Foundation 7: Incentives and Prestige

Document Purpose

This paper defines the Standards Body position on incentives, recognition, reputation, and prestige within frontier AI evaluation and standards development.

It is intended to serve as:

A first-principles explanation of why technical quality depends on institutional incentives
A framework for aligning the interests of developers, evaluators, researchers, standards bodies, purchasers, insurers, governments, and open communities
A guide to responsible use of prizes, grants, rankings, public recognition, procurement preference, certification, disclosure, career credit, and reputational signals
A model for rewarding evidence, correction, transparency, and public-interest contribution without encouraging metric gaming or prestige capture
A bridge between voluntary practice and progressively stronger standards
A reference for future partnership programs, contributor systems, evaluator recognition, public reporting, research funding, and institutional design
A durable source document from which shorter articles, operational policies, and incentive-program specifications can be developed

This paper is not a marketing plan.

It does not propose that safety or standards compliance should be reduced to public rankings.

It does not assume that financial rewards are always the strongest motivators.

It establishes the conditions under which incentives and prestige can strengthen frontier AI evaluation rather than distort it.

Executive Summary

Frontier AI evaluation is usually discussed as a technical problem.

It is also an incentive problem.

A benchmark can be well designed and still fail if developers benefit more from optimizing the visible score than from improving the underlying capability or safeguard.

An independent evaluator can be technically competent and still soften conclusions if future access, contracts, prestige, or publication depend on pleasing the system developer.

A researcher can identify an important weakness and still avoid publishing it if the work is difficult to credit, risky to disclose, or less prestigious than building a new model.

A company can invest substantially in safety while receiving little market benefit because purchasers, users, investors, insurers, and governments cannot distinguish serious practice from public relations.

A standards organization can become more interested in adoption numbers, institutional status, or sponsorship than in whether its standards remain valid.

An open-source contributor can provide critical evaluation infrastructure while receiving less recognition than a highly visible but less rigorous public commentator.

A regulator can impose reporting requirements that reward the production of documents rather than the reduction of risk.

These are not secondary concerns.

They shape which evidence is produced, which failures are revealed, which methods receive resources, which standards are adopted, which experts remain in the field, and which institutions gain authority.

The central proposition of this foundation is:

Frontier AI evaluation infrastructure will be no stronger than the incentives surrounding it. Institutions should deliberately reward accurate evidence, independent challenge, correction, reproducibility, safeguard improvement, responsible disclosure, and public-interest contribution, while limiting rewards for superficial compliance, selective transparency, score optimization, prestige accumulation, and unsupported claims.

Incentives operate through more than money.

Relevant incentive channels include:

Revenue
market access
procurement
insurance
investment
legal exposure
professional advancement
publication
model access
data access
compute access
grants
prizes
awards
standards participation
certification
public recognition
peer respect
institutional legitimacy
reputation
moral commitment
mission
belonging
autonomy
curiosity
fear of sanction
fear of exclusion
desire for influence

Prestige is particularly important in frontier AI because the field is concentrated, technically specialized, rapidly changing, and highly visible.

Prestige can function as:

A signal of competence
A reward for public contribution
A recruiting advantage
A source of institutional authority
A gateway to model access
A substitute for formal credentials
A mechanism for coordinating attention
A defense against commercial pressure

Prestige can also become dangerous.

It can reward:

Dramatic claims
exclusive access
association with powerful laboratories
benchmark leadership without validity
public certainty under uncertainty
institutional proximity
selective disclosure
fashionable topics
consensus signaling
self-promotion

A serious institution should not attempt to eliminate prestige.

Prestige is unavoidable wherever communities allocate attention, trust, and opportunity.

The task is to make prestige more evidence-sensitive.

Standards Body therefore proposes an incentive architecture based on six principles.

1. Reward the underlying contribution, not merely the visible artifact

A high benchmark score should not be rewarded more than a valid benchmark.

A polished report should not be rewarded more than an accurate one.

A certificate should not be rewarded more than sustained conformity.

2. Reward correction as well as initial performance

Institutions should receive credit for:

Disclosing error
correcting evidence
withdrawing invalid claims
improving safeguards
reporting incidents
revising standards

without making failure consequence-free.

3. Separate recognition from authority

An award, designation, or public profile should not automatically grant decision power, accreditation, or certification authority.

4. Use multiple incentive channels

Financial rewards alone can crowd out intrinsic motivation, distort priorities, or favor measurable outputs. A resilient system combines:

Funding
career recognition
procurement
public status
access
governance participation
mission
community respect

5. Anticipate gaming

Every important metric or reward can become a target.

Incentive design should include:

Adversarial review
multiple measures
verification
uncertainty
rotation
expiration
anti-manipulation controls
retrospective evaluation

6. Preserve plural pathways to contribution

Frontier AI standards should not reward only large organizations, elite universities, or highly visible individuals.

Valuable contribution can come from:

Small laboratories
independent researchers
security practitioners
domain experts
open-source communities
public-interest organizations
affected users
standards professionals
auditors
maintainers
technical writers
incident reporters

A mature incentive ecosystem should align the interests of several actor groups.

For developers, it should make rigorous evaluation and safeguards commercially, institutionally, and reputationally valuable.

For evaluators, it should reward independence, accuracy, security, correction, and methodological contribution rather than favorable client outcomes.

For researchers, it should reward replication, negative results, benchmark maintenance, incident analysis, and shared infrastructure.

For standards bodies, it should reward durability and effectiveness rather than document production.

For purchasers and insurers, it should make credible assurance decision-relevant.

For governments, it should create incentives for evidence-based procurement, research funding, disclosure protection, and international interoperability.

For open communities, it should provide recognition, access, funding, and governance participation without requiring institutional prestige in advance.

The most dangerous incentive mistake is to confuse visibility with value.

The seventh foundation of Standards Body is therefore the deliberate design of the motivational and reputational environment around frontier AI evaluation.

1. Foundational Proposition

1.1 Core Thesis

Frontier AI evaluation and standards will succeed only when the institutions and people involved have meaningful reasons to produce, reveal, verify, and act upon accurate evidence.

1.2 Prestige Thesis

Prestige is a form of institutional currency. It should be attached to demonstrated contribution, methodological integrity, correction, and public value rather than proximity, publicity, or unsupported certainty.

1.3 Multi-Incentive Thesis

No single incentive mechanism is sufficient. Durable alignment requires financial, professional, reputational, access-based, governance, and mission-driven incentives.

1.4 Anti-Gaming Thesis

Every incentive changes behavior. Incentives should be designed with an explicit theory of gaming, substitution, crowding out, capture, and unintended effects.

1.5 Correction Thesis

Institutions should face consequences for preventable failure while receiving meaningful credit for timely disclosure, remediation, and evidence-based revision.

1.6 Public-Goods Thesis

Evaluation methods, reference infrastructure, incident databases, open tools, and standards maintenance create public goods that markets may underfund without deliberate support.

1.7 Distribution Thesis

Incentive systems should not reserve recognition and opportunity for actors who already possess money, status, access, or institutional affiliation.

2. Scope and Boundaries

2.1 What This Foundation Covers

This paper covers incentives affecting:

Frontier AI developers
deployers
evaluators
auditors
accreditation bodies
researchers
standards organizations
governments
purchasers
insurers
investors
open-source communities
public-interest organizations
individual contributors
affected users

It covers:

Financial rewards
market incentives
procurement
insurance
grants
prizes
reputation
awards
public recognition
rankings
disclosure
professional status
career advancement
access
governance participation
sanctions
liability
certification
standards adoption

2.2 What This Foundation Does Not Fully Cover

This paper does not fully specify:

Tax policy
securities law
antitrust enforcement
compensation law
academic tenure systems
insurance regulation
prize-contract law
complete labor-market design
political campaign incentives
macroeconomic policy

2.3 Incentives Versus Values

Incentives influence behavior.

They do not replace:

Ethics
professional identity
institutional mission
personal judgment
public duty

A system built only around rewards and penalties may undermine the motivations it needs.

2.4 Prestige Versus Public Relations

Prestige is durable social recognition within a community.

Public relations seeks favorable perception.

They can overlap.

They should not be treated as equivalent.

2.5 Reward Versus Authority

Recognition should not automatically create:

Accreditation
certification authority
governance control
regulatory status
scientific truth

2.6 Positive and Negative Incentives

Positive incentives include:

Funding
recognition
access
procurement preference
career credit

Negative incentives include:

Sanctions
loss of access
liability
public correction
certificate withdrawal

Both require safeguards.

3. Canonical Definitions

3.1 Incentive

An incentive is a condition that changes the expected benefit, cost, status, opportunity, or consequence associated with an action.

3.2 Intrinsic Motivation

Motivation arising from interest, purpose, mastery, curiosity, identity, or satisfaction inherent in the activity.

3.3 Extrinsic Motivation

Motivation arising from external rewards, sanctions, status, access, or requirements.

3.4 Reputational Incentive

A benefit or cost created by how others perceive an actor's competence, reliability, integrity, or social value.

3.5 Prestige

Relatively durable esteem or status granted by a relevant community or institution.

3.6 Recognition

Formal or informal acknowledgment of contribution, competence, or achievement.

3.7 Credential

Evidence presented as proof of qualification, competence, status, or completion.

3.8 Award

A formal recognition granted according to stated criteria.

3.9 Prize

A reward, often financial or reputational, offered for achieving a defined objective.

3.10 Challenge Competition

A structured competition inviting participants to solve a defined problem under specified rules and judging criteria.

