cyber/truth/standard inference.md

the naive first solution to the true-false problem — a single-factor contextual weighting that preceded the full cyber/truth architecture

the algorithm

given a query particle Q, compute a contextual score for each candidate answer:

candidates = particles cyberlinked with Q
for each candidate P in candidates:
    links = cyberlinks between Q and P
    weight = 0
    for each link in links:
        neuron = link.neuron
        avg_will = neuron.will_balance / neuron.total_cyberlinks
        weight += avg_will
    score(P) = cyberank(P) × weight
return candidates sorted by score

the intuition: a neuron who concentrates will across few cyberlinks signals stronger conviction per link. a neuron who spreads will across thousands of links contributes less per link. the score multiplies global cyberank (what the graph thinks matters) by concentrated will in context (what committed neurons think matters here)

why it works against the true-false problem

if true has cyberank 10 and false has cyberank 9, global rank always picks true. but if the neurons who linked a specific question to false have higher concentrated will than those who linked it to true, the contextual score can flip the answer. the concentration signal breaks the global rank tie

what it lacks

standard inference addressed the true-false problem but left three gaps:

1. no local reconvergence — still uses global cyberank as base, just reweighted. the full tri-kernel reconverges locally given context particles, producing relevance instead of adjusted global rank

2. no honesty mechanism — neurons can vote strategically. serum with valence creates an equilibrium where honest reporting dominates

3. no market correction — incorrect answers persist until neurons manually reweight. ICBS markets suppress false edges economically and continuously

lineage

true-false problem → standard inference → cyber/truth (tri-kernel + BTS + ICBS)