trinity-dialectic

A single-file dialectical evaluation engine that forces three independent reasoning lenses before any LLM output is finalized.

Status

Experimental. Maintained by Sage / 0SxD as part of an ongoing research portfolio focused on prompt engineering, agent skills, and LLM evaluation.

What this is

trinity.py is a reference implementation of the Trinity Dialectic: a structured self-evaluation protocol built into the 143_protocol_a operating contract for multi-agent AI systems. It addresses premature commitment in LLM agents by requiring every proposed output to pass three independent gates - Logos (logical rigor), Pathos (creative alignment), and Ethos (practical executability) - before finalization. The fight between Logos and Pathos, arbitrated by Ethos, is the productive mechanism. This is not a pipeline; it is a dialectic grounded in Aristotle's Nicomachean Ethics and phronesis (practical wisdom).

Approach

• Three evaluation gates fire in sequence; a score below threshold triggers a LOOP_BACK_REVIEW_REPLAN_RETURN cycle rather than suppressing the output
• Trinity of Trinities: each gate applies its own internal three-axis evaluation, yielding a 3x3 grid with a maximum of 15 evaluation tokens
• No external dependencies beyond the Python standard library
• Designed for agentic systems where outputs become inputs to downstream agents and output drift compounds over long task horizons
• Source rooted in trinity.py; examples in examples/

Layout

• trinity.py - the full dialectical engine (TrinityEngine, DefaultDialecticGate, TrinityOfTrinities)
• examples/basic_eval.py - architecture decision walkthrough through all three gates
• examples/socratic_gate.py - Socratic variant: surface the better question before answering the stated one
• _visuals/ - evaluation flow diagram (PNG; EXIF stripped)

Usage / How to read this

No install required beyond Python 3.x:

from trinity import TrinityEngine, DefaultDialecticGate, ThreatLevel

engine = TrinityEngine(gate=DefaultDialecticGate())
context = {
    "logos_claim": "Use the proven solution.",
    "logos_reasoning": "Three independent benchmarks confirm this approach.",
    "logos_evidence": ["benchmark A", "benchmark B", "benchmark C"],
    "logos_confidence": 0.82,
    "pathos_claim": "The disruptive approach yields 3x gains if it works.",
    "pathos_reasoning": "No one has tried combining these two methods before.",
    "pathos_evidence": ["theoretical basis from paper X"],
    "pathos_confidence": 0.60,
    "threat_level": ThreatLevel.NONE,
    "violations": [],
}
result = engine.run_dialectic(context)
print(result.phase.name)          # VERIFIED
print(result.final_verdict.name)  # LOGOS_WINS
print(result.final_position)      # "Use the proven solution."

Start with examples/basic_eval.py for a full walkthrough. See also 143-protocol (link below) for the full operating contract this engine is part of.

Prior art and citations

• GEPA: Reflective Prompt Evolution, arXiv:2507.19457 - reflective self-evaluation loop that motivates the LOOP_BACK_REVIEW_REPLAN_RETURN cycle
• AGENTS.md spec (Linux Foundation Agentic AI Foundation) - behavioral contract model
• Aristotle, Nicomachean Ethics Books I-II, VI - foundational source for Ethos/Logos/Pathos framing and phronesis; cited verbatim in trinity.py module docstring

See also: 143-protocol for the full operating contract.

License

MIT. Copyright (c) 2026 Sage / 0SxD.

Notes

This repo is part of an active R&D portfolio. Content may move, change, or be withdrawn. Issues and PRs welcome but reviews are best-effort.