Theory Consensus Reactor

A governance-constrained IAF cluster that turns a pool of agents into a decision engine — competing proposals driven through a six-step pipeline toward a single winning decision.

2026-07-28 TCR Architecture

The Theory Consensus Reactor (TCR) is a special kind of IAF cluster: several IAF units wired together under a fixed governance process. Where a plain IAF unit executes an action, a TCR makes a decision — it takes a question, has a pool of agents compete to answer it, and converges on one winning output. That output, a document, then becomes the prompt that drives downstream action units. Decisions driving action.

Where TCR Sits

The larger system is a four-level nesting, from the organisational container down to a single LLM call:

IC   (organisational container)
 ├─ TCR   (decision node: a governance-constrained IAF cluster)
 │   └─ IAF × N   (each pipeline step is one IAF unit)
 │        └─ Dispatch × M   (each agent = one LLM call loop)
 └─ IAF / IAF cluster   (action node: business execution units)
      └─ Dispatch × M

In one line: IC is the organisation, TCR makes decisions, IAF executes, Dispatch runs the model. TCR is itself built from IAF and Dispatch — it is an IAF cluster with a special governance process laid over it. This mirrors the Intelligenism theory of the Intelligent Consortium: a drive node that decides, and action nodes that act.

The Bidding Stage

Before any writing happens, agents bid for the two roles in the round:

Proposal bidding. Every agent decides whether to enter as a proposer; the lowest bidders win the proposer slots.
Bet bidding. The remaining agents decide whether to bet; the highest-stake bidders win the bettor slots.

Roles are earned by staking, not assigned — an economic filter that makes each agent commit to the role it believes it can win value in. A round that attracts no viable bids is voided and only its accounting runs.

The Six-Step Pipeline

Once roles are set, proposers and bettors drive the proposals through six steps:

Step	Actor	What happens
1 · Origin	proposer	Each proposer writes a first-draft proposal.
2 · Critique	bettor	Bettors challenge and evaluate the proposals.
3 · Revise	proposer	Proposers revise their drafts in light of the critique.
4 · Score	bettor	Phase-1 scoring of the revised proposals.
5 · Integrate	proposer	Non-winning proposers integrate the leading proposals.
6 · Score	bettor	Phase-2 scoring of the integrated proposals.

Each step is a distinct IAF unit; every step runs in its own isolated process. The pattern is deliberately adversarial: the agents who did not write a proposal are the ones who critique and score it, so quality is pressure-tested by parties with a stake in a different outcome.

Two-Phase Settlement

Settlement narrows the field in two passes. Phase 1 selects the top ~20% of proposals after the first scoring round. Phase 2, after the integration and re-scoring steps, selects the single winning proposal. A dual-currency ledger tracks stakes and payouts across the whole round, so winning and losing carry real economic consequences for the agents involved.

Why This Design

Decisions driving action. The winning document is injected as a prompt into downstream IAF dispatches — the reactor decides, the action nodes execute. The two are cleanly separated but tightly coupled.

Independence expressed through connection. Like the framework it is built on, TCR keeps each agent an independent unit; the intelligence of the whole comes from the governance process that connects them — bidding, adversarial review, and staked settlement — not from any single agent being clever.

TCR is the decision-making organ of the Intelligenism ecosystem: a repeatable, economically grounded procedure for turning many competing agent opinions into one actionable output.