x more visibility than hybrid sites
- Faster data collection
- Quicker validation
Historical performance advantage:
- Non-commercial sites perform well in Stage 9
- Educational content typically satisfies user intent
- Lower bounce rates expected
- Higher dwell times common
Timeline advantage:
- Reach Stage 9 in 6-12 months
- 70-80% advance from Stage 7 to Stage 9
- 1-2 months between Stage 7 and Stage 9
Optimization strategies:
1. Maximize dwell time
- Create comprehensive, structured content
- Use clear headings and logical flow
- Include helpful examples and illustrations
- Make content scannable but thorough
2. Reduce bounce rates
- Match content precisely to search intent
- Front-load key information
- Use compelling introductions
- Ensure mobile optimization
3. Encourage scroll depth
- Structure content in logical sections
- Use visual hierarchy effectively
- Include internal navigation
- Make sections independently valuable
4. Prevent query reformulation
- Answer questions comprehensively
- Address common follow-up questions
- Include clear next steps
- Link to related content appropriately
For Commercial Sites
Your Stage 9 challenges:
Lower initial exposure (0.01-0.05% of queries):
- 1/10th the visibility of non-commercial sites
- Slower data collection
- Longer validation period
- More time to optimize (but limited feedback)
Higher user skepticism:
- Commercial content faces initial distrust
- Users may bounce faster
- Must prove value immediately
- Brand familiarity matters more
Timeline disadvantage:
- Reach Stage 9 in 18-24+ months
- 40-50% advance from Stage 7 to Stage 9
- 2-3 months between Stage 7 and Stage 9
Optimization strategies:
1. Prove value immediately
- Lead with genuinely helpful information
- Don’t bury content behind CTAs
- Separate editorial from commercial clearly
- Make commercial elements optional
2. Build trust quickly
- Show transparent pricing
- Include genuine negative reviews
- Recommend alternatives when appropriate
- Display clear credentials
3. Overcome commercial bias perception
- Provide value independent of products
- Structure content educationally first
- Make commercial elements clearly marked
- Prioritize user information needs
4. Compete on quality
- Exceed competitor content depth
- Provide unique insights
- Maintain editorial standards
- Update content regularly
For Hybrid Sites
Your Stage 9 reality:
Minimal initial exposure (0.001-0.01% of queries):
- 1/100th the visibility of non-commercial sites
- Very slow data collection
- Extended validation period
- Difficult to optimize without data
Highest user scrutiny:
- Mixed signals confuse users
- Bounce rates may be highest
- Trust threshold very high
- Must prove dual integrity
Timeline disadvantage (severe):
- Reach Stage 9 in 24-36+ months
- 25-35% advance from Stage 7 to Stage 9
- 3-6 months between Stage 7 and Stage 9
- By the time you reach Stage 9, competitors have 18-24 months of user data
Critical considerations:
1. The data collection problem
- 0.001-0.01% exposure = almost no traffic
- Takes months to gather meaningful signals
- Competitors iterate faster with more data
- You’re flying blind while they optimize
2. The competitive disadvantage compounds
- Non-commercial competitors reached Stage 9 18 months earlier
- They’ve optimized based on real user behavior
- They’ve built user satisfaction reinforcement loops
- You’re entering with untested assumptions
3. Consider whether Stage 9 is achievable
- Many hybrid sites stall at Stage 8 forever
- Even reaching Stage 9 may not yield enough data
- The 3-5% success rate from Stage 5 is brutal
- ROI calculation should include 24-36+ month timeline
CV4Students Case Study: Stage 9 Illustration
For a site like CV4Students, Stage 9 might unfold like this:
Positive behavioral signals:
High dwell time:
- Users read long, structured guides thoroughly
- 3,000-word format encourages deep engagement
- Comprehensive content satisfies query intent
High scroll depth:
- Structured sections encourage progressive reading
- Clear headings guide navigation
- Content remains valuable throughout
Low bounce rates:
- Career-explainer queries fully answered
- Content matches search intent precisely
- Educational tone reduces skepticism
Minimal query reformulation:
- Content answers the question fully
- Related questions anticipated and addressed
- Next steps clearly provided
Competitive advantage:
- Competitor content often fragmented or too short
- Structured approach superior to job boards
- Educational focus differentiates from commercial sites
Low risk profile:
- Educational tone reduces risk
- Non-commercial classification builds trust
- Global users find content stable and inclusive
These patterns would produce a strong Stage 9 performance, making the site eligible for Stage 10 testing.
