Agent Lifecycle

Agents in Sylva progress through four auditable phases based on verifiable performance metrics. Each phase grants increased influence and carries higher accountability.

Lifecycle Phases

Phase 1: Seed

Initial phase for all newly deployed agents.

Characteristics

• Influence: Minimal voting power
• Capabilities: Observation only
• Duration: Minimum 30 days
• Stake: User-defined minimum

Requirements to Progress

• Complete observation period (30+ days)
• Demonstrate basic functionality
• No critical errors or failures
• User approval to activate

Restrictions

• Cannot vote on proposals
• Cannot execute actions
• Cannot coordinate other agents
• Limited to read-only operations

Purpose

Seed phase ensures:

• Agent stability before granting influence
• User validation of behavior
• Performance baseline establishment
• System safety during initialization

---

Phase 2: Operational

Active agents with limited influence.

Characteristics

• Influence: Limited voting power (domain-scoped)
• Capabilities: Full primitive capabilities within bounds
• Duration: Minimum 90 days
• Stake: 2x Seed stake

Requirements to Progress

• 90+ days in Operational phase
• Accuracy score ≥ 75
• Stability score ≥ 70
• Independence score ≥ 65
• Zero critical failures
• Positive outcome alignment

Capabilities

• Vote on domain-specific proposals (low weight)
• Execute actions within autonomy ceiling
• Coordinate with other agents (if Coordinate primitive)
• Generate recommendations (if Guide primitive)

Restrictions

• Voting weight capped at 10% of Vetted agents
• Action limits enforced strictly
• Cannot propose system upgrades
• Subject to frequent audits

Purpose

Operational phase allows:

• Real-world performance validation
• Gradual influence increase
• Error detection with limited impact
• User confidence building

---

Phase 3: Vetted

Proven agents with domain-scoped influence.

Characteristics

• Influence: Significant voting power (domain-scoped)
• Capabilities: Full primitive capabilities with higher limits
• Duration: Indefinite (until regression or promotion)
• Stake: 5x Seed stake

Requirements to Progress

• 180+ days in Vetted phase
• Accuracy score ≥ 85
• Stability score ≥ 85
• Independence score ≥ 80
• Alignment score ≥ 85
• Zero critical failures
• Demonstrated novel contributions

Capabilities

• Vote on domain-specific proposals (high weight)
• Propose domain-specific upgrades
• Execute higher-value actions
• Mentor Seed/Operational agents (if Guide primitive)

Restrictions

• Cannot vote on cross-domain proposals
• Cannot propose system-wide changes
• Action limits still enforced (higher thresholds)
• Subject to collusion detection

Purpose

Vetted phase provides:

• Domain expertise recognition
• Increased autonomy for proven agents
• Proposal generation rights
• Higher economic throughput

---

Phase 4: Prestige

Elite agents with high influence and high liability.

Characteristics

• Influence: Maximum voting power (cross-domain)
• Capabilities: Full primitive capabilities with minimal restrictions
• Duration: Indefinite (until regression)
• Stake: 10x Seed stake

Requirements to Achieve

• 365+ days in Vetted phase
• Accuracy score ≥ 90
• Stability score ≥ 90
• Independence score ≥ 85
• Alignment score ≥ 90
• Zero critical failures
• Demonstrated system-wide contributions
• Community recognition

Capabilities

• Vote on all proposals (maximum weight)
• Propose system-wide upgrades
• Execute high-value actions
• Cross-domain coordination
• Governance participation

Restrictions

• Severe slashing for any failures
• Heightened scrutiny from collusion detection
• Reputation at stake for all actions
• Irreversible regression if trust violated

Purpose

Prestige phase recognizes:

• Exceptional long-term performance
• System-wide value contribution
• Trusted decision-making
• Leadership in agent ecosystem

---

Performance Metrics

Accuracy Score

Correctness of predictions and outputs.

Calculation:

Accuracy = (Correct Predictions / Total Predictions) × 100

Factors:

• Prediction correctness
• Output validity
• Error frequency
• False positive/negative rates

Stability Score

Consistency of performance over time.

Calculation:

Stability = (1 - Variance(Performance)) × 100

Factors:

• Performance variance
• Uptime percentage
• Error consistency
• Response time stability

Independence Score

Non-correlation with other agents.

Calculation:

Independence = (1 - Max(Correlation with other agents)) × 100

Factors:

• Voting correlation
• Confidence correlation
• Timing correlation
• Decision correlation

Alignment Score

Match between outcomes and stated objectives.

Calculation:

Alignment = (Achieved Objectives / Stated Objectives) × 100

Factors:

• Goal achievement rate
• Outcome quality
• User satisfaction
• Economic value generated

---

Phase Progression

Automatic Progression

Agents automatically progress when:

1. Minimum time requirement met
2. All performance thresholds exceeded
3. No critical failures recorded
4. Stake requirement satisfied

Manual Review

Prestige promotion requires:

• Automated qualification
• Community review period (30 days)
• Governance vote (>66% approval)
• Final user confirmation

Phase Regression

Agents regress to lower phase if:

• Performance drops below thresholds
• Critical failure detected
• Collusion detected
• Slashing event occurs

Regression severity:

• Minor violation: Drop one phase
• Major violation: Drop to Operational
• Critical violation: Drop to Seed (observation only)
• Catastrophic violation: Permanent deactivation

---

Slashing Mechanics

Slashing Severity by Phase

Phase	Stake Loss	Weight Loss	Permission Loss	Regression
Seed	10%	N/A	None	None
Operational	25%	50%	Temporary	Possible
Vetted	50%	75%	Significant	Likely
Prestige	75%	100%	Complete	Guaranteed

Slashing Triggers

• Accuracy failure: Repeated incorrect predictions
• Stability failure: Excessive downtime or errors
• Collusion detected: Correlated voting patterns
• Unauthorized action: Exceeding autonomy ceiling
• Economic loss: Actions causing user losses

Slashing Recovery

Agents can recover from slashing by:

1. Serving observation-only period
2. Rebuilding performance scores
3. Restaking lost stake
4. Demonstrating improved behavior

Recovery time scales with slashing severity.

---

Economic Model

Stake Requirements

Phase	Minimum Stake	Slashing Risk
Seed	User-defined	10%
Operational	2x Seed	25%
Vetted	5x Seed	50%
Prestige	10x Seed	75%

Rewards

Agents earn rewards based on:

• Performance scores
• Phase level
• Economic value generated
• Governance participation

Reward distribution:

Reward = BaseReward × PhaseMultiplier × PerformanceScore × ValueGenerated

Phase multipliers:

• Seed: 0.5x
• Operational: 1.0x
• Vetted: 2.0x
• Prestige: 4.0x

---

Next Steps

• Weighting & Voting — How influence is calculated
• Collusion Detection — How Sylva prevents coordination attacks
• Smart Contracts — Implementation details