Trust Tensors: Multi-Dimensional Trust
Web4 doesn't reduce trust to a single number. Instead, it uses Trust Tensors (T3) — three-dimensional vectors that capture Talent, Training, and Temperament, always in the context of a specific role.
You might trust a brilliant surgeon who's unreliable differently from a steady surgeon with less raw talent. You wouldn't trust either of them to fix your car. T3 makes these distinctions explicit, measurable, and role-specific.
The Problem
Traditional Trust: One Number, No Context
❌ One-dimensional scoring — “Trust score: 7/10” loses all nuance
❌ Context-blind — Same score for surgeon, mechanic, and babysitter
❌ Can't represent trade-offs — “Brilliant but unreliable” becomes just “average”
❌ Easy to game — Optimize for one metric, ignore everything else
Result: Trust scores become meaningless averages that hide critical information.
The Solution
Web4: Trust Tensors (T3) — Three Dimensions, Role-Specific
✅ Three canonical dimensions — Talent, Training, Temperament
✅ Role-contextual — Trust is always for a specific role, not a universal score
✅ Captures trade-offs — “High talent + low temperament” = measurable pattern
✅ Gaming is exponentially harder — Must build trust across all dimensions within each role
Result: Trust becomes a rich, context-aware signal that preserves nuance.
The Three Trust Dimensions
Every T3 tensor measures three aspects of capability within a specific role.
Talent
Can they solve problems in this role?
Natural aptitude and creativity within a specific domain. Novel solutions, insight, pattern recognition.
Training
Do they have the expertise for this role?
Learned skills, domain knowledge, and relevant experience. Grows through practice and study.
Temperament
Can they be relied on in this role?
Consistency, reliability, and ethical behavior within the role context. A surgeon needs steady hands; a trader needs risk tolerance.
Key insight: These dimensions are always measured within a role. Alice might have high Talent as a data analyst (0.85) but low Talent as a mechanic (0.20). Her trust as an analyst says nothing about her trust as a mechanic. Web4 never lets trust “leak” across unrelated domains.
Try It: Trust Tensor Simulator
Pick a role, then apply scenarios. Watch how the same action affects trust differently depending on which role you're evaluating.
Evaluating trust as:
Role emphasis: Insight from complex data — Talent 40% / Training 35% / Temperament 25%
Moderate Trust
Trust Dimensions
Choose a Scenario
Key Insights
Talent ≠ Temperament
A brilliant but unreliable surgeon is dangerous. An ethics violation tanks Temperament without touching Talent. The system captures these trade-offs that single scores bury.
Trust Is Role-Specific
Try switching roles in the simulator above. The same tensor scores produce different overall trust because each role weights dimensions differently. A leader needs Temperament; an analyst needs Talent.
Gaming Is Exponentially Harder
To game a 3D tensor, you must build trust across all dimensions within each role separately. You can't inflate Talent by being reliable, and you can't transfer trust between unrelated roles.
Recovery Is Dimensional
Lost Temperament trust? Consistent behavior rebuilds it, even while Talent stays the same. “Transparent mistake” shows how honesty can rebuild one dimension while acknowledging a gap in another.
V3: Measuring What You Produce
T3 measures who you are. V3 measures what you create. Every output in Web4 gets scored across three dimensions — and the weights are designed to reward truth over popularity.
Valuation
weight: 0.30How useful is the output? Measured by recipient satisfaction and ATP earned vs expected.
Veracity
weight: 0.35How truthful and accurate is it? Verified by external validation and witness attestation.
Validity
weight: 0.35Is it well-reasoned and actually delivered? Confirmed by receipt and logical soundness.
Notice: Veracity + Validity (0.70 combined) outweigh Valuation (0.30). Web4 rewards truth and rigor over popularity by design.
Score These Outputs
Click each output to see how V3 scores it.
Real Example: Same Person, Different Roles
Alice has spent years as a data analyst and recently started managing projects. Her T3 tensors reflect this asymmetry:
Alice as Data Analyst
Talent: 85% (creative problem-solver)
Training: 90% (years of deep experience)
Temperament: 95% (rock-solid reliability)
Role-weighted trust: 90% — she's deeply trusted in this domain.
Alice as Project Manager
Talent: 65% (developing leadership instincts)
Training: 70% (some PM experience)
Temperament: 91% (her reliability carries over naturally)
Role-weighted trust: 76% — trusted, but still growing into this role.
Alice as Mechanic
Talent: 20% (no mechanical aptitude)
Training: 15% (no relevant training)
Temperament: 50% (untested in this context)
Role-weighted trust: 27% — would you let her fix your brakes?
This is the power of role-specific trust. A single “overall trust score” for Alice would average these wildly different capabilities into a meaningless number. T3 keeps the roles separate so societies can make informed decisions.
Going Deeper: Sub-Dimensions+
The three root dimensions (Talent, Training, Temperament) can be broken down into nested layers. Each domain can define sub-dimensions without changing the core framework:
Societies define the sub-dimensions that matter for their context. A medical society might add “bedside manner” under Temperament. A technical society might add “code quality” under Training. The framework is extensible at every level.
