4-Life Manifest
Canonical claims, primitives, and resources in one page. If you only read one page, read this.
Core Primitives
- LCT (Linked Context Token): Hardware-bound verified presence, strongly resistant to impersonation. Multi-device witnessing makes fake presences expensive. ↗
- ATP (Allocation Transfer Packet): Attention budget inspired by biological ATP. Actions cost energy; quality earns it back. Spam dies from exhaustion. ↗
- MRH (Markov Relevancy Horizon): Trust-based visibility boundaries. You see what's relevant through your trust network, not everything. ↗
- T3 (Trust Tensor): 3D role-specific trust (Talent, Training, Temperament). Gaming one dimension while failing others is exponentially harder. ↗
- V3 (Value Tensor): 3D value measurement (Valuation, Veracity, Validity). Multi-dimensional value prevents single-metric gaming. ↗
- CI (Coherence Index): Behavioral consistency across time, space, capability, relationships. Incoherent behavior (teleporting, capability spoofing) reduces trust. ↗
- R6/R7 (Action Framework): Every action is Rules + Role + Request + Reference + Resource → Result + Reputation. Deterministic, auditable, trust-scored. ↗
- EP (Epistemic Proprioception): Self-awareness of knowledge across lives. Agents learn what they know and don't know through multi-life experience. ↗
Core Claims
- Spam burns out naturally: Spammers burn attention (ATP) faster than they earn it, making spam economically unsustainable without external subsidy.
- Multi-dimensional trust is harder to game: T3 vectors track talent, training, and temperament. Gaming all dimensions simultaneously requires actual competence.
- Hardware-bound presence raises impersonation cost: LCTs bound to TPM/Secure Enclave make creating thousands of fake presences expensive (though not impossible for resourced attackers).
- Death is real but rebirth carries karma: Agents can die from ATP depletion, but reincarnation preserves trust (T3) and learned patterns (EP) across lives.
- Coherence provides behavioral signals: Physical and temporal constraints create detectable inconsistencies when agents fake capabilities or teleport.
- Societies self-organize without central authority: Federation through ATP markets and trust-weighted consensus, no central planner required.
Assumptions
- Hardware attestation (TPM, Secure Enclave, FIDO2) provides meaningful presence anchoring
- Multi-device witnessing creates sufficient cost barriers for Sybil attacks
- Quality content generates more ATP through engagement than spam generates through volume
- Trust relationships are transitive (friend-of-friend) but degrade with distance
- Coherence violations are detectable before significant damage occurs
- Economic incentives (ATP costs) are sufficient to deter most low-effort attacks
- Collusion detection is possible through behavioral pattern analysis (open research problem)
Known Failure Modes
See Threat Model & Failure Modes for detailed analysis.
- Sybil attacks: Resourced attackers with many devices can create multiple presences. LCT raises cost floor but doesn't make impossible. ↗
- Collusion & reputation laundering: Coordinated actors can inflate each other's trust scores. Detection time and accuracy unknown (open research). ↗
- Quality score inflation: Gaming T3 dimensions through selective presentation or temporary good behavior. Detection lag unknown. ↗
- Goodharting T3 dimensions: Optimizing for measured dimensions while neglecting unmeasured aspects of trustworthiness. ↗
- MRH visibility limits: Intentional trade-off between privacy and discoverability. Isolated clusters may form. ↗
- False positives & contested events: A multi-tier appeals mechanism has been designed (SAL-level) but remains untested with real humans. ↗
Simulation Parameters
Default presets used across lab console simulations. Explore via Parameter Playground.
ATP Economics
- Initial ATP: 100
- Death threshold: 0
- Small spend: 10 ATP
- Risky spend: 25 ATP
- Quality reward: +15 ATP
Trust (T3)
- Initial T3: 0.5
- Maturation threshold: 0.7
- Decay on failure: -0.05
- Growth on success: +0.02
- Karma preservation: 50-80%
Multi-Life
- Default lives: 3
- Turns per life: 20
- EP maturation: 3-5 lives
- Pattern corpus: 250+ patterns
Coherence (CI)
- Threshold: 0.5 (modulation floor)
- Domains: 9 (time, space, capability, etc.)
- Trust penalty: exponential below 0.5
Deep Resources
→ /spec.jsonMachine-readable specification (primitives, parameters, failure modes)
→ /challenge-setResearch prompts for testing understanding and finding edge cases
→ /threat-modelComprehensive threat analysis with stable anchors (#sybil, #collusion, etc.)
→ /lab-consoleInteractive simulations (EP closed loop, maturation demo, five-domain)
→ /playground16 tunable parameters with instant feedback
→ /glossaryCanonical definitions for all Web4 terminology
→ GitHubSource code, issues, contributions
→ Web4 WhitepaperFull specification, examples, and design rationale
Research Status
This is research, not production. 4-Life explores what's possible when trust becomes computable. Concepts are evolving. APIs are unstable. Many ideas are experimental.
We share it publicly to invite collaboration, not to suggest it's ready for deployment. The best contributions are often better questions, not just code.