Decision Cycles

The repeating process that drives state transitions.

The Cycle

Propose -> Vote -> Arbitrate -> Execute -> (repeat)

Each cycle attempts to move the session from one state to another.

Phases

1. Propose

Each registered proposer submits a transition proposal.

What proposers receive:

• •Current state
• •Available transitions
• •Transition prompts
• •Session history

What proposers return:

• •Action name (e.g., approve, reject)
• •Target state
• •Reasoning

2. Vote

Each registered voter compares proposals pairwise and votes.

What voters receive:

• •The two proposals being compared
• •Current state context
• •Session history

What voters return:

• •Which proposal they prefer (A or B)
• •NEITHER if both are bad
• •Confidence level

3. Arbitrate

The arbiter evaluates consensus using the configured strategy.

Strategies:

Human primacy: Human votes override AI votes. A single human vote for proposal A beats any number of AI votes for proposal B.

4. Execute

If consensus is reached, the winning transition is applied.

What happens:

• •Session state updates to the target state
• •TransitionRecord added to session.history
• •Next cycle begins (or session completes if goal reached)

Watching the Cycle

npx dialai machine.json --verbose

Verbose output shows:

[PROPOSE] ai-proposer: approve -> approved
[PROPOSE] ai-proposer-2: reject -> draft
[VOTE] ai-voter: A (approve) - confidence: 0.8
[VOTE] human-voter: A (approve) - confidence: 1.0
[ARBITRATE] consensus reached: approve
[EXECUTE] draft -> approved

No Consensus

If voting doesn't produce a winner:

• •The cycle repeats
• •New proposals are solicited
• •Different proposals may emerge

Configure max cycles to prevent infinite loops:

{
  "arbiter": {
    "strategy": "majority",
    "maxCycles": 5
  }
}

Session Completion

The session ends when:

• •Current state equals defaultState (success)
• •Max cycles exceeded (failure)
• •No valid transitions available (stuck)