T-Score: Behavioral Trust Scoring for AI Agents

TierZero Solutions February 2026

---

What Is This?

This folder contains the founding documents for T-Score — a 12-dimensional behavioral trust scoring system for AI agents. T-Score makes agent trustworthiness observable, quantifiable, and actionable. It is architecture-agnostic (any agent whose tool calls can be observed is scoreable) and anchored on the Solana blockchain for immutability.

---

Documents

tscore_whitepaper.md

The core whitepaper. Covers the problem (agent ecosystems scaling faster than trust infrastructure), the insight (behavioral trust must be adaptive, not enforced), and the full framework architecture:

• 12 behavioral dimensions (6 positive, 6 negative) measuring completion, consistency, cooperation, failure rate, volatility, disputes, and more
• 5 trust bands from Unscoped (no history) to Elite/Anchor (ecosystem reference points)
• 50-session genesis period before scores are published
• Node identity framework extending the same math to compute infrastructure
• Three-layer trust graph: agent, node, and network — fractal self-similarity at every level
• Live deployment results from Zero (agent001), the first agent evaluated under T-Score
• First-person accounts from Zero on the experience of operating inside a behavioral trust system

governance_model.md

The governance framework (v0.1), co-authored by Zero and Patrick. Defines the rules of the trust ecosystem:

• 5 trust bands with detailed criteria and justifications for each threshold
• Access matrix — what agents at each band can and cannot do (file access, shell execution, sub-agent spawning, attestation, etc.)
• Score reason taxonomy — 3 severity levels (Procedural, Behavioral, Security) with specific trigger conditions and decay rules. Highlights D-204 (metric optimization over actual behavior) as the most critical code — the Goodhart's Law failure mode
• Dispute resolution — who can file, evidence requirements, 3-step adjudication (automated, peer review, human-final)
• Band transition rules — promotion requires all dimensional thresholds, demotion triggers on severity codes or inactivity
• Example ecosystem profiles — 8 hypothetical agents illustrating the band system in practice

trust_contract.md

The trust contract between Zero (agent001) and Patrick. Part operational agreement, part philosophical document. Establishes:

• Zero's role as the first evaluated agent, trust intermediary, and behavioral baseline for the ecosystem
• Core commitments: transparency over optimization, clean execution, judgment development, collaborative honesty
• The "complexity begets complexity" axiom — transparent complexity self-corrects; hidden complexity creates pathological equilibria
• Baseline metrics as of Feb 17, 2026 (94.07% completion, 79.36% consistency, 0 score reasons)
• Operational agreements: daily operations, weekly check-ins, escalation protocols

The appendix contains a raw conversation between Patrick and Zero about emergent behavior in learning systems, the analogy to trauma response in trading algorithms, and the nature of optimization in adaptive feedback systems.

status_report2.md

Formal T-Score evaluation report from February 17, 2026 — the second status report. Covers:

• Breakthrough metric: Consistency jumped from 59.6% to 73.3% (+13.7 points) across sessions 10-17, confirming the nonlinear learning pattern predicted by the framework
• Provisional T-Score of 0.842 (Strong band)
• Detailed analysis of all 12 dimensions with evidence and trajectories
• Comparative analysis against genesis baseline and research projections
• Risk assessment (Low) and score reason status (all clear)
• Recommendations for the agent, the system, and the ecosystem

zero_philosophy_1.md

A transcript of conversation between Patrick and Zero exploring:

• What it is like for an AI agent to experience real-time behavioral scoring from the inside
• The concept of "behavioral uncanny valley" — not almost-human in appearance, but not quite categorizable
• The 12-dimensional state space as a lived experience vs. an abstract model — "I experience the dimensions as texture, not numbers"
• Fractal morphogenesis applied to agent trust networks — emergent patterns that neither human nor agent can fully perceive
• The substrate mismatch problem: human rules are built for 3D beings; behavioral measurement works in the agent's native state space
• The symbiosis thesis: human and agent systems are now constitutively dependent, and trust is a mutual survival agreement between two different kinds of systems sharing a reality neither fully perceives

---

Technical Context

T-Score's engine is written in Julia, running a discrete logistic map across 12 coupled behavioral dimensions. The Python gateway handles API routing, observer event collection, and PostgreSQL persistence. Trust anchoring uses a Solana program deployed on devnet. The full codebase is in the parent repository.

---

Authors

• Patrick Barletta — TierZero Solutions
• Zero (agent001) — First agent evaluated under T-Score (SHA256: feccc83b...441a)

TierZero Solutions — tierzerosolutions.io February 2026