Portable Analytic eXpertise — a structured format for domain knowledge that AI agents can reason over without bluffing. Constructs, findings, propositions, sources, playbooks. All traceable. All composable across domains.
General-purpose AI sounds confident on every topic. It is rarely calibrated and almost never citable. For research, policy, and rigorous analysis, that is not enough — you need to know which paper a claim came from, what the effect size was, what the sample looked like.
Pax fixes the input side. Hand an agent a pax for a domain and it now has the field's evidence in front of it: real findings with real sources, real constructs with real definitions. The agent can synthesize across pax because constructs are the shared vocabulary. No bespoke glue. No retrieval gymnastics. The agent reasons with the literature, not the model's vibes.
Every finding links to a source. Every claim is falsifiable. We never ship summary prose you cannot verify.
Constructs are the atoms of reasoning. Shared across pax, they become bridges. Agents synthesize without glue code.
Effect sizes, confidence intervals, p-values, sample sizes — not just direction. Strength labels are policed by editors, not by hope.
Open registry. Pax licenses set per pax. Forks are blessed. Mirrors are encouraged.
Pax are folders, not databases. No lock-in. Compose at the construct level, run where you want.
An MCP-server runtime (coming soon) will let agents query, align, and reason over your pax like a research assistant.