NEXUS is the calibrated AI substrate that the rest of Aequara runs on. Multi-LLM panel routing across OpenAI, Anthropic, Google, DeepSeek and local models. Brier-tracked dispatch logging. Methodology hash custody. The α'' workload classifier and β'' execution-surface classifier — one operating layer for AI work that has consequences.
Every dispatch is classified across 12 workload axes (code-cheap, code-novel, long-ctx, reasoning-batch, trivial-classify, etc.). Each axis routes to the substrate that wins on the 4-Criterion Pareto check — cheapest, most productive, most efficient, most suitable.
The complement to α'': where α'' picks WHICH substrate, β'' picks WHICH execution surface (HTTP API · MCP server · file edit · CLI · subprocess · browser-headless). Together they form the smallest-possible-substrate discipline.
Every dispatch logs to llm-learnings.jsonl. Every prediction registers as a Brier score with stated horizon. Every error gets published. The substrate gets cheaper, faster, and more accurate as time progresses — never plateaus.
NPX-100 publishes through NEXUS. LUMEN reads from NEXUS. CIPHER stores NEXUS pre-publication research. AEQUARA Craft teaches the NEXUS craft. COMPASS is built on NEXUS for creators. NEXUS is not a product you buy — it's the operating layer the rest of Aequara is built on.
The financial-data and ratings industry runs on platforms whose incentives diverge from the people they serve. Bloomberg pricing is licensed by issuers. Rating-agency methodology drifts silently. Index providers refuse to publish their constituent decisions. NEXUS is built on the opposite stack: methodology hash stable, anti-issuer-pay, every dispatch logged, every error published, every classifier Brier-tracked.
NEXUS is what the next decade of fiduciary measurement runs on. The mic, occasionally, drops.
"After hours, the calibration continues."