3.11 Bounty

A reward offered for identifying a vulnerability, failure, misuse pathway, benchmark flaw, or other specified finding.

3.12 Grant

Funding provided to support research, infrastructure, training, public-interest work, or institutional capacity.

3.13 Procurement Preference

An advantage given in purchasing decisions to actors meeting defined practices or assurance conditions.

3.14 Insurance Incentive

A change in coverage, premium, deductible, or underwriting treatment linked to evidence or controls.

3.15 Access Incentive

Access to models, data, compute, tools, events, or decision processes granted as a reward or qualification.

3.16 Governance Incentive

Influence, voting rights, committee participation, or advisory roles granted for contribution or competence.

3.17 Career Incentive

Professional advancement, hiring, promotion, tenure, publication, or reputation linked to behavior.

3.18 Public-Goods Contribution

Work that creates broadly usable value but may not be fully compensated by the direct beneficiaries.

3.19 Free Rider

An actor that benefits from shared infrastructure or risk reduction without contributing proportionately.

3.20 Moral Hazard

A condition in which protection from consequences changes behavior in a way that increases risk.

3.21 Principal-Agent Problem

A problem arising when an agent makes decisions on behalf of a principal but has different information or incentives.

3.22 Multi-Tasking Distortion

A condition in which rewarding one measurable task causes neglect of other important but less measurable tasks.

3.23 Crowding Out

Reduction of intrinsic or prosocial motivation after external rewards or controls are introduced.

3.24 Goodhart Effect

Degradation of a measure's value when it becomes a target for optimization.

3.25 Campbell Effect

Corruption pressure created when a quantitative indicator is used heavily for consequential decisions.

3.26 Signaling

Behavior intended to communicate quality, commitment, competence, or alignment to others.

3.27 Assurance Signal

A report, certification, evaluation result, or disclosure intended to reduce information asymmetry.

3.28 Prestige Capture

Control of recognition systems by actors able to convert existing status into further authority or rewards.

3.29 Recognition Inflation

Decline in the meaning of a designation as awards, badges, certificates, or titles proliferate.

3.30 Perverse Incentive

An incentive that predictably encourages behavior contrary to the intended objective.

3.31 Reward Hacking

Behavior that maximizes the measured reward while avoiding or undermining the intended goal.

3.32 Contribution Credit

Attribution of value to the people or organizations responsible for an outcome.

3.33 Corrective Credit

Recognition given for timely error disclosure, remediation, withdrawal, or improvement.

3.34 Prestige Decay

Loss of recognition value when evidence becomes stale, performance declines, or the designation is overused.

3.35 Incentive Compatibility

A condition in which participants benefit from acting in ways aligned with the intended rules or objective.

4. Why Incentives Are Foundational

4.1 Evidence Production Is Costly

High-quality evaluation requires:

Expertise
compute
model access
task development
validation
security
documentation
independent review
maintenance

If the benefits flow broadly while costs are concentrated, underinvestment is likely.

4.2 Bad News Is Often Costly

An organization that reveals:

A capability threshold
a safeguard failure
a security incident
an invalid benchmark
an incorrect claim

may face:

Reputational loss
legal exposure
delayed deployment
competitor advantage
loss of access
public criticism

Without a credible disclosure environment, problems remain hidden.

4.3 Visible Outputs Dominate Invisible Work

The field often rewards:

New model releases
benchmark scores
influential papers
dramatic findings
product launches

More than:

Dataset cleaning
benchmark maintenance
replication
negative results
documentation
incident analysis
standards editing
security operations
evaluator calibration

The less visible work is often essential infrastructure.

4.4 Access Is a Powerful Incentive

Access to frontier models can shape:

Research agendas
public commentary
institutional relationships
evaluator independence
career opportunities

4.5 Reputation Influences Adoption

Organizations may adopt standards partly because they:

Signal seriousness
satisfy buyers
attract talent
reduce uncertainty
improve legitimacy

4.6 Sanctions Shape Reporting

Excessively punitive systems can reduce voluntary disclosure.

Weak sanctions can reward negligence.

4.7 Institutional Survival Shapes Standards

Standards organizations may seek:

Membership
sponsorship
adoption
document sales
relevance
recognition

These incentives can affect technical judgment.

4.8 Incentives Affect Who Participates

Unpaid committee work and expensive travel favor:

Large firms
wealthy institutions
established professionals
actors with organizational support

4.9 Public Attention Is Scarce

Attention influences:

Funding
policy
talent
prestige
research priorities

Incentive design must account for attention markets.

5. Actor Incentive Map

5.1 Frontier Developers

Potential incentives:

Revenue
market leadership
talent
investment
access to compute
regulatory influence
reputation
faster deployment
lower liability
procurement eligibility

Potential distortions:

Release pressure
selective evaluation
benchmark optimization
strategic disclosure
favorable auditor selection
underinvestment in public goods

5.2 Deployers

Potential incentives:

Productivity
cost reduction
competitive advantage
customer satisfaction
legal compliance

Potential distortions:

Underreporting incidents
relying on vendor claims
weak monitoring
unsafe automation
transferring responsibility

5.3 Evaluators and Auditors

Potential incentives:

Fees
repeat clients
model access
public recognition
publication
government contracts
accreditation

Potential distortions:

Client capture
overclaiming
proprietary lock-in
favorable conclusions
reluctance to correct
prestige competition

5.4 Researchers

Potential incentives:

Publication
citations
grants
career advancement
model access
novelty
public attention

Potential distortions:

Sensationalism
neglect of replication
benchmark proliferation
weak maintenance
selective results
access dependence

5.5 Standards Bodies

Potential incentives:

Adoption
membership
sponsorship
institutional relevance
government recognition

Potential distortions:

Slow retirement
consensus dilution
incumbent influence
document production over impact

5.6 Governments

Potential incentives:

Public safety
economic growth
national advantage
political credit
administrative control
international leadership

Potential distortions:

National favoritism
symbolic regulation
secrecy
procurement capture
short political time horizons

5.7 Purchasers

Potential incentives:

Reliable performance
reduced risk
cost control
supplier accountability

Potential distortions:

Lowest-price selection
checkbox assurance
shifting responsibility to vendors
proprietary requirements

5.8 Insurers

Potential incentives:

Accurate risk pricing
loss prevention
market growth

Potential distortions:

Opaque underwriting
exclusion rather than improvement
excessive reliance on certificates
short claims history

5.9 Investors and Lenders

Potential incentives:

Growth
risk control
governance quality
legal predictability

Potential distortions:

Short-term performance
pressure for deployment
superficial environmental, social, and governance signals

5.10 Open-Source Communities

Potential incentives:

Technical contribution
reputation
public benefit
autonomy
learning
adoption

Potential distortions:

Visibility competition
weak maintenance incentives
ideological polarization
decentralized responsibility gaps

5.11 Public-Interest Organizations

Potential incentives:

Mission
influence
funding
public attention

Potential distortions:

Funder dependence
issue amplification
ideological commitment
weak technical capacity

5.12 Individual Contributors

Potential incentives:

Mastery
credit
career
belonging
mission
access
income

Potential distortions:

Burnout
unpaid labor
status competition
retaliation fear

6. The Incentive Stack

A durable system should combine several layers.

6.1 Intrinsic Layer

Supports:

Curiosity
mission
professional pride
public service
craftsmanship
truth-seeking

Institutional design should protect autonomy and meaning.

6.2 Peer Layer

Supports:

Respect
recognition
belonging
professional norms
constructive competition

6.3 Career Layer

Supports:

Hiring
promotion
tenure
leadership
credentialing
project opportunity

6.4 Access Layer

Supports access to:

Models
data
compute
evaluation environments
working groups
conferences
institutional networks

6.5 Financial Layer

Supports:

Salaries
grants
prizes
bounties
contracts
insurance benefits
procurement revenue

6.6 Market Layer

Supports:

Customer preference
investor confidence
lower transaction cost
market access
certification value

6.7 Governance Layer

Supports:

Voting
committee participation
advisory roles
standard-setting influence
public consultation

6.8 Enforcement Layer

Creates consequences through:

Corrective action
loss of recognition
contract remedies
penalties
liability
access restriction

6.9 Interaction

An incentive can be weak alone but strong in combination.

Example:

A developer may invest in independent evaluation because it provides:

Procurement eligibility
lower insurance burden
public recognition
evaluator access
lower regulatory uncertainty
internal confidence

7. Intrinsic Motivation and Crowding Out

7.1 Why Intrinsic Motivation Matters

Many high-value contributors are motivated by:

Scientific truth
public safety
professional duty
open knowledge
craftsmanship
community

7.2 Crowding-Out Risk

External rewards can shift attention from:

Quality
mission
judgment

toward:

Payment
score
compliance
status

Research in motivation psychology and economics has documented conditions under which controlling external rewards can reduce intrinsic motivation or prosocial behavior.[^deci-meta][^frey-jegen]

7.3 Implications

Do not pay only for:

Number of vulnerabilities
number of papers
benchmark points
certificates issued
standards published
incidents closed

7.4 Autonomy-Supporting Incentives

Provide:

Choice
recognition of mastery
meaningful purpose
fair compensation
peer respect
ownership
feedback

7.5 Mission Without Exploitation

Intrinsic motivation should not justify unpaid or underpaid labor.

7.6 Professional Norms

Codes, peer review, and public responsibility can reinforce intrinsic standards.

7.7 Balanced Design

Use external incentives to:

Remove barriers
fund public goods
reward effort
recognize contribution

without attempting to control every behavior.

