A distributed reasoning architecture in which civilizational signals
become structured hypotheses, validated frameworks, and the next set of signals.
Signal agents are the engine's interface with the world. They continuously ingest the streams of evidence on which civilizational thinking must be conditioned — scientific literature, geopolitical movements, financial architecture, regulatory drafting, climate data, infrastructure decisions, and the moving frontier of AI capability itself.
The discipline here is selectivity over volume. Most "news" is noise from the perspective of century-scale design. Signal agents discard the cyclical and preserve the structural — the moments where an underlying system reconfigures rather than merely fluctuates.
Hypothesis agents perform the most consequential transformation in the engine: from event to proposition. They do not ask what happened. They ask what new systemic pattern is now legible because of what happened.
A hypothesis, in the KRYONIS sense, is not a guess. It is a structured statement bound to its evidentiary base, framed in language precise enough to be tested. Every proposition that leaves this stage carries a thesis, a domain, and a candidate explanation.
The most common error in civilizational analysis is to mistake a rename for an insight. Novelty scoring is the engine's defence against this: every hypothesis is measured against the existing literature of capital, value, governance, and verification.
The output is not a binary "new / not new" but a structured map of where the hypothesis genuinely departs from inheritance — and where it merely re-clothes ideas that have already been priced.
Critique agents are adversarial by design. Each one carries a specific institutional voice — the regulator, the investor, the scientist, the sceptic, the geostrategist, the bioethicist — and each attempts to break the hypothesis before reality does.
A hypothesis that survives this stage does so because its weaknesses are known and bounded. The engine prefers a fragile hypothesis it understands to a confident hypothesis it does not.
Surviving hypotheses are translated into operational form: policy scenarios, economic models, adoption pathways, risk maps, protocol designs, dashboard prototypes. Simulation does not mean physical modelling alone — it means converting a proposition into something an institution could actually test.
Experiment agents define the next deployable step. A hypothesis without a next move is just an opinion. The output of this stage is concrete: a paper, a calculator, a public registry, a pilot architecture, a memorandum, an investor brief, a deliberate test of language in the open.
The standard is reversibility. Every experiment is sized so that its failure teaches something the engine could not have learned by reasoning alone.
The world's response to the engine — engagement, silence, contestation, adoption, invitation — is itself data. Feedback agents re-encode this response as signal and return it to the start of the loop. The engine learns not by introspection but by contact.
This is the closing element of operationalized foresight. Foresight that does not adjust to reception is prophecy. Foresight that adjusts is method.
Every cycle returns more structured than the last. This is the discipline of the engine: not faster prediction, but better-tested civilizational thinking.