Glossary
A concept hub for the core terms used across the RCT ecosystem, designed to improve discoverability, comprehension, and internal linking.
FDIA Equation
F = (D^I) × A — A constitutional equation where Future output (F) is determined by Data quality (D) raised to the power of Intent (I), multiplied by Architect authorization (A). When A=0, F=0 always.
JITNA Protocol
Just In Time Nodal Assembly — an open agent-to-agent communication protocol (RFC-001 v2.0) often called 'the HTTP of Agentic AI.' Defines PROPOSE → COUNTER → ACCEPT negotiation flow with Ed25519 signatures.
SignedAI
The multi-LLM consensus verification system of the RCT Ecosystem. Routes critical queries through 4–8 models simultaneously and requires agreement before releasing results. Achieves 0.3% hallucination rate vs 12–15% industry average.
RCTDB
Universal memory schema with 8 dimensions: query_hash, fdia_scores, subject_uuid, model_chain, consensus_result, delta_chain, timestamp, provenance. Enables PDPA-compliant right to erasure via UUID tombstone pattern.
Delta Engine
Memory compression system that stores only incremental state changes (deltas) rather than full state snapshots. Achieves 74% average lossless compression with sub-1ms reconstruction, enabling warm recall under 50ms.
HexaCore
The 7-model AI infrastructure of the RCT Ecosystem: 3 Western models + 3 Eastern models + 1 Thai regional model (Typhoon v2). Provides 3.74x cost reduction vs single-model deployments through intelligent task routing.
Constitutional AI
A governance-oriented approach where model behavior and system operation are shaped by explicit rules, review logic, and safety boundaries rather than by output generation alone.
Constitutional Kill Switch
In the RCT Ecosystem, when the FDIA Architect variable (A) is set to 0, all AI output is unconditionally blocked — F=0 regardless of input quality. This is implemented as a hard mathematical gate, not a soft preference.
Deterministic AI
An AI system property where identical inputs always produce identical outputs, or where safety constraints hold unconditionally regardless of model behavior. In the RCT Ecosystem, constitutional constraints (A=0→F=0) are deterministic.
Intent Operating System
An AI orchestration layer that functions like an operating system — managing resource allocation (model routing), access control (FDIA gate), state management (RCTDB), and audit trails (provenance). The RCT Ecosystem is an Intent OS.
Agentic AI
AI systems that can autonomously plan, execute, and iterate on multi-step tasks without requiring step-by-step human instruction. Agentic AI requires inter-agent communication standards like JITNA.
Multi-LLM Consensus
The practice of routing a query to multiple LLM providers simultaneously and requiring agreement between outputs before accepting results. Reduces hallucination by catching model-specific biases. Used in SignedAI (Tier 4/6/8).
Dynamic Routing
Choosing different model or workflow paths depending on risk, complexity, cost, latency, or evaluation requirements rather than sending every request through the same path.
Hallucination Control
The system-level discipline of reducing unsupported or overconfident outputs through retrieval quality, memory design, routing, verification, and evaluation loops. RCT achieves 0.3% vs 12–15% industry average.
Verification Layer
A step that checks generated outputs against references, policies, rules, schemas, or second-pass evaluators before high-value actions are accepted.
Intent Operations
A framing for AI behavior that emphasizes understanding goals, context, and constraints before deciding what actions or outputs should follow.
Prompt Injection
A security attack where malicious text in user input attempts to override the system's intended behavior. RCT's JITNA Normalizer automatically strips known injection patterns on every request before LLM processing.
Consensus Threshold
The minimum percentage of model agreement required before a SignedAI result is accepted. Tier 4 = 50%, Tier 6 = 67%, Tier 8 = 75%, Tier S = 100%. Higher consensus thresholds are used for higher-stakes decisions.
Warm Recall
Serving an AI response from the Delta Engine hot-zone semantic cache (similarity threshold 0.95) without calling an LLM. Achieves under 50 milliseconds response time. Only possible for queries semantically similar to cached responses.
Cold Start
A novel AI query that cannot be served from cache and must be fully processed through the LLM pipeline. Cold start time in the RCT Ecosystem is 3–5 seconds, compared to under 50ms for warm recall.
UUID Tombstone Pattern
A PDPA-compliant data erasure method where a subject's uuid is marked as 'tombstoned' rather than physically deleted, ensuring no data is retrievable while maintaining referential integrity. Used in RCTDB for right-to-erasure requests.
