The trust modelBy Odebode Toluwalase Esq · 5 min read

Multi-agent debate (DyTopo)

A tiered bench of specialist agents proposes and challenges findings in parallel rounds, in an order set by a learning loop. A claim has to survive scrutiny before you see it.

LexCore doesn't ask one model for an answer. It runs a tiered bench of specialists: a root orchestrator, domain managers, and epistemic, conceptual and writing agents. Each one is specialised by its instructions, not by being a different model.

Specialists include a contract-risk analyst, clause comparator, precedent researcher, corporate-governance and data-protection agents, materiality and enforceability analysts, a redliner, and a summary writer.
Each specialist's knowledge lives in a versioned, reviewable file, not a prompt buried in code. It carries Nigerian grounding for the area that specialist covers, from company governance and data protection to dispute clauses.
That material is orientation, never a citation. It tells a specialist where to look. The specialist may not state a section number it cannot see, and has to send an ungrounded point to a human gate instead.
Agents run in parallel and blind to one another in each round, so their findings are independent commitments. Nobody piles in behind the loudest output.
A Q-learning loop sets the order specialists run in. An optional recruitment mode lets domain managers pick a domain-scoped subset each round (deterministic, with no extra model call).
A Precedent Board brings prior outcomes into the debate.

Findings are proposed, challenged in debate, put through CitationGate, and only then synthesised. The debate is why a finding is more than one model's first guess.

Related documentation