The LVN initiative has produced a white paper, an evidence base, and a protocol specification. But specifications don't solve problems — people do. These ten provocations represent the questions where we need the most help, framed as genuine debates where reasonable people can disagree. Each one could change the architecture. We're looking for the strongest arguments, not consensus.
The Signal Exchange Protocol converts private context into abstract signal vectors using a learned embedding model. But who trains and controls that model determines who controls what "complementarity" means — and therefore who gets matched with whom.
Zero-knowledge proofs work beautifully for exact attribute matching ("prove you're over 18"). But the LVN needs fuzzy, semantic matching ("find someone whose situation complements mine in ways I can't articulate"). Can this be done under encryption with current techniques?
The LVN's core promise is matching based on context people are already sharing with their AI assistant. But a system that understands your needs well enough to match you is also a system that understands your vulnerabilities well enough to exploit you. Is there a meaningful distinction?
The Chetty et al. data shows that the people who most benefit from cross-boundary connections are the least likely to have them. The LVN promises to fix this — but technology adoption follows power law distributions. The early adopters will be educated, tech-savvy, already-networked people.
The white paper commits to "pluralism of value" — accommodating gifts, reciprocity, barter, and transactions. But when a match surfaces between a retired accountant and a freelancer who needs tax help, who decides whether this is a favor, a mentorship, or a $200 consultation?
The Trust & Reputation Fabric (Layer 4) flows trust through relational chains. But relational trust networks are homophilous — people trust people like themselves. If trust bootstraps from existing social networks, it inherits their biases.
The LVN insists it's a protocol layer, not a platform. But protocols need implementations, implementations need infrastructure, infrastructure needs funding, and funding creates power. Email is a protocol; Gmail dominates it. HTTP is a protocol; a handful of companies control the web.
The LVN assumes AI assistants will implement cooperative protocols. But Claude is built by Anthropic. GPT by OpenAI. Gemini by Google. These companies compete. Their incentive is to keep user context proprietary — it's their moat. Why would they implement a protocol that shares it?
The LVN thesis is that a massive amount of cooperative potential goes unrealized due to informational barriers. But economists might counter: if the value were really there, markets would have found it. TaskRabbit, Nextdoor, and LinkedIn exist precisely to reduce cooperation friction. Maybe the "gap" is just cooperation that isn't actually worth the coordination cost.
The LVN needs a first pilot community. The wrong choice wastes years. The right choice proves the concept and generates the data needed to refine the protocol. The community must be small enough to iterate, diverse enough to test cross-boundary matching, and trust-rich enough to tolerate early failures.