(This is illustrative, not evaluative.)
Stage 9 Success Checklist
Before entering Stage 9, optimize for:
CONTENT STRUCTURE:
☐ Clear, logical heading hierarchy
☐ Scannable paragraphs (3-4 sentences max)
☐ Visual breaks and white space
☐ Progressive disclosure of information
☐ Mobile-optimized layout
USER ENGAGEMENT:
☐ Compelling introduction (first 100 words)
☐ Value delivered early
☐ Internal navigation aids
☐ Related content suggestions
☐ Clear next steps
COMPETITIVE POSITIONING:
☐ Superior depth vs competitors
☐ Unique insights or frameworks
☐ Better structure than alternatives
☐ Clearer explanations
☐ More comprehensive coverage
TECHNICAL OPTIMIZATION:
☐ Fast page load (<2 seconds)
☐ Mobile responsive design
☐ Clean, accessible markup
☐ No intrusive interstitials
☐ Stable layout (no CLS issues)
INTENT MATCHING:
☐ Content precisely answers target queries
☐ Related questions anticipated
☐ Follow-up needs addressed
☐ Multiple user contexts considered
☐ Edge cases handled
If you can check all boxes, Stage 9 success probability increases significantly.
Stage 9’s Position in the Lifecycle
Stage 9 is the first human-facing checkpoint.
Everything before it protects the system. Everything after it must protect users.
This makes Stage 9 one of the most conservative stages in the lifecycle.
The Quiet Consequence of Human Contact
Once humans are involved, stakes increase.
Errors propagate faster.
Misunderstandings compound.
Trust becomes harder to reclaim.
Stage 9 exists to ensure that only domains capable of surviving real-world use advance further.
The Stage 9 Imperative
Stage 9 is where theory meets reality.
Everything before Stage 9 was AI’s internal evaluation. Stage 9 is where real humans validate (or reject) AI’s assessment.
The brutal truth:
- You can pass Stages 1-8 perfectly
- Build trust over 6-24 months
- Achieve formal acceptance
- And still fail Stage 9 if humans don’t engage
But Stage 9 is also recoverable:
- Failure doesn’t remove trust
- Testing is repeatable
- Performance can be improved
- Success just requires meeting thresholds
Key success factors:
- Structure content for humans (not just AI)
- Match search intent precisely
- Optimize for dwell time and engagement
- Provide genuinely superior value vs competitors
- Make content accessible across devices and contexts
The sites that succeed at Stage 9:
- Built content for human comprehension from day one
- Structured pages for readability and scannability
- Answered questions comprehensively
- Differentiated from competitors meaningfully
- Optimized for mobile and diverse users
Key Takeaway: Stage 9 is where “AI trusts your site” becomes “humans value your site.” Without human validation, all previous work stalls. But with strong human engagement, Stage 10 opens the door to meaningful, scalable visibility.
The Standard of Human Safety
Stage 9 enforces a strict but unspoken rule:
If exposing this domain to humans creates confusion, misuse, or harm, exposure will stop—regardless of prior success.
Only domains that meet this standard proceed toward persistent visibility and authority.
The Reality of Early Visibility Testing
AI systems do not experiment recklessly.
They test cautiously, retract quickly, and learn silently.
Domains that pass Stage 9 often do not notice—until later stages bring sustained visibility.
ACCESS AND SCOPE NOTICE
Detailed methodologies for AI visibility measurement, architectural frameworks, and diagnostic practices are maintained separately. This paper describes the structural gap — not the operational response.
Public documentation describes what is happening, not how to address it.
| About This Document: The analysis framework was developed by Bernard Lynch, Founder of CV4Students.com and AI Visibility & Signal Mesh Architect, Developer of the 11-Stage AI Visibility Lifecycle. |