Decision-quality routing across every model.

meta-vengine is the routing engine that decides which model runs which task. It is multi-model by default (Claude, GPT, Gemini, Grok) with explicit cost gates, recovery hooks, and a logged decision-quality (DQ) score on every call. Average DQ across 4,954 routing decisions: 0.889.

The router does not assume one model wins. Hiring Committee runs Tech Lead on Grok by design, because deep technical probing benefits from a different posterior than the Hiring Manager (Claude) and HR Partner (GPT). Each agent gets the model whose decision profile fits the role, not the cheapest one.

Recovery is first-class. The engine ships with a 94% recovery rate against transient model failures and a 99.88% cache hit rate against repeated subqueries. When a model 404s on a stale ID, the router falls back to the canonical pricing.json registry, picks a current model, and continues without dropping the task.

meta-vengine is what lets Career Board enforce the BYO-keys promise. Every model call is routed through the operator's own keys when present, with an explicit fallback to platform keys only on tiers that include them. The receipt records which key paid for which call, every time.