STRATON-LLM
Protocol & Negotiation
The distinguishing feature of STRATON-LLM is that negotiation is not treated as a loose conversation. It is executed as a protocol-driven, multi-turn process with explicit dialogue acts, bounded transitions, and evaluation at the end of the session.
Protocol flow
1. Trigger detection
An incoming message is intercepted. The term is checked against the local ontology and mapping store. If confidence is low or the concept is unknown, negotiation begins.
2. Context-aware classification
The trigger classifier distinguishes semantic mismatches from contextual ones so the system does not respond to every failure the same way.
3. Proposal planning
The system prepares negotiation acts such as propose, support, attack, defend, or request_evidence depending on the chosen argumentation mode.
4. Protocol-governed dialogue
A finite-state protocol engine manages turns, validates acts, and prevents the exchange from collapsing into uncontrolled free-form prompting.
5. Agreement assessment
The final candidate agreement is checked for trust, confidence, and dialogue quality before it is accepted.
6. Persistence and reuse
Accepted mappings and traces are stored so future interactions can resolve the same mismatch more efficiently.
Worked scenario: "price" vs "rate"
A simulation showing how the system moves from mismatch to reusable agreement.
TravelAgent
Requests hotel price for a booking package.
FinanceAgent
Recognizes “price” is not native in its ontology, where “rate” is the active term.
Trigger Layer
Lookup fails or returns low confidence, so negotiation is started.
Negotiation Module
Exchanges definitions, constraints, and contextual intent for the requested concept.
Protocol Engine
Validates support, attack, defend, and evidence-request acts during the dialogue.
Evaluation Layer
Scores the final “price ↔ rate” mapping before it is accepted into the mapping store.
Structured argumentation modes
Quick resolution
Used when the conflict is narrow and confidence is already close to the acceptance threshold. The goal is fast clarification with minimal turns.
Thorough argumentation
Used when the mapping is plausible but still requires stronger challenge, support, and clarification before trust can be earned.
Evidence-heavy debate
Used when the semantic stakes are high or the mismatch is severe. The system pushes for explicit definitions, examples, and stronger supporting evidence.
STRATON-LLM pages
Overview
Problem framing, project goals, research contribution, and end-to-end system story.
Architecture
Six-layer architecture, module boundaries, and system walkthrough.
Protocol & Negotiation
Triggering, dialogue acts, FSM control, and a worked negotiation scenario.
Design Decisions
The major choices, trade-offs, and constraints that shaped STRATON-LLM.
Evaluation & Learning
Agreement trust, heuristic scoring, persistence, and evolution roadmap.
Dialogue acts
These acts make the conversation measurable and machine-checkable, which is essential for later evaluation and learning.
Why protocol matters
Free-form dialogue is flexible but hard to trust. A protocol gives the system replayability, legal-turn validation, bounded sessions, and a clear record of how an agreement was reached.