the foundational problem of cyber inference
if true has cyberank 10 and false has cyberank 9, then for any question cyberlinked to both, the answer is always true — regardless of context. global rank dominates
the problem generalizes: any high-rank particle wins every contextual query it appears in. a question "what causes malaria?" linked to both "plasmodium" (rank 50) and "bad air" (rank 5000) answers "bad air" — not because it is correct, but because it is popular. cyberank measures what the graph attends to globally, not what is true locally
why global rank fails for inference
cyberank is a per-particle score. it answers "how important is this particle across the whole cybergraph?" — not "how relevant is this particle to this question?" a system that answers every question with the most popular connected particle is a search engine, not intelligence
the insight: inference requires contextual truth. the same particle can be the right answer to one question and wrong for another. a single global number cannot encode this
the solutions
cyber/truth/standard inference — the naive first attempt. multiply global cyberank by concentrated will per cyberlink in context. breaks global dominance by introducing a per-neuron conviction signal. simple and zero-cost, but still a single-factor approximation with no honesty guarantee and no market correction
cyber/truth — the full architecture. three layers that together make contextual truth emerge:
| Layer | Mechanism | What it solves |
|---|---|---|
| tri-kernel local reconvergence | context particles shift the probability distribution locally | global rank dominance |
| serum + valence | honesty is a Bayes-Nash equilibrium | strategic voting |
| ICBS markets | capital flows against false edges | persistence of incorrect answers |