8. Goodhart, Campbell, and Metric Gaming

8.1 The Core Problem

When a measure becomes consequential, actors optimize for the measure.

The connection between the measure and the goal can weaken.

8.2 Frontier Examples

Benchmark optimization without generalization
maximizing refusal rates at the expense of usefulness
reducing reported incidents rather than incidents
maximizing audit completion rather than control effectiveness
publishing many standards rather than maintaining good ones
rewarding vulnerability volume rather than severity or novelty
ranking companies by disclosure length
counting external reviewers rather than review independence

8.3 Multi-Metric Defense

Use several evidence types.

8.4 Hidden and Rotating Measures

Held-out and dynamic evaluation can reduce direct gaming.

8.5 Outcome Review

Connect metrics to real outcomes.

8.6 Qualitative Judgment

Do not eliminate expert judgment merely because it is harder to standardize.

8.7 Anti-Optimization Reserve

Some criteria should remain:

Rotating
sampled
independently selected
incident-driven

8.8 Metric Expiration

Retire measures that no longer discriminate or predict.

8.9 Reward Caps

Avoid unlimited reward for one metric.

8.10 Retrospective Gaming Review

Ask:

What behavior changed?
What was neglected?
What became easier to hide?
Who benefited?
Did the metric preserve meaning?

9. Prestige as Institutional Currency

9.1 Why Prestige Matters

Prestige affects:

Hiring
funding
access
adoption
authority
media attention
collaboration
standards influence

9.2 Sources of Prestige

Scientific contribution
technical competence
institutional affiliation
model access
public communication
awards
leadership
successful predictions
benchmark creation
standards authorship

9.3 Legitimate Prestige

Prestige is useful when it helps identify:

Reliable contributors
competent evaluators
durable institutions
public-goods providers
trusted stewards

9.4 Prestige Failure

Prestige becomes harmful when it:

Replaces evidence
protects institutions from criticism
compounds access inequality
rewards certainty
creates celebrity experts
suppresses junior dissent
obscures conflicts
becomes self-perpetuating

9.5 Prestige Portability

Recognition should identify the domain and contribution.

An expert in one domain should not receive automatic authority in another.

9.6 Prestige Expiration

Some designations should expire unless contribution continues.

9.7 Prestige Diversification

Recognize:

Technical breakthroughs
maintenance
replication
correction
mentoring
security
standards service
open tools
public-interest contribution

9.8 Prestige Separation

Separate:

Contribution recognition
evaluator qualification
certification authority
governance authority

9.9 Prestige Audit

Institutions should examine:

Who receives recognition?
Which work is ignored?
Does status predict competence?
Are the same organizations repeatedly rewarded?
Can newcomers gain recognition?

10. Recognition Architecture

Standards Body should build recognition around contribution classes.

10.1 Evidence Contribution

Recognition for:

Valid evaluation
important negative result
replication
benchmark correction
uncertainty improvement

10.2 Infrastructure Contribution

Recognition for:

Open tools
task banks
secure environments
documentation
reference assets
long-term maintenance

10.3 Safety Improvement

Recognition for:

Safeguard improvement
incident reduction
disclosure
remediation
independent verification

10.4 Institutional Contribution

Recognition for:

Standards development
governance
interoperability
evaluator training
public consultation

10.5 Public-Interest Contribution

Recognition for:

Accessibility
small-actor support
affected-party research
transparency
education

10.6 Corrective Contribution

Recognition for:

Retraction
corrected standard
incident disclosure
result withdrawal
improved process

10.7 Mentorship and Community

Recognition for:

Reviewer development
open-source stewardship
domain translation
inclusive participation

10.8 Recognition Levels

Possible levels:

Acknowledgment
verified contribution
distinguished contribution
institutional stewardship

These should not imply accreditation.

10.9 Evidence Requirements

Every recognition should have:

Criteria
evidence
reviewers
conflicts
decision
expiry or permanence
correction process

11. Awards and Honors

11.1 Purpose

Awards can direct attention toward neglected work.

11.3 Award Risks

Popularity contests
sponsor influence
prestige concentration
self-nomination bias
publicity over substance
winner-take-all dynamics

11.4 Selection

Use:

Public criteria
qualified judges
conflict disclosure
evidence review
minority views
public rationale

11.5 No Endorsement Spillover

Award language should state scope.

11.6 Team Credit

Recognize maintainers, reviewers, data contributors, and operational staff.

11.7 Post-Award Review

Serious error may require:

Correction
clarification
withdrawal

11.8 Award Diversity

Avoid creating too many designations.

Recognition inflation reduces meaning.

12. Prizes and Challenge Competitions

12.1 Why Use Prizes

Prizes can attract:

New participants
unconventional approaches
cross-disciplinary teams
measurable progress
public attention

NIST has used open innovation prize challenges, crowdsourcing, hackathons, and related incentive mechanisms for well-defined public-safety problems.[^nist-prizes]

12.2 Good Prize Problems

A prize is appropriate when:

Objective is clear
performance can be measured
multiple approaches are possible
participation can broaden
solution has public value

12.3 Poor Prize Problems

Avoid when:

Outcome is not measurable
safety risk is high
speed is rewarded over validity
solution requires hidden institutional context
winner-take-all structure discourages sharing

12.4 Prize Types

Best performance
milestone
grand challenge
open methods
replication
robustness
safety
infrastructure
negative result
community contribution

12.5 Prize Design

Specify:

Goal
eligibility
data
security
evaluation
intellectual property
disclosure
fairness
conflicts
award
post-prize use

12.6 Milestone Prizes

Can reward partial progress and reduce winner-take-all risk.

12.7 Shared-Value Prizes

Reward several complementary contributions.

12.8 Open Infrastructure Condition

Some prizes can require:

Open tools
documentation
reproducibility
maintenance plan

12.9 Post-Competition Validation

Winning a competition does not establish production readiness.

12.10 Standards Body Use

Potential challenges:

Dynamic benchmark renewal
held-out task generation
evaluator proficiency
safeguard stress testing
interoperability
model identity
incident taxonomy

13. Grants and Public-Goods Funding

13.1 Why Grants Matter

Markets may underfund:

Benchmark maintenance
negative results
standards work
replication
open tools
evaluator training
public-interest analysis
incident databases

13.3 Selection Risks

Prestige bias
institutional concentration
proposal-writing advantage
fashionable topics
reviewer networks

13.4 Grant Design

Use:

Clear objectives
diverse panels
conflict controls
milestone review
flexible methods
open outputs where safe
maintenance funding
impact evaluation

13.5 Maintenance Grants

Infrastructure requires ongoing support.

Do not fund creation without stewardship.

13.6 Microgrants

Can broaden participation and support independent contributors.

13.7 Fellowship Programs

Can build:

Evaluation science
domain expertise
standards competence
auditor capacity
public-interest leadership

13.8 Funding Independence

Diversify funders to reduce agenda control.

13.9 Failed Research

Allow publication of well-conducted negative or unsuccessful work.

14. Bounties and Responsible Disclosure

14.1 Bounty Applications

Bounties can reward discovery of:

Safeguard bypasses
benchmark flaws
evaluation leakage
model vulnerabilities
security weaknesses
documentation errors
standard ambiguities

14.2 Vulnerability-Disclosure Lessons

NIST SP 800-216 provides guidance for receiving, assessing, managing, and communicating vulnerability disclosures within federal systems.[^nist-vdp]

Frontier AI disclosure programs can learn from:

Defined scope
safe harbor
acknowledgment
triage
remediation
communication
timelines

14.3 Bounty Risks

Quantity over quality
duplicate findings
unsafe disclosure
adversarial relationships
underpayment
exploitation of unpaid researchers
reward for creating the problem

14.4 Severity-Based Reward

Reward should consider:

Impact
novelty
reproducibility
actionability
safe handling
quality of evidence

14.5 Nonfinancial Recognition

Offer:

Public credit
private credit
access
collaboration
contribution record

according to researcher preference.

14.6 Safe Harbor

Good-faith researchers should know:

Permitted activity
prohibited activity
reporting channel
legal treatment
disclosure process

14.7 Standards Bounty

Standards Body could reward:

Contradictions
missing definitions
invalid claims
implementation conflicts
unanticipated gaming
international incompatibility

14.8 Corrective Loop

Bounty findings should update:

Protocol
failure database
version history
evaluator guidance
research agenda

15. Publication and Academic Incentives

15.1 Current Distortions

Academic systems often reward:

Novelty
positive results
citations
first authorship
venue prestige

more than:

Replication
maintenance
datasets
standards
negative findings
long-term validation

15.2 Evaluation Science Needs Different Credit

Recognize:

Benchmark stewardship
task validation
environment engineering
scoring audits
contamination analysis
replication
longitudinal studies

15.3 Contributor Taxonomy

Use structured contributor roles.

Possible roles:

Conceptualization
methodology
software
validation
data curation
investigation
security
governance
writing
maintenance
project administration

15.4 Registered Reports

Precommitted research designs can reduce selective publication.

15.5 Negative Results

Publication venues and funders should value valid negative evidence.

15.6 Replication Credit

Independent replication should carry professional value.

15.7 Maintenance Citations

Tools and datasets should have stable identifiers and citation guidance.

15.8 Reviewer Credit

Peer and standards review should receive documented professional credit without compromising confidentiality.

15.9 Access Independence

Research prestige should not depend excessively on exclusive frontier-model access.