Semantic Similarity
A numerical measure (typically 0.0–1.0) of how similar two text inputs are in meaning, regardless of exact wording. The Delta Engine uses a 0.95 similarity threshold to determine if cached responses can be served for new queries.
Hot Zone
The fastest-access tier of the RCTDB memory hierarchy. Keeps most-frequently-accessed semantic cache entries in-memory for under 1ms access. Capacity-limited; entries migrate to warm and cold zones based on access frequency.
Property-Based Testing
A testing methodology where mathematical properties (invariants) are specified, and a framework like Python Hypothesis automatically generates thousands of test cases to find edge cases that violate those properties.
Chaos Engineering
A testing discipline where controlled failure scenarios (server outage, network partition, service timeout) are deliberately injected into a system to discover failure modes before they occur in production.
Vibe Testing
The informal practice of evaluating an AI system's quality by looking at its outputs and judging whether they 'feel right.' Not sufficient for enterprise AI deployments — formal evaluation harnesses are required.
Evaluation Harness
A systematic, automated set of quality gates that an AI system must pass before deployment. The RCT Ecosystem runs 4,849 tests across 8 levels (Unit, Integration, Service, Contract, Performance, Security, Chaos, Property).
Mathematical Invariant
A property that must hold unconditionally for all valid inputs, regardless of edge cases. Example: FDIA invariant — when A=0, F must equal exactly 0 for all possible D and I values.
PDPA Section 33
Thailand's Personal Data Protection Act Section 33 grants data subjects the right to request an explanation of automated decisions made about them. RCT's RCTDB provenance trail (dimension 8) automatically satisfies this requirement.
PDPA Right to Erasure
The right of data subjects under PDPA (and GDPR) to request permanent deletion of their personal data. In RCTDB, this is implemented via the UUID tombstone pattern — marking subject data as erased without breaking referential integrity.
Audit Trail
A chronological record of all AI decisions, with sufficient detail to reconstruct what happened, when, and why. RCTDB's 8-dimensional schema automatically generates audit trails for every RCT Ecosystem query.
E-E-A-T
Google's quality signal framework: Experience, Expertise, Authoritativeness, Trustworthiness. For AI content, E-E-A-T requires first-hand experience, verifiable credentials, transparent authorship, and fact-checked claims.
Topical Authority
The degree to which a website is recognized by search engines as a comprehensive, reliable source on a specific subject. Built through content depth, internal linking between related articles, and entity-level structured data.
AEO (Answer Engine Optimization)
The practice of structuring content to be selected as direct answers by AI systems (ChatGPT, Perplexity, Google AI Overviews). Requires FAQPage JSON-LD schema, clear question-answer format, and verified factual claims.
FAQPage Schema
A JSON-LD structured data format that marks up question-and-answer content with schema.org/FAQPage markup. Eligible for Google Featured Snippets and rich results, significantly improving click-through rates.
DefinedTerm Schema
A JSON-LD structured data type (schema.org/DefinedTerm) used to mark up glossary entries and technical terms. Helps search engines understand proprietary concepts and enables Knowledge Graph entity recognition.
Internal Linking
Hyperlinks between pages on the same website that distribute PageRank, help search engines discover content, and guide users through related topics. High-quality internal links include descriptive anchor text.
Architect Genome
Creator's DNA — the foundational design genome of the RCT 7-Genome System. Encodes the architectural vision and structural patterns of the RCT Ecosystem.
ARTENT Genome
Creation intelligence — the generative intelligence genome of the RCT 7-Genome System responsible for creative AI output capabilities.
JITNA Genome
Protocol layer — the communication genome of the RCT 7-Genome System. Implements the JITNA Protocol standard for agent-to-agent communication.
Codex Genome
Knowledge vault — the long-term knowledge store genome of the RCT 7-Genome System. Manages structured knowledge retrieval and organization.
SignedAI Genome
Verification layer — the trust and consensus genome of the RCT 7-Genome System. Implements multi-model consensus and cryptographic verification of all system outputs.
RCT-KnowledgeVault Genome
Memory architecture — the persistent storage genome of the RCT 7-Genome System. Manages the RCTDB schema, Delta Engine compression, and the hot/warm/cold zone memory hierarchy.
RCT-7 Genome
Continuous improvement — the meta-learning genome of the RCT 7-Genome System. Monitors system performance, identifies improvement opportunities, and governs the evolution of the entire ecosystem.