Iterative Reasoning Engine

Multi-agent reasoning system that takes a single user prompt, autonomously generates the problem scenario, creates specialised agents with distinct roles, and orchestrates structured debate across iterative rounds. The user provides the question — the system builds the entire reasoning apparatus around it, including which perspectives are needed and how they should interact.

The orchestrator manages agent lifecycle: spawns new agents when a perspective gap is identified, removes agents that aren't contributing, redirects agents that drift off-topic, and adjusts micro-parameters continuously based on debate quality. Agents self-modulate their own parameters across iterations — confidence levels, assertion strength, and willingness to concede shift as the debate develops evidence for or against their positions.

A data collection layer monitors for information gaps during debate — when agents make claims without supporting evidence or when the discussion reveals unknown territory, the system can source additional data to feed into subsequent rounds. An analysis layer runs in parallel detecting logical fallacies, confirmation bias, circular reasoning, and blind spots across all agents' outputs.

Users can inject new agents mid-debate to introduce perspectives the system didn't anticipate. The orchestrator keeps running notes summarising the state of the discussion, key agreements, unresolved conflicts, and confidence levels per conclusion. Built in Swift for macOS with a real-time reasoning observatory interface showing agent interactions, parameter shifts, and convergence patterns as they happen.