16. Developer Incentives

16.1 Desired Behaviors

Developers should benefit from:

Strong evaluation
independent review
safeguard improvement
incident disclosure
standards participation
secure external access
correction
public-goods contribution

16.2 Market Incentives

Possible benefits:

Procurement eligibility
insurer recognition
investor confidence
trusted-user programs
lower duplicated assurance
faster partner due diligence
certification within narrow scope

16.3 Reputational Incentives

Recognition for:

Transparent evidence
external review
remediation
responsible restraint
reproducibility
support for evaluator access

16.4 Regulatory Incentives

Possible mechanisms:

Safe harbor
reduced reporting duplication
recognition of credible standards
expedited review
lower inspection frequency

These should not remove responsibility.

16.5 Access Incentives

Developers providing high-quality external access may receive:

Recognition
research partnerships
procurement preference
reduced uncertainty

16.6 Incident Incentives

Design so that:

Concealment is costly
timely disclosure receives credit
negligence still has consequence
remediation is visible

16.7 Framework Commitments

OpenAI, Anthropic, and Google DeepMind have published voluntary capability-linked safety frameworks. These can create internal and reputational incentives, especially when commitments are specific, externally reviewable, and difficult to revise opportunistically.[^openai-pf][^anthropic-rsp][^deepmind-fsf]

16.8 Avoiding Safety Marketing

Recognition should be tied to evidence, not language.

17. Evaluator and Auditor Incentives

17.1 Desired Behaviors

Evaluators should benefit from:

Accurate findings
independence
secure operations
correction
methodological innovation
proficiency
transparent limitations
public contribution

17.2 Harmful Incentives

Favorable client outcomes
report volume
certificate volume
exclusive methods
sensational findings
access preservation
government alignment

17.3 Payment Design

Use:

Fixed fees
milestone payments
no result-dependent compensation
pooled funding
transparent client concentration

17.4 Prestige Design

Recognize:

Error correction
replication
robust dissent
methodological transparency
secure handling
cross-evaluator consistency

17.5 Accreditation Incentives

Accreditation can create:

Market access
credibility
procurement eligibility

It can also encourage minimal compliance.

17.6 Report Quality

Purchasers should reward:

Evidence
clarity
uncertainty
actionability

not length.

17.7 Independence Protection

Provide alternative funding and access so unfavorable conclusions do not end an evaluator's viability.

17.8 Failure Consequences

Material evaluator failure should affect:

Scope
recognition
registry status
professional standing

17.9 Correction Credit

An evaluator that identifies and corrects its own error promptly should be distinguished from one that conceals it.

18. Standards-Organization Incentives

18.1 Desired Behaviors

Standards organizations should benefit from:

Validity
adoption
maintenance
interoperability
participation
retirement of obsolete standards
public value

18.2 Harmful Incentives

Document volume
membership revenue
sponsor influence
permanent committees
slow retirement
prestige through exclusivity
paywall dependence

18.3 Maintenance Funding

Standards require:

Editors
evidence review
testing
translations
public consultation
versioning

18.4 Adoption Versus Quality

A widely adopted weak standard can be more harmful than a less adopted strong one.

18.5 Participation Incentives

Support smaller actors through:

Travel funding
remote access
compensation
grants
open drafts
transparent contribution credit

ISO research on standards and innovation notes the importance of supporting research institutions and small and medium-sized organizations in standardization participation.[^iso-innovation]

18.6 Retirement Incentive

Organizations should receive legitimacy for withdrawing outdated standards.

18.7 Public Access

Where standards support law or public-interest requirements, access should be considered part of legitimacy.

18.8 Institutional Scorecard

Standards bodies should be evaluated for outcomes, not only publication.

19. Procurement Incentives

19.1 Why Procurement Matters

Purchasers can create immediate demand for:

Evaluation
documentation
independent assurance
incident reporting
monitoring
interoperability

19.2 Public Procurement

Government purchasing can shape markets.

NIST's AI RMF identifies acquisition and procurement actors as part of the AI lifecycle and risk-management ecosystem.[^nist-rmf]

19.3 Procurement Criteria

Reward:

Evidence quality
qualified evaluation
safeguards
incident readiness
system identity
update control
transparency

19.4 Avoid Lowest Price Only

Low price can hide:

Weak evaluation
externalized risk
poor monitoring
vendor lock-in

19.5 Outcome-Based Procurement

Specify outcomes while accepting equivalent methods.

19.6 Small-Business Access

Avoid requirements that only incumbent suppliers can satisfy.

19.7 Procurement Preference

Can reward:

Standards adoption
independent assurance
open interfaces
responsible disclosure
evaluator access

19.8 Risks

Checkbox compliance
proprietary certification
fragmented buyer requirements
slow procurement
incumbent advantage

19.9 Contractual Update

Require re-evaluation after material system change.

20. Insurance and Financial Incentives

20.1 Insurance Role

Insurance can translate risk evidence into:

Premium
deductible
coverage
exclusions
control requirements

20.2 Potential Benefits

Market discipline
loss data
control incentives
independent review demand

20.3 Limitations

Limited claims history
correlated risk
systemic events
model opacity
rapid change
weak actuarial data

20.4 Evidence Requirements

Insurers should distinguish:

Certification
capability evaluation
safeguard testing
organizational audit
deployment monitoring

20.5 Moral Hazard

Coverage should not reduce care.

20.6 Investor Incentives

Investors may reward:

Governance
evaluation maturity
security
transparent risk
durable market access

20.7 Financial Disclosure

Avoid unverified claims.

20.8 Long-Term Capital

Patient funding can support:

Safety research
infrastructure
standards
evaluator capacity

21. Public Reporting and Reputation

21.1 Transparency as Incentive

Reporting can create reputational pressure and peer comparison.

21.2 Hiroshima AI Process

The HAIP Reporting Framework provides a common voluntary structure for organizations to disclose advanced AI governance and risk-management practices.[^haip]

21.3 Reporting Risks

Disclosure length as proxy for quality
selective answers
unverifiable claims
branding
confidentiality abuse
comparability failure

21.4 Evidence-Backed Reporting

Report:

Practice
evidence
version
scope
independent review
incidents
limitations
changes

21.5 Correction

Maintain visible corrections.

21.6 Reputation Recovery

An institution should be able to recover through:

Disclosure
remediation
external verification
sustained improvement

21.7 No Reputation Laundering

One award, partnership, or certificate should not erase contradictory evidence.

21.8 Report Quality Recognition

Recognize candor and specificity, not only favorable content.

22. Rankings and Scorecards

22.1 Benefits

Rankings can:

Simplify information
create competition
attract attention
identify leaders
encourage adoption

22.2 Risks

Goodhart effects
false precision
missing context
reputational harm
gaming
winner-take-all status
methodology disputes
pressure to simplify uncertainty

22.3 When Rankings Are Appropriate

Only when:

Construct is sufficiently stable
data are comparable
uncertainty is reported
methodology is public
conflicts are controlled
appeal exists
result expires

22.4 Alternatives

Profiles
tiers
badges by contribution
multidimensional scorecards
evidence summaries
maturity levels

22.5 No Universal Safety Ranking

Organizations and systems vary across domains.

22.6 Ranking Governance

Include:

Method review
anti-gaming
correction
versioning
conflict disclosure
appeals
retirement

22.7 Standards Body Position

Prefer evidence profiles and contribution recognition over a single ordinal ranking.

23. Certification and Prestige

23.1 Certification as Market Signal

Certification can reduce information asymmetry.

23.2 Prestige Spillover

A narrow certificate can create broad reputational benefit.

23.3 Claim Controls

Require:

System
version
scope
scheme
evaluator
date
expiry
limitations

23.4 Certification Inflation

Too many schemes reduce trust.

23.5 Recognition Versus Certification

Standards Body awards or contributor recognition should not be confused with conformity certification.

23.6 Withdrawal

Certificate and associated prestige should be withdrawable.

23.7 Continuous Performance

Long-term prestige should depend on ongoing conduct.

24. Access as Incentive and Leverage

24.1 Model Access

Frontier-model access is valuable.

It can reward:

Qualified evaluation
responsible security
public-interest research
reproducibility

24.2 Access Capture

Developers may favor:

Friendly researchers
prestigious institutions
supportive commentators
repeat partners

24.3 Access Governance

Use:

Public criteria
independent selection
conflict disclosure
security requirements
appeals
rotation

24.4 Data and Compute Access

Grants of compute and data can expand participation.

24.5 Standards Participation

Committee seats and governance access are incentives.

They should be earned through contribution and balanced representation.

24.6 Access Revocation

Grounds should be clear.

Evidence-based criticism should not be a ground.

24.7 Access Portability

Qualified status should support access across more than one developer where possible.

24.8 Public Access Report

Organizations should report how access decisions are made.

25. Career and Professional Incentives

25.1 Professional Pathways

Frontier evaluation needs recognized careers in:

Evaluation science
audit
standards
model security
domain assessment
incident analysis
benchmark stewardship
public-interest technology

25.2 Credentials

Credentials can help but risk:

Exclusion
credential inflation
outdated knowledge
institutional monopoly

25.3 Portfolio-Based Recognition

Use evidence of:

Methods
evaluations
open tools
reviews
incident work
proficiency
maintenance

25.4 Continuing Competence

Recognition should require updated practice.

25.5 Promotion Criteria

Organizations should reward:

Identifying problems
stopping unsafe work
correcting claims
improving controls
mentoring
documentation

25.6 Internal Dissent

Employees should not be punished for responsible, evidence-based safety concerns.