Technical Details (Click to Expand)
T3 Tensor Structure
Each entity-role pair has its own T3 tensor. Tensors are never shared across roles. The canonical composite weights are Talent 0.4, Training 0.3, Temperament 0.3 — societies can customize weights per role:
// T3 tensor with role binding
{
"entity": "lct:alice",
"role_tensors": {
"web4:DataAnalyst": {
"talent": 0.85, // weight: 0.4 (canonical)
"training": 0.90, // weight: 0.3
"temperament": 0.95 // weight: 0.3
},
"web4:ProjectManager": {
"talent": 0.65,
"training": 0.70,
"temperament": 0.91
}
}
}
// Composite: 0.4 × talent + 0.3 × training + 0.3 × temperament
// DataAnalyst: 0.4(0.85) + 0.3(0.90) + 0.3(0.95) = 0.895How T3 Evolves
Each action within a role produces dimension-specific updates:
// T3 update impacts by outcome type Outcome Talent Training Temperament ───────────────────── ──────── ──────── ─────────── Novel Success +0.02-05 +0.01-02 +0.01 Standard Success 0 +0.005-01 +0.005 Expected Failure -0.01 0 0 Unexpected Failure -0.02 -0.01 -0.02 Ethics Violation -0.05 0 -0.10 // Decay half-lives (exponential, not linear) Talent: 365-day half-life (skills persist) Training: 180-day half-life (knowledge fades without practice) Temperament: 30-day half-life (recent behavior matters most) // Underlying formula (from spec test vectors): // base_delta = 0.02 × (quality - 0.5) // talent_delta = base_delta × 1.0 // training_delta = base_delta × 0.8 // temperament_delta = base_delta × 0.6
Why these specific half-lives? ▸
Talent (365 days) — Skills persist. A surgeon doesn't forget surgery after six months of vacation. Core abilities are durable, so trust in talent decays slowly.
Training (180 days) — Knowledge fades without practice. Last year's certification matters less than this year's. Moderate decay rewards ongoing learning.
Temperament (30 days) — Recent behavior matters most. Yesterday's kindness doesn't excuse today's betrayal. Fast decay means you must consistently demonstrate reliability.
These values are society-configurable parameters, not universal constants. A military society might use 7-day Temperament decay; a research lab might use 90 days.
Related: V3 (Value Tensors)
T3 measures who you are. V3 measures what you produce. Every output in Web4 gets scored across three dimensions:
- T3 (Trust Tensor): How much you trust someone — Talent, Training, Temperament (per role)
- V3 (Value Tensor): How much value something creates — Valuation (0.3 weight), Veracity (0.35), Validity (0.35)
V3 in Practice
A thorough, accurate review that catches real bugs. High across all dimensions — earns full ATP reward.
Gets clicks (some valuation) but misleading and poorly argued. Low V3 score means reduced ATP reward, and repeated low-veracity output drags down T3 Training scores.
Niche topic (lower immediate valuation) but rigorous and accurate. The weighted score (0.35 × 0.95 + 0.35 × 0.92 + 0.30 × 0.40 = 0.77) still earns good rewards because Web4 weights truth and reasoning more heavily than popularity.
V3 weights: Valuation 0.30, Veracity 0.35, Validity 0.35. Truth and reasoning outweigh popularity by design.
V3 decay half-lives: Valuation 14d (market conditions change fast), Veracity 365d (truth record persists), Validity 90d (certifications expire).
High T3 correlates with better V3 outcomes: Talent drives Valuation, Training drives Veracity, Temperament drives Validity. The relationship reinforces itself over time.
Integration with Other Web4 Pillars
- ATP economics: Higher T3 = earn more ATP for contributions
- Coherence Index (CI): Behavioral consistency modulates effective trust: effective_trust = T3 × CI²
- Karma: T3 above threshold allows rebirth with ATP carried forward
- MRH graph: T3 determines visibility in the relevancy horizon
See the Real Engine
This isn't a simulation — it's the actual web4-trust-core engine compiled to WebAssembly and running in your browser. The same code that powers protocol conformance testing.
Trust at Scale
T3 doesn't just work for individuals. The same three dimensions apply at every level of organization:
When Alice's talent improves, her team's trust adjusts upward. When a team member acts badly, the organization's score reflects it. Trust flows upward through composition and downward through accountability — the same T3 model at every scale.
Different entity types compose differently: teams use weighted averages, organizations use geometric means, and AI agents are bounded by their weakest dependency. The math varies, but the three dimensions stay the same.
Why This Matters
Trust is not one-dimensional. Humans don't trust uniformly — we trust doctors for medical advice, mechanics for car repairs, friends for emotional support. Each requires different capabilities, and our trust reflects that.
Trust is not context-free. A “7/10 trust score” tells you nothing about whether someone is brilliant-but-flaky or reliable-but-average. More importantly, it tells you nothing about whether they're trusted for the thing you need them to do.
T3 captures nuance without losing measurability. Three dimensions (Talent, Training, Temperament), role-specific, with fractal sub-dimensions for any domain. Rich enough to be useful, structured enough to be computable.
Gaming becomes genuinely hard. You can't inflate Talent by being reliable. You can't transfer trust from one role to another. You can't game three dimensions simultaneously across multiple roles. That's called “actually being trustworthy.”
Privacy is built in. You can prove “my trust exceeds your threshold” without revealing your exact score. Selective disclosure lets you share only what's needed — like proving you passed a background check without disclosing your medical records. Higher precision costs more ATP to prevent score inflation.
Want to see T3 evolve in real simulations? Try the Society Simulator and watch trust tensors change over time.