a neuron is truthful when its cyberlinks report its actual private beliefs — not adjusted for social pressure, predicted popularity, or gaming the reward signal
a truthful link: the neuron creates the connection because it genuinely believes it reflects reality, stakes according to that conviction, and sets valence to match its honest prediction of where the ICBS market will settle
truthfulness in mechanism design
in mechanism design, a protocol is truthful (dominant strategy incentive compatible, DSIC) when honest reporting is a dominant strategy — the best response regardless of what others do. this is stronger than a Nash equilibrium, where honesty is optimal only given that others are also honest.
Bayesian Truth Serum achieves Bayes-Nash equilibrium truthfulness: honest reporting is optimal when the neuron believes others will also report honestly. whether the full veritas protocol achieves the stronger DSIC property is an open question — see cyber/epistemology §6.1.
truthfulness and syntropy
a truthful link increases syntropy: the cyberlink sharpens the collective picture, reducing uncertainty ($D_{KL}(\pi^*_{\text{after}} \| u) > D_{KL}(\pi^*_{\text{before}} \| u)$). a spammy or false link decreases syntropy — it moves $\pi^*$ toward noise.
karma is the accumulated truthfulness record: the running sum of BTS scores across all a neuron's links. high karma means a consistent track record of signal over noise. karma enters effective adjacency as the multiplier $\kappa(\nu)$, making past truthfulness a structural property of current influence.
the incentive structure
Bayesian Truth Serum makes truthfulness rational:
- inflating valence toward predicted popularity loses information gain (surprise drops to zero once the neuron has predicted its own position)
- setting valence contrarian without genuine signal loses prediction accuracy
- the unique score-maximizing strategy is accurate reporting of both belief (link + stake) and meta-belief (valence)
over time: truthful neurons earn stake from noise producers, accumulating influence. noise producers lose stake to truthful neurons, losing influence. the graph self-selects toward truthful contributors in proportion to epistemic accuracy.
truthfulness and trust
truthfulness is a property of track record, not individual acts. a single truthful link earns a positive BTS score. systematic truthfulness earns high karma. high karma is the formal analog of trust: the network has observed that this neuron's private signals are genuine.
trust is not agreement. a truthful neuron can consistently disagree with the majority — setting $v = -1$ on links others rate highly — and earn high karma if its contrarian predictions repeatedly prove accurate.
see truth for the probabilistic truth signal. see truth model for the two-layer structure. see valence for the ternary field that carries the honesty signal. see Bayesian Truth Serum for the scoring formula. see karma for the accumulated truthfulness record.