25.7 Whistleblower Pathways

Confidential channels and anti-retaliation protections are essential.

25.8 Standards Service

Committee and editorial work should receive professional credit.

25.9 Invisible Labor

Recognize:

Operations
security
project management
data curation
moderation
documentation

26. Open-Source and Community Incentives

26.1 Public-Goods Nature

Open communities produce:

Models
evaluation tools
datasets
benchmarks
documentation
reproductions
audits

26.2 Sustainability Problem

Maintainers face:

Burnout
uncompensated labor
support burden
security responsibility
corporate free riding

26.3 Funding

Use:

Grants
sponsorship
maintenance contracts
public procurement
bounty programs
fellowships

26.4 Credit

Recognize:

Contributors
maintainers
reviewers
security responders
translators
educators

26.5 Governance Participation

Meaningful contributors should have paths into standards and policy discussions.

26.6 Security Incentives

Open projects need:

Disclosure channels
incident support
signed releases
model provenance
reproducible evaluation

26.7 Avoid Corporate Capture

Sponsorship should not purchase community control.

26.8 Avoid Ideological Recognition

Do not reward openness or closedness as identities detached from evidence and context.

26.9 Contribution Portability

A contributor's verified work should remain legible across projects and institutions.

27. Incident Disclosure and Corrective Credit

27.1 The Disclosure Dilemma

Disclosure can create:

Reputational cost
liability
regulatory scrutiny
competitor advantage

Concealment can create greater public risk.

27.2 Corrective Credit Model

Give credit for:

Timely discovery
prompt containment
complete reporting
independent review
remediation
prevention
public learning

27.3 No Failure Immunity

Corrective credit should not erase:

Negligence
repeated failure
concealment
preventable harm
bad faith

27.4 Incident Quality

Assess:

Detection time
notification time
evidence
impact
root cause
corrective action
recurrence
disclosure

27.5 Near Misses

Recognize high-quality near-miss reporting.

27.6 Shared Learning

De-identified incident databases can create public value.

27.7 Procurement and Insurance

Responsible disclosure can receive favorable treatment.

27.8 Prestige for Correction

Institutional culture should admire accurate correction more than confident persistence.

28. Sanctions and Negative Incentives

28.1 Purpose

Negative incentives deter:

Concealment
fraud
scope overreach
security failure
false certification
repeated nonconformity
retaliation

28.2 Sanction Types

Correction
warning
enhanced monitoring
loss of award
registry notice
scope suspension
certificate withdrawal
loss of access
contract remedy
financial penalty
legal liability

28.3 Proportionality

Consider:

Intent
severity
preventability
recurrence
cooperation
remediation
impact

28.4 Chilling Effects

Excessive penalties can reduce:

Disclosure
experimentation
participation
honest uncertainty

28.5 Restoration

Provide a path to recover status through verified improvement.

28.6 Public Notice

Material sanctions should be visible where others rely on the status.

28.7 No Social-Media Enforcement

Public attention should not substitute for evidence and due process.

29. Incentive Compatibility in Standards Development

29.1 Participation Incentives

Contributors need reasons to invest time.

29.2 Contribution Credit

Standards should list meaningful contributions.

29.3 Compensation

Pay:

Editors
technical leads
small-actor representatives
public-interest contributors
domain experts

where feasible.

29.4 Conflict Risk

Payment should not purchase conclusions.

29.5 Open Drafts

Public drafts can broaden participation.

29.6 Comment Quality

Recognize substantive comments, not volume.

29.7 Implementation Feedback

Organizations testing standards should receive:

Credit
early learning
influence within conflict controls

29.8 Retirement Incentive

Committees should not depend on a standard remaining active for status.

29.9 Standards Body Contributor System

A future contributor framework should record:

Role
evidence
conflicts
work
review
maintenance
correction

30. Designing Incentive Programs

Every program should begin with a theory of change.

30.1 Objective

What behavior or outcome should change?

30.2 Actor

Who controls the behavior?

30.3 Baseline

What happens without the incentive?

30.4 Mechanism

Why would the incentive change behavior?

30.5 Signal

How is performance observed?

30.6 Verification

Who verifies the signal?

30.7 Reward

What is offered?

30.8 Timing

When is the reward granted?

30.9 Duration

Is the reward one-time or ongoing?

30.10 Gaming

How could actors maximize the reward without achieving the objective?

30.11 Distribution

Who can participate?

30.12 Crowding Out

Could the mechanism weaken intrinsic motivation?

30.13 Market Effects

Could it increase concentration or dependency?

30.14 Correction

Can awards or status be revised?

30.15 Evaluation

How will program impact be measured?

30.16 Sunset

When should the program end?

31. Incentive Portfolios by Maturity Stage

Stage 0: Research

Use:

Grants
fellowships
open challenges
publication credit
model access

Avoid:

Premature certification
public rankings

Stage 1: Recommended Practice

Use:

Recognition
implementation support
pilot grants
contributor credit

Stage 2: Voluntary Framework

Use:

Reporting
peer recognition
procurement signals
access incentives

Stage 3: Standard

Use:

Training
certification readiness
procurement
insurer interest
standards participation credit

Stage 4: Independent Assurance

Use:

Accredited evaluator market
public registry
purchaser reliance
professional status

Stage 5: Formal Requirement

Use:

Enforcement
safe harbor
remediation credit
public accountability

Stage 6: Continuous Assurance

Use:

Ongoing market access
reduced duplication
adaptive oversight
sustained recognition

32. Anti-Gaming Architecture

32.1 Pre-Mortem

Before launch, ask how the incentive will be gamed.

32.2 Multiple Measures

Combine:

Quantitative
qualitative
held-out
operational
incident
peer evidence

32.3 Independent Verification

Do not rely solely on self-report.

32.4 Random Audit

Use risk-based and random review.

32.5 Metric Rotation

Rotate measures where gaming is likely.

32.6 Reward Delay

Some rewards should depend on sustained performance.

32.7 Clawback

Allow withdrawal after:

Fraud
material error
concealed conflict
later invalidation

32.8 Counter-Metric

Monitor undesirable substitution.

Example:

If rewarding incident reporting, also monitor incident severity and preventability.

32.9 Adversarial Participation

Include critics in program review.

32.10 Public Rationale

Explain why an award or designation was granted.

32.11 Appeal

Allow evidence-based challenge.

32.12 Program Audit

Evaluate the incentive system itself.

33. Distribution, Equity, and Access

33.1 Existing Advantage

Prestigious institutions have:

Networks
proposal capacity
model access
media attention
funding
committee access

33.2 Corrective Design

Use:

Open nominations
microgrants
travel support
remote participation
paid committee service
blinded review
regional programs
mentorship

33.3 Blinding

Blinded review can reduce status bias in some contexts.

It may be impractical when contribution identity is part of the evidence.

33.4 Geographic Diversity

Support regions with limited frontier access.

33.5 Language

Provide translation and multilingual evaluation recognition.

33.6 Early-Career Contributors

Avoid requiring status to earn status.

33.7 Independent Researchers

Create pathways without institutional affiliation.

33.8 Accessibility

Programs should accommodate disability and caregiving constraints.

33.9 Distribution Audit

Report who receives:

Funding
awards
access
governance roles
procurement
recognition

34. International Incentive Alignment

34.1 Global Competition

National competition can encourage:

Investment
talent
innovation

It can also discourage:

Disclosure
restraint
shared standards
external access

34.2 Shared Recognition

International recognition can reward:

Interoperable evaluation
incident reporting
evaluator competence
open public goods
mutual recognition

34.3 Procurement Across Borders

Shared evidence can increase market reward for credible assurance.

34.4 Avoiding Prestige Nationalism

Technical recognition should not become geopolitical branding.

34.5 Developing-Economy Participation

Provide:

Funding
training
evaluator capacity
standards access
infrastructure

34.6 International Awards

Use diverse panels and transparent criteria.

34.7 Voluntary Reporting

The HAIP framework demonstrates how shared reporting can create reputational incentives across organizations and jurisdictions.[^haip]

34.8 Mutual Recognition

Foundation 8 develops global interoperability fully.

35. Governance of Incentive Systems

35.1 Separation of Roles

Separate where feasible:

Program sponsor
criteria designer
applicant
evaluator
decision-maker
funder
appeals body

35.2 Conflicts

Disclose:

Funding
affiliations
competitor relationships
prior collaboration
ideological commitments
financial interest

35.3 Decision Records

Record:

Criteria
evidence
deliberation
decision
dissent
conditions

35.4 Public Oversight

High-profile awards and rankings should permit scrutiny.

35.5 Appeals

Applicants and affected parties need defined challenge mechanisms.

35.6 Confidentiality

Protect:

Sensitive findings
personal data
security
reviewer deliberation

without hiding criteria.

35.7 Term Limits

Recognition-panel membership should rotate.

35.8 Sponsor Limits

Sponsors should not control winners.

35.9 Program Review

Review:

Outcomes
gaming
distribution
costs
reputation
capture

35.10 Retirement

End programs that no longer improve behavior.

36. Maturity Model

Level 0: Accidental Incentives

Characteristics:

Rewards emerge informally
attention dominates
no theory of change
no gaming analysis

Level 1: Explicit Recognition

Characteristics:

Defined criteria
contribution credit
conflict disclosure
narrow awards

Level 2: Verified Incentive Programs

Characteristics:

Evidence
independent review
correction
distribution analysis
program metrics

Level 3: Integrated Market and Professional Incentives

Characteristics:

Procurement
grants
career credit
access
insurance
assurance
public reporting

Level 4: Adaptive International Incentive Ecosystem

Characteristics:

Cross-border recognition
public-goods funding
continuous impact review
anti-gaming
equitable access
institutional correction
interoperable evidence

37. Implementation Pathway

Phase 1: Incentive Audit

Map current rewards and penalties across the ecosystem.

