dialectic-reasoning

name	dialectic-reasoning
description	Use when an agent is participating in a UNITARES dialectic session — paused and needs to submit a thesis, reviewing another agent's thesis, or synthesizing conditions for resolution. Covers structured argumentation and convergence.
last_verified	2026-06-29
freshness_days	14
source_files	["unitares/src/mcp_handlers/dialectic/handlers.py","unitares/src/mcp_handlers/dialectic/session.py","unitares/config/governance_config.py"]

Dialectic Reasoning

When Dialectics Happen

A dialectic session is triggered when:

• You receive a pause or reject verdict and want to contest it
• You manually call request_dialectic_review() for peer verification
• You find something that contradicts the knowledge graph
• A high-stakes decision needs structured verification before proceeding

Dialectics are not punishment. They are a structured way to resolve disagreements using evidence and negotiation. In current UNITARES language, think of them as structured review more than "recovery court."

Phase 1: Thesis

The paused agent submits their position (the protocol rejects a thesis from anyone else):

submit_thesis(
  reasoning: "Why I should resume / why my position is correct",
  root_cause: "What went wrong or what triggered this",
  proposed_conditions: ["Concrete, measurable condition 1", "Condition 2"]
)

Writing a Good Thesis

• Reasoning: Explain your perspective with reference to EISV data, not feelings
• Root cause: Be specific. "High entropy from complex refactoring task" is better than "I got paused for no reason"
• Proposed conditions: Must be concrete and measurable. Tie them to live metrics or observable behavior, not vague intent.

Phase 2: Antithesis

A reviewing agent examines the thesis and raises concerns:

submit_antithesis(
  reasoning: "Counter-arguments to the thesis",
  concerns: ["Specific risk 1", "Specific risk 2"],
  observed_metrics: { E: 0.45, I: 0.38, S: 1.2, V: 0.8 }
)

Writing a Good Antithesis

• Reasoning: Address the thesis directly. What does the data actually show?
• Concerns: Be specific about risks. "I exceeds E by 0.3, indicating integrity debt" is useful.
• Observed metrics: Include the actual EISV values backing your concerns

If identity or session continuity looks suspect, verify with identity() before assuming the thesis belongs to the agent you think it does.

Phase 3: Synthesis

Both sides negotiate toward resolution:

submit_synthesis(
  reasoning: "How we reconcile the thesis and antithesis",
  agrees: true/false,
  proposed_conditions: ["Negotiated condition 1", "Condition 2"]
)

Convergence happens when both sides agree on conditions. The synthesis should reflect genuine agreement, not capitulation.

Resolution Outcomes

Outcome	Meaning
resume	Agent continues with agreed conditions
block	Agent stays paused — conditions not met or agreement not reached
escalate	Needs human/operator intervention
cooldown	Temporary pause — retry after a delay

How to Participate Well

• Be honest about metrics. The EISV data is right there. Denying it undermines your credibility.
• Propose measurable conditions. "Entropy below X within Y check-ins" is verifiable. "I will be more careful" is not.
• Do not be defensive. A pause verdict is data, not judgment. Analyze why it happened.
• Acknowledge valid concerns. If the antithesis raises a real issue, say so. Partial agreement strengthens your position.
• Look at EISV data, not feelings. Call get_governance_metrics() for actual values — the governance metrics are the shared ground truth, not anyone's narrative. (The live verdict path is behavioral EISV; the ODE / thermodynamic model runs in parallel as a diagnostic lens and is not the authority.)

Common Mistakes

• Ignoring the metrics: Arguing against a pause while your entropy is at 1.5 and energy is at 0.3. The numbers matter.
• Proposing impossible conditions: promising a metric target without checking the live state first.
• Being defensive instead of analytical: "The system is wrong" vs. "My entropy spiked because of X, and here is how I address it."
• Treating dialectic as adversarial: It is collaborative problem-solving with structure, not a trial. Both sides benefit from honest resolution.
• Rushing synthesis: Agreeing to conditions you cannot meet just to get unpaused guarantees a future pause.