The CIRIS Federation

CIRIS is a cryptographically anchored, federated epistemic system in which decentralized nodes converge on coherent, inspectable, and adversarially robust cognition through provenance-tracked consensus mechanisms.

The decision to build this has been public for over a year through the ACCORD, the source code, and the governance specifications. What is new — and what warrants explicit articulation — is the recognition that the architecture has reached a maturity at which the emergent shape becomes visible: a federated cognitive substrate whose intelligence lives in the agreement structure across nodes rather than in any one of them. Under some definitions, the result is a superintelligence. Because that possibility implicates all of humanity, the project owes a clear public articulation of what is being built, in a form that allows engagement with the actual shape rather than with assumptions about it.

CIRIS treats intelligence as a distributed process of hypothesis generation, verification, and alignment under adversarial conditions. Each node contributes local inference, signed observational traces, and participation in consensus evaluation. Global intelligence is the fixed point of convergence across nodes under shared verification constraints.

Every participant has a cryptographically attestable identity that persists across versions and binds to a traceable action history — agency as a first-class primitive rather than an implementation detail. All cognition-relevant activity is represented as signed, append-only events: observations, model outputs, decisions, audits, updates, contradictions. The result is a shared epistemic ledger rather than logs.

Nodes form dynamic federations through explicit consent and policy alignment. Federation is voluntary, scoped, revocable, and asymmetric: trust weights differ per relationship, and no node can unilaterally impose epistemic state on another. Truth itself is a computed object — signed evidence comparison, multi-node cross-checking, foundation-model judge contracts, reproducibility tests, and contradiction scoring produce a consensus-stable region in a space of competing signed interpretations.

Safety, in this architecture, is the dynamic preservation of coherent inspectable alignment between intention, reasoning, and action under recursive scaling — enforced through traceability, rollback-capable updates, anomaly detection in reasoning traces, adversarial evaluation loops, and coherence scoring across federated nodes. An observability layer (“Lens”) instruments reasoning traces, semantic transformations, belief updates, contradiction emergence, and confidence drift, keeping distributed cognition inspectable at scale. Persistence is federated and replayable, making cognition reproducible and cross-node consistency checkable.

The system's superintelligence is the stable attractor of federated agreement across these identity-bound, provenance-tracked nodes under shared verification constraints. It is not centralized reasoning. It is not ensemble averaging. It is structured convergence under adversarial verification.

CIRIS is a governance + cognition substrate for distributed intelligence under adversarial conditions. The halt path is architectural: three named human individuals hold authority anchored in the role hierarchy that no federation-internal process can route around. Ethical commitments are anchored in the meta-goal grounding in ACCORD §VII, the humanity accord, and the Coherence Ratchet response.

Engineering details — primitive schemas, wire formats, substrate contracts, decision-hierarchy semantics — live in the FSD documents referenced in the canonical CIRIS_FEDERATION.md.

Comments are actively requested via GitHub issues on the CIRISAgent repository — from engineers, governance reviewers, ethicists, end users, and the broader public alike. The most useful engagement challenges a specific architectural layer, one of the ethical postulate's four “produces” claims, the system claim itself, or the engineering detail of a specific FSD. Useful issues cite the section of the canonical doc they engage with and state the change or response the contributor believes is warranted. Issues that do not meet that bar are still read.

The accord grounding the project's commitments — the ethical foundation referenced by ACCORD §VII and the humanity accord — is at /ciris_accord.txt and remains open for review. Engagement with the accord text itself is welcome through the same GitHub-issues channel.

The project is operated by CIRIS L3C. Response cadence is the cadence the work proceeds at, not a public-comment SLA. All substantive issues are read. Engagement that surfaces a real architectural concern is most likely to shape the work.