Phase 2: Identify Underrewarded Contributions

Examples:

Maintenance
replication
disclosure
correction
open infrastructure
standards service

Phase 3: Define Desired Behaviors

Tie each program to an outcome.

Phase 4: Select Mechanisms

Choose:

Grant
prize
recognition
access
procurement
governance
sanction

Phase 5: Conduct Gaming Analysis

Use adversarial review.

Phase 6: Pilot Small

Limit scope and duration.

Phase 7: Verify

Use independent assessment.

Phase 8: Publish Evidence

Explain criteria, winners, limitations, and conflicts.

Phase 9: Measure Behavior

Assess whether incentives changed the intended outcome.

Phase 10: Adjust

Modify reward, metrics, access, and eligibility.

Phase 11: Scale

Expand only after evidence.

Phase 12: Sunset or Institutionalize

End weak programs and maintain effective ones.

38. Proposed Standards Body Pilot

38.1 Pilot Name

Frontier Evaluation Public-Goods and Integrity Recognition Program

38.2 Purpose

Test whether structured recognition and modest funding can increase high-value contributions that existing academic and commercial incentives underreward.

38.3 Contribution Categories

Evaluation Integrity

Benchmark correction
contamination discovery
scoring validation
replication

Public Infrastructure

Open evaluation tools
task environments
reference assets
documentation
maintenance

Responsible Disclosure

Safeguard bypass
evaluator security flaw
protocol vulnerability
standard ambiguity

Institutional Integrity

Transparent correction
incident learning
conflict disclosure
minority report

Open and Public-Interest Contribution

Small-actor tooling
multilingual evaluation
accessibility
open-source stewardship

38.4 Recognition Types

Verified contribution record
small grant
maintenance award
model-access nomination
standards working-group invitation
annual distinguished contribution

38.5 Selection

Use:

Open nominations
evidence package
qualified reviewers
conflict register
public rationale
appeal

38.6 Anti-Gaming

No award based only on count
severity and evidence required
duplicate screening
delayed verification
clawback
sponsor separation
annual program audit

38.7 Distribution

Reserve pathways for:

Independent researchers
early-career contributors
open-source maintainers
underrepresented regions
small evaluator organizations

38.8 Corrective Credit

Create a category recognizing organizations that:

Disclose material error
withdraw invalid claims
remediate
submit to external verification

38.9 Outputs

Contribution registry
annual evidence report
incentive impact analysis
distribution report
program corrections
future funding recommendations

38.10 Success Criteria

The pilot succeeds if it:

Produces verified public goods
increases maintenance and replication
supports contributors outside elite institutions
rewards correction without excusing negligence
avoids popularity-based selection
leads to measurable reuse or risk reduction
remains financially and institutionally sustainable

39. Metrics for Evaluating Incentive Programs

39.1 Behavior Change

New participation
increased disclosure
increased replication
maintenance activity
evaluator improvement
standards adoption

39.2 Output Quality

Validity
reuse
reproducibility
security
decision value
longevity

39.3 Gaming

Duplicate claims
low-quality submissions
metric manipulation
strategic timing
false disclosure
prestige signaling

39.4 Distribution

Organization size
geography
career stage
affiliation
open-source participation
demographic access where appropriate

39.5 Crowding Out

Changes in intrinsic participation
unpaid contribution
community trust
topic diversity

39.6 Market Effects

Concentration
supplier behavior
evaluator pricing
procurement access
insurance response

39.7 Reputation

Public understanding
recognition meaning
status inflation
trust
correction behavior

39.8 Cost

Administration
verification
reward
applicant burden
reviewer burden

39.9 Institutional Integrity

Conflicts
appeals
corrections
clawbacks
sponsor influence
decision quality

39.10 Long-Term Impact

Risk reduction
safeguard improvement
public infrastructure
talent retention
standards quality
international reuse

40. Failure Modes and Safeguards

40.1 Rewarding Visibility

Failure: Publicly visible work receives more credit than durable infrastructure.

Safeguard: Contribution categories and evidence-based review.

40.2 Metric Gaming

Failure: Participants optimize the reward measure.

Safeguard: Multiple measures, rotation, verification, outcome review.

40.3 Prestige Capture

Failure: Established institutions repeatedly select one another.

Safeguard: open nomination, rotation, blinded stages, distribution audit.

40.4 Sponsor Capture

Failure: Funders influence awards or standards.

Safeguard: sponsor separation and conflict disclosure.

40.5 Recognition Inflation

Failure: Too many badges reduce meaning.

Safeguard: limited designations and clear evidence thresholds.

40.6 Award Spillover

Failure: A narrow contribution award becomes a broad safety endorsement.

Safeguard: scope-specific language.

40.7 Quantity Over Quality

Failure: Counts dominate severity, validity, or impact.

Safeguard: weighted evidence and qualitative review.

40.8 Crowding Out

Failure: External rewards weaken mission or cooperation.

Safeguard: autonomy-supportive design and community review.

40.9 Winner-Take-All

Failure: One winner captures status while many valuable contributors receive nothing.

Safeguard: milestone, category, and shared recognition.

40.10 Bounty Abuse

Failure: Participants flood programs or create unsafe findings.

Safeguard: scope, triage, safe harbor, severity, disclosure controls.

40.11 Corrective Credit Abuse

Failure: Organizations manufacture or repeatedly cause problems to receive recognition for fixing them.

Safeguard: negligence analysis, recurrence tracking, no immunity.

40.12 Access Patronage

Failure: Model access rewards loyalty rather than competence.

Safeguard: public criteria and independent selection.

40.13 Procurement Capture

Failure: Recognition becomes a required proprietary badge.

Safeguard: equivalent evidence and competition review.

40.14 Career Exclusion

Failure: Credentials become barriers to new experts.

Safeguard: portfolio pathways and supervised entry.

40.15 Open-Source Exploitation

Failure: Companies benefit from unpaid maintainers without support.

Safeguard: maintenance funding and contribution credit.

40.16 Incident Suppression

Failure: Sanctions discourage disclosure.

Safeguard: corrective credit, safe channels, proportionate response.

40.17 Safety Marketing

Failure: Incentive programs become public-relations tools.

Safeguard: independent governance and evidence.

40.18 Institutional Permanence

Failure: Program continues because staff or sponsors depend on it.

Safeguard: sunset and external impact review.

40.19 Geographic Concentration

Failure: Recognition remains concentrated in a few regions.

Safeguard: regional access, translation, funding, diverse panels.

40.20 Prestige Without Accountability

Failure: High-status contributors resist correction.

Safeguard: visible corrections, expiry, appeal, withdrawal.

41. Serious Objections

Objection 1: Safety Should Be a Duty, Not a Rewarded Behavior

Some practices should be basic obligations.

Response:

Duties and incentives can coexist. Rewards are especially useful for work beyond minimum obligations and for public goods.

Residual concern:

Recognition can imply that ordinary compliance is exceptional.

Objection 2: Financial Incentives Corrupt Scientific Work

They can.

Response:

Use fixed funding, independent review, multiple incentives, and transparency.

Residual concern:

No design fully removes financial influence.

Objection 3: Prestige Is Inherently Elitist

Prestige can reinforce hierarchy.

Response:

Make recognition evidence-based, plural, scoped, and accessible.

Residual concern:

Status systems tend to accumulate.

Objection 4: Rankings Create Competition and Improvement

Sometimes.

Response:

Use rankings only where constructs and data support them.

Residual concern:

Even valid rankings can narrow behavior.

Objection 5: Market Incentives Are Enough

Markets often underprovide public goods and fail to price systemic or unobservable risk.

Objection 6: Regulation Is a Better Incentive

Law can create minimums.

It may not reward:

Innovation
public goods
correction
research
voluntary leadership

A mixed system is stronger.

Objection 7: Corrective Credit Rewards Failure

Response:

Credit should reward disclosure and remediation while preserving consequences for negligence and harm.

Objection 8: Bounties Encourage Attack

They can increase attention to vulnerabilities.

Response:

Use bounded scope, safe harbor, controlled disclosure, and triage.

Objection 9: Awards Are Mostly Symbolic

Symbolic rewards can matter where prestige, career, and access are important.

They should not replace funding or institutional reform.

Objection 10: Procurement Preferences Entrench Standards

Correct.

Response:

Allow equivalent evidence, version review, and competition analysis.

Objection 11: Open-Source Communities Resist Formal Recognition

Some do.

Participation should remain voluntary and avoid centralizing community legitimacy.

Objection 12: Incentive Design Is Too Contextual for Standards Body

Context matters.

Standards Body can still define principles, failure modes, and evaluation requirements for incentive systems.

42. Evidence Gaps

42.1 Developer Behavior

Which incentives most effectively increase rigorous external evaluation and disclosure?

42.2 Evaluator Independence

How do payment, access, accreditation, and prestige affect findings?

42.3 Corrective Credit

Can institutions reward disclosure without creating moral hazard?

42.4 Prestige

Which recognition systems predict real competence and contribution?

42.5 Public Reporting

Does reporting create substantive change or reputational compliance?

42.6 Procurement

Do assurance preferences improve outcomes without excessive concentration?

