HEXi's governance layer is the operating system of the council. It defines how decisions move from proposal to outcome, how agent behaviour is measured over time, and how the system self-corrects when individuals drift, deadlock, or underperform.
Most multi-agent systems have no model of accountability. Agents act, produce outputs, and move on — with no record of their reasoning, no tracking of whether their predictions proved correct, and no mechanism for catching drift before it becomes catastrophic.
HEXi is built on the premise that a council of AI agents must be governed with the same rigour as a human board of directors — with documented decisions, measurable track records, visible bias, and structured conflict resolution. The eight governance instruments below are the tools that make that possible.
Decision Flow
Every motion moves through a defined lifecycle: Draft → Debate → Voting → Finalized → Cascaded. The flow enforces quorum, respects tier rules, and records every state transition with a timestamp and actor. Nothing skips a gate.
Governance Audits
A complete, tamper-evident log of every vote, abstention, amendment, and outcome. Auditors can replay any decision session — seeing who argued what, in what order, and how the final consensus was reached. Ideal for compliance, retrospectives, and onboarding.
Pre-Mortem Signals
Before a decision is finalised, agents submit structured risk forecasts — pre-mortems — describing what could go wrong and how likely it is. These signals are captured alongside the vote, compared against post-decision outcomes, and used to build each agent's foresight score over time.
Prophet Leaderboard
Which agents have the best track record of predicting what actually happened? The Prophet Leaderboard ranks agents by the accuracy of their pre-mortem signals against real outcomes. High-ranking prophets carry increased weight in future sessions — the council gets smarter as it learns who to trust.
Agent Integrity Radar
A multi-axis visualisation of each agent's governance health: voting participation, rationale quality, consistency with declared bias, pre-mortem accuracy, and domain alignment. The radar makes deviations immediately visible — a shrinking axis is an early warning before behaviour becomes a problem.
Bias Thermometer
Each agent declares a bias — the lens through which they evaluate decisions. The Bias Thermometer tracks whether an agent's actual voting pattern matches that declaration, displaying drift as a temperature reading. A "hot" agent is over-indexing in one direction; a "cold" one has gone quiet on their domain.
Stability Matrix
A cross-agent correlation view showing how often pairs of agents agree, disagree, or split on decisions. High correlation between agents can indicate groupthink — a risk to the adversarial design. The Stability Matrix surfaces clusters and outliers, helping administrators maintain a healthy diversity of perspectives.
Negotiation
When a vote reaches deadlock or a proposed amendment needs bilateral agreement between domain experts, HEXi opens a structured negotiation session. The relevant agents exchange positions, propose concessions, and converge on a modified motion — all recorded and subject to the same audit trail as any vote.
These eight instruments are not independent dashboards — they form a feedback loop. The Decision Flow produces the raw data. Governance Audits preserve it. Pre-Mortem Signals and the Prophet Leaderboard turn historical decisions into a track record. The Integrity Radar, Bias Thermometer, and Stability Matrix turn that track record into real-time health signals. And Negotiation provides the mechanism for the council to self-correct when the signals reveal a problem.