Reference documentation for the FORETELLION cognitive engine: how a request is classified, how a council of reasoning lenses is selected, and how deliberation is recorded as provenance. Kernel v4.1.0-evidence-ledger, engine 1.1.0-wired.
Every request the engine accepts moves through three deterministic steps. Each one produces an artifact that later becomes provenance, so the reasoning behind an answer can be replayed and audited.
A problem classifier assigns the problem type, 2 to 6 discipline tags from a 50-group taxonomy, a complexity rating, keywords, and a suggested stage focus.
Every thinker in the 300-mind library is scored against the classified domains, keywords, and stage focus. The chosen mode fixes the council size, and adversary anchors are boosted so dissent is always seated.
The council deliberates through the seven reasoning stages, on the Council surface or inside a chat session, and every lens selection and stage trace is recorded.
Deliberation walks the same seven questions every time, so reasoning is comparable across requests and traceable in the audit record.
Anchor thinkers: Turing, Shannon, Hume, Darwin, Feynman, Russell
Anchor thinkers: Aristotle, Descartes, Euclid, Newton, Simon, Ackoff
Anchor thinkers: Socrates, Kahneman, Tversky, Hume, Taleb, Feynman
Anchor thinkers: Meadows, Wiener, Bertalanffy, Forrester, Holland, Pearl
Anchor thinkers: Leonardo, Fuller, Jobs, Christensen, Drucker, Munger
Anchor thinkers: Darwin, Meadows, Taleb, Clausewitz, Diamond
Anchor thinkers: Aristotle, Kant, Rawls, Dworkin, Hart, Ostrom, Nussbaum, Sen
Modes set how large the council is and what posture it takes. Adversary modes guarantee falsificationist thinkers are seated.
Fast triangulation across a few strong lenses
Thorough multi-domain analysis
Adversary seatedLarge council with explicit dissent
Adversary seatedAdversarial reasoning and failure discovery
Adversary seatedStrip the problem to fundamental assumptions
Adversary seatedFeedback loops, incentives, second-order effects
Adversary seatedDivergent thinking and unconventional solutions
Options compared by evidence, uncertainty, reversibility, risk
Adversary seatedCompeting ethical frameworks, no auto-resolution
Adversary seatedEvidence quality, source verification, uncertainty
Adversary seatedFourteen canonical contrarians and falsificationists are anchored into adversarial modes. When the engine runs Red Team or Deep Think, these minds are boosted in selection, so every conclusion faces its strongest critic.
The cognitive engine is the Reason authority and the ForetellionGate kernel is the Govern authority. The two are architecturally separate: reasoning never bypasses governance, and governance never fabricates reasoning. The engine sits above the model as a substrate, it does not replace it.
Provenance: council selection, stage traces, and mode metadata are stored per session and per response, so any answer can be traced back to the lenses and stages that produced it.