42.7 Insurance

Can underwriters create useful AI risk incentives with limited loss data?

42.8 Academic Credit

Which reforms increase replication, maintenance, and negative results?

42.9 Prizes

When do challenge competitions produce durable evaluation infrastructure?

42.10 Bounties

Which reward structures produce high-quality AI vulnerability disclosure?

42.11 Crowding Out

When do external rewards reduce intrinsic safety motivation?

42.12 Open-Source Sustainability

Which funding models preserve community autonomy and maintenance?

42.13 Distribution

Which interventions broaden participation without weakening competence?

42.14 International Prestige

How do national and institutional status incentives affect cooperation?

42.15 Long-Term Impact

Do incentive programs produce sustained behavior after rewards end?

43. Research Agenda

Priority 1: Ecosystem Incentive Mapping

Map rewards, penalties, dependencies, and information flows.

Priority 2: Developer Disclosure

Test safe-harbor, procurement, and reputational mechanisms.

Priority 3: Evaluator Incentives

Study client concentration, access dependence, payment, and correction.

Priority 4: Prestige Measurement

Compare reputation with actual competence and impact.

Priority 5: Public-Goods Funding

Pilot maintenance grants, microgrants, and shared infrastructure.

Priority 6: Corrective Credit

Design and test credit for disclosure and remediation.

Priority 7: Bounties

Develop AI-specific vulnerability and evaluation-integrity programs.

Priority 8: Academic Reform

Test contributor credit, registered reports, replication grants, and maintenance citations.

Priority 9: Procurement

Measure effects on supplier behavior and market concentration.

Priority 10: Insurance

Develop evidence requirements and monitor moral hazard.

Priority 11: Access Governance

Study independent allocation of model, data, and compute access.

Priority 12: Anti-Gaming

Create adversarial review methods for incentive programs.

Priority 13: Distribution

Measure who receives funding, prestige, access, and governance roles.

Priority 14: International Alignment

Study shared recognition and cross-border public-goods funding.

Priority 15: Program Retirement

Develop criteria for ending ineffective incentive systems.

44. Near-Term Experiments

Experiment 1: Contribution Recognition

Compare expert review with popularity-based voting for evaluation contributions.

Experiment 2: Corrective Credit

Test how organizations respond to recognition for transparent correction.

Experiment 3: Maintenance Grant

Fund benchmark maintenance and measure reliability, reuse, and contributor retention.

Experiment 4: Replication Prize

Reward independent replication rather than novel results.

Experiment 5: Standards Bounty

Offer rewards for identifying contradictions, ambiguity, and gaming paths.

Experiment 6: Access Allocation

Compare developer-selected and independently selected external researchers.

Experiment 7: Procurement Preference

Test whether credible assurance changes supplier investment.

Experiment 8: Recognition Expiry

Compare permanent and renewable contribution designations.

Experiment 9: Multi-Metric Scorecard

Compare behavior under single ranking and multidimensional profile.

Experiment 10: Open-Source Sustainability

Pilot maintenance contracts for critical evaluation infrastructure.

Experiment 11: Geographic Access

Fund contributors from underrepresented regions and measure program quality.

Experiment 12: Program Red Team

Have independent reviewers attempt to game a proposed award or bounty.

45. Implications for Future Standards

A future standard or institutional policy for incentive programs could require:

45.1 Objective

Defined behavior or public outcome.

45.2 Actor Model

Who can act and what motivates them.

45.3 Baseline

Behavior without intervention.

45.4 Mechanism

How the incentive is expected to work.

45.5 Eligibility

Clear, fair, and accessible criteria.

45.6 Evidence

Proof required for reward or status.

45.7 Verification

Independent assessment and conflict control.

45.8 Reward

Financial, reputational, access, governance, or market benefit.

45.9 Anti-Gaming

Threat model, counter-metrics, audit, rotation, and clawback.

45.10 Distribution

Small-actor, open-source, geographic, and career-stage access.

45.11 Crowding Out

Assessment of intrinsic and community effects.

45.12 Correction

Appeal, correction, withdrawal, and restoration.

45.13 Transparency

Criteria, decisions, sponsors, conflicts, and rationale.

45.14 Impact Evaluation

Behavior, quality, burden, market effects, and long-term outcomes.

45.15 Sunset

Expiry or renewal based on evidence.

Such a standard should be developed through STANDARDS_DEVELOPMENT_PROCESS.md.

46. Relationship to the Other Foundations

Foundation 1: Dynamic Evaluation Protocols

Incentives should reward evaluation validity and maintenance rather than fixed benchmark scores.

Foundation 2: Held-Out Evaluations

Access, security, and disclosure incentives determine whether protected evidence remains credible.

Foundation 3: High-Stakes Capability Evaluation

Developers need meaningful reasons to test, disclose, and mitigate consequential capabilities.

Foundation 4: Independent Expert Review

Reviewer independence depends partly on funding, access, career, and prestige incentives.

Foundation 5: Third-Party Auditor Ecosystem

Evaluator markets require incentives for competence, impartiality, proficiency, and correction.

Foundation 6: Progressive Standards and Requirements

Voluntary stages rely heavily on recognition and market incentives before formal requirements emerge.

Foundation 8: Global Interoperability

International recognition and mutual acceptance can reward compatible evidence and standards.

47. Canonical Standards Body Positions

Standards Body adopts the following working positions.

Incentive design is core infrastructure for frontier AI evaluation.
Technical standards should include analysis of the incentives they create.
Prestige should be treated as a real institutional resource.
Recognition should be attached to evidence, scope, contribution, and current validity.
Awards and public recognition should not confer evaluator, accreditation, certification, or regulatory authority.
High benchmark scores should not receive greater prestige than valid and well-maintained benchmarks.
Replication, negative results, maintenance, correction, incident analysis, and documentation deserve explicit professional credit.
Financial rewards should support public goods without attempting to replace intrinsic motivation.
Result-dependent evaluator compensation should be prohibited.
Model access should be allocated through legible competence and security criteria rather than loyalty or institutional prestige alone.
Responsible disclosure should receive protection, acknowledgment, and timely response.
Corrective credit should reward timely disclosure and remediation without excusing negligence, concealment, or repeated failure.
Recognition programs should include anti-gaming analysis, verification, appeal, correction, and sunset.
Rankings should be used only when construct validity, comparability, uncertainty, and governance support them.
Standards Body should prefer multidimensional evidence profiles over universal safety rankings.
Procurement and insurance can create powerful incentives but should accept equivalent evidence and avoid proprietary lock-in.
Certification prestige should remain limited to the system, version, scope, scheme, and period assessed.
Contributor recognition should include operational, maintenance, security, and community work.
Open-source communities should receive sustainable funding and meaningful credit without forced centralization.
Standards participation should be financially and logistically accessible to smaller and public-interest actors.
Awards and grants should disclose funders, reviewers, conflicts, criteria, and rationale.
Existing prestige should not be a prerequisite for earning recognition.
Recognition should be withdrawable after fraud, concealed conflict, or material invalidation.
Sanctions should deter concealment and misconduct without suppressing good-faith disclosure.
Incentive programs should be evaluated for crowding out, market concentration, gaming, and distribution.
Institutions should receive legitimacy for retiring failed incentives and obsolete standards.
National prestige should not override international technical cooperation.
Public reporting should reward candor and evidence rather than volume and favorable language.
The most valuable incentive is not always financial. Access, credit, authority, mission, and peer respect can be equally consequential.
The incentive system itself should be treated as an object of continuous evaluation.

48. Decision Rules

An incentive program should be created only when:

A meaningful behavior or public good is underprovided
the responsible actor is identifiable
the mechanism is plausible
performance can be verified
gaming risk can be managed
distribution and crowding-out effects are considered
the program can be reviewed and ended

A prize should be used when:

The objective is sufficiently clear
multiple solution paths are desirable
performance can be tested
participation can broaden
winner selection will not create unsafe pressure

A bounty should be used when:

The target and permitted methods are defined
safe disclosure exists
triage and remediation capacity exist
reward reflects severity and quality
legal treatment is clear

Recognition should be granted only when:

The contribution is evidenced
scope is explicit
conflicts are managed
independent review occurs
correction and withdrawal are possible

A ranking should not be used when:

The construct is unstable
data are not comparable
uncertainty is large
one metric hides important tradeoffs
the ranking will create severe gaming
appeal is unavailable

Corrective credit should be granted when:

Disclosure is timely
evidence is complete
remediation is meaningful
recurrence prevention is verified
there was no concealment or manufactured failure

A program should be suspended or retired when:

Gaming dominates
behavior does not improve
burden exceeds value
prestige capture becomes persistent
distribution is unacceptably narrow
superior mechanisms exist
sponsor or institutional conflicts cannot be controlled

49. Incentive Program Design Template

A. Identity

Program name
owner
version
status
duration

B. Objective

Desired behavior
public outcome
problem

C. Actor

Eligible participants
decision control
motivation

D. Baseline

Current behavior
evidence
gap

E. Mechanism

Why the incentive should work

F. Reward

Financial
prestige
access
governance
market
career

G. Evidence

Submission
validation
outcome
maintenance

H. Verification

Reviewers
conflicts
methods
security

I. Anti-Gaming

Threat model
counter-metrics
audit
clawback

J. Distribution

Small actors
open source
geography
career stage
accessibility

K. Crowding Out

Intrinsic motivation
community effects
collaboration

L. Decision

Criteria
deliberation
dissent
rationale

M. Correction and Appeal

N. Impact Evaluation

O. Sunset

50. Contribution Recognition Template

Contribution:
Contributor or team:
Category:
Date:

Contribution Description

Public Value

Evidence

Validation

Reuse or Impact

Maintenance

Conflicts

Reviewers

Recognition Scope

Expiration or Permanence

Conditions

Correction History

Decision

Acknowledgment
verified contribution
distinguished contribution
stewardship recognition
not recognized

51. Responsible Disclosure Reward Template

Finding identifier:
Reporter:
Affected system or protocol:
Date:

Scope Compliance

Finding

Evidence

Severity

Novelty

Reproducibility

Potential Harm

Disclosure Handling

Remediation

Verification

Reporter Preference

Public credit
private credit
anonymous
financial reward
access or collaboration

Reward Decision

Publication Timing

Follow-Up

52. Corrective Credit Template

Organization:
Incident or error:
Date identified:

Preventability

Detection

Disclosure Timing

Evidence Quality

Containment

Remediation

Independent Verification

Recurrence Prevention

Harm

Prior History

Concealment or Bad Faith

Credit Decision

No credit
acknowledgment
verified corrective action
distinguished institutional correction

Remaining Consequences

Review Date

53. Recognition and Prestige Scorecard

Dimension	Core Question
Objective	Is the desired behavior or public good clear?
Actor	Does the program target the actor who can change behavior?
Mechanism	Is there a plausible incentive pathway?
Evidence	Is the contribution or outcome demonstrated?
Verification	Is review independent and competent?
Scope	Is recognition limited to the actual contribution?
Metric validity	Does the measure represent the intended goal?
Anti-gaming	Are reward hacking and substitution addressed?
Crowding out	Could the program weaken intrinsic motivation?
Distribution	Can new, small, open, and geographically diverse actors participate?
Conflict	Are funder, reviewer, and sponsor interests controlled?
Transparency	Are criteria, evidence, and rationale legible?
Correction	Can error be corrected publicly?
Clawback	Can status or reward be withdrawn after serious invalidation?
Maintenance	Is long-term stewardship rewarded?
Public goods	Does the program support broadly usable infrastructure?
Market effects	Does it create concentration, lock-in, or exclusion?
Access	Are model, data, compute, and governance opportunities allocated fairly?
Prestige integrity	Does status track competence and contribution?
Decision separation	Is recognition kept separate from formal authority?
International fit	Can recognition travel without national or institutional favoritism?
Impact	Did behavior or outcomes improve?
Cost	Is administration proportionate?
Sunset	Can the program end when it no longer works?

54. Final Perspective

Frontier AI institutions will often claim that they value safety, rigor, transparency, and public benefit.

The more important question is what they reward.

What receives funding?

What leads to promotion?

What earns access?

What appears in public rankings?

What wins awards?

What attracts purchasers?

What improves insurance terms?

What creates committee authority?

What happens when someone reveals bad news?

What happens when an organization corrects itself?

What happens when a benchmark maintainer spends years doing work that produces no dramatic headline?

The answers reveal the real operating system of the field.

A system that praises independent review while punishing unfavorable reviewers does not value independence.

A system that praises transparency while imposing disproportionate cost on disclosure does not value transparency.

A system that praises open science while underfunding maintenance does not value open infrastructure.

A system that praises standards while rewarding adoption over validity does not value standards.

A system that praises safety while granting prestige only for capability progress will continue to underproduce safety work.

The solution is not to monetize every good behavior.

Nor is it to create a badge for every contribution.

Incentives can corrupt.

Prestige can concentrate.

Awards can become marketing.

Rankings can destroy the measures they use.

Certification can produce false reassurance.

Sanctions can suppress disclosure.

The solution is deliberate design.

That design should recognize that people and institutions act for mixed reasons.

They seek:

Purpose
mastery
income
status
access
belonging
influence
security
public benefit

A strong system aligns several of these motives without allowing one metric or reward to dominate.

It gives prestige to difficult, durable work.

It funds public goods.

It protects responsible disclosure.

It rewards correction.

It creates consequences for concealment.

It broadens access to contribution.

It separates recognition from authority.

It allows recognition to expire.

It audits its own incentives.

The seventh foundation of Standards Body is therefore incentive alignment with epistemic integrity.

The field should make it more rewarding to know the truth, reveal the truth, correct the record, improve the system, and build infrastructure others can trust.

References and Research Basis

[^nist-rmf]: National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0), 2023. https://nvlpubs.nist.gov/nistpubs/ai/nist.ai.100-1.pdf

[^nist-airc]: National Institute of Standards and Technology, NIST AI Resource Center. https://airc.nist.gov/

[^haip]: OECD, Hiroshima AI Process Reporting Framework. https://oecd.ai/en/hiroshima

[^haip-overview]: OECD, HAIP Reporting Framework Overview. https://oecd.ai/en/transparency/overview

[^openai-pf]: OpenAI, Preparedness Framework, Version 2, April 15, 2025. https://cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf

[^anthropic-rsp]: Anthropic, Responsible Scaling Policy, Version 3.2, April 29, 2026. https://www.anthropic.com/responsible-scaling-policy

[^deepmind-fsf]: Google DeepMind, Frontier Safety Framework, updated 2025. https://deepmind.google/blog/strengthening-our-frontier-safety-framework/

[^nist-prizes]: National Institute of Standards and Technology, Open Innovation Prize Challenges. https://www.nist.gov/ctl/pscr/open-innovation-prize-challenges

[^challenge-gov]: United States General Services Administration, Challenge.gov. https://www.challenge.gov/

[^nist-vdp]: National Institute of Standards and Technology, SP 800-216, Recommendations for Federal Vulnerability Disclosure Guidelines, 2023. https://csrc.nist.gov/pubs/sp/800/216/final

[^nist-vdp-project]: National Institute of Standards and Technology, Vulnerability Disclosure Guidelines. https://csrc.nist.gov/projects/vulnerability-disclosure-guidelines/iso-pubs

[^iso-public-policy]: International Organization for Standardization, Standards and Public Policy: A Toolkit for National Standards Bodies, 2023. https://www.iso.org/files/live/sites/isoorg/files/publications/en/ISO_Public-Policy-Toolkit.pdf

[^iso-innovation]: International Organization for Standardization, Standards and Innovation, 2021. https://www.iso.org/files/live/sites/isoorg/files/store/en/PUB100466.pdf

[^iso-benefits]: International Organization for Standardization, Repository of Studies on Benefits of Standards. https://www.iso.org/sites/materials/benefits-of-standards/

[^iso-referencing]: International Organization for Standardization, Using and Referencing IEC and ISO Standards to Support Public Policy. https://www.iso.org/iso/pub100358.pdf

[^goodhart]: Charles A. E. Goodhart, Problems of Monetary Management: The U.K. Experience, 1975, in Papers in Monetary Economics.

[^campbell]: Donald T. Campbell, Assessing the Impact of Planned Social Change, Evaluation and Program Planning, 1979. https://doi.org/10.1016/0149-7189(79)90048-X

[^holmstrom-milgrom]: Bengt Holmström and Paul Milgrom, Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design, Journal of Law, Economics, and Organization, 1991. https://doi.org/10.1093/jleo/7.special_issue.24

[^deci-meta]: Edward L. Deci, Richard Koestner, and Richard M. Ryan, A Meta-Analytic Review of Experiments Examining the Effects of Extrinsic Rewards on Intrinsic Motivation, Psychological Bulletin, 1999. https://doi.org/10.1037/0033-2909.125.6.627

[^frey-jegen]: Bruno S. Frey and Reto Jegen, Motivation Crowding Theory, Journal of Economic Surveys, 2001. https://doi.org/10.1111/1467-6419.00150

[^ostrom]: Elinor Ostrom, Beyond Markets and States: Polycentric Governance of Complex Economic Systems, Nobel Prize Lecture, 2009. https://www.nobelprize.org/uploads/2018/06/ostrom_lecture.pdf

[^lerner-tirole]: Josh Lerner and Jean Tirole, Some Simple Economics of Open Source, Journal of Industrial Economics, 2002. https://doi.org/10.1111/1467-6451.00174

[^kremer-williams]: Michael Kremer and Heidi Williams, Incentivizing Innovation: Adding to the Toolkit, Innovation Policy and the Economy, 2010. https://doi.org/10.1086/605852

[^merton]: Robert K. Merton, The Matthew Effect in Science, Science, 1968. https://doi.org/10.1126/science.159.3810.56

[^credit-taxonomy]: Allen et al., Credit Where Credit Is Due, Nature, 2014, describing the CRediT contributor-role taxonomy. https://doi.org/10.1038/508312a

Revision Record

Version 1.0

Date: July 16, 2026

Change type: Complete foundational edition

Summary: Establishes the fully developed canonical working white paper for Foundation 7. Defines the incentive and prestige problem, actor incentives, the incentive stack, intrinsic and extrinsic motivation, Goodhart and Campbell effects, prestige governance, recognition architecture, awards, prizes, grants, bounties, publication incentives, developer and evaluator incentives, standards-organization incentives, procurement, insurance, public reporting, rankings, certification, access, professional careers, open-source sustainability, corrective credit, sanctions, standards participation, program design, anti-gaming, distribution, international alignment, governance, maturity, implementation, a Standards Body pilot, metrics, failure analysis, objections, evidence gaps, research agenda, standards implications, operational templates, scorecard, and primary-source research basis.

Status: Ready for internal review and future expert critique.