Tier 14 — Live Intelligence Dashboard

What the router is actually doing, right now

Real provider health, real composite score distribution, real confidence trends, and real safety bounds activations — pulled live from the same intelligence layer that routes every real decision. No mock data, no placeholders: every number below is read directly from production.

Weight provenance loading

The composite scoring weights currently in effect — baseline unless the adaptive weight engine has genuinely earned the right to adjust them.

Loading real weight data…

Provider health real telemetry

Every real provider this platform has actually used, with its live health score and recent history. Health is computed from real latency stability, uptime, breaker state, recency, and consistency — never estimated.

Loading real provider data…

Composite score distribution last 1,000 decisions

How real routing decisions have actually scored, across every real agent and buyer this session.

Loading…

Confidence trend over time

Real confidence values for real decisions, in the order they actually happened.

Loading…

Safety bounds activations real gating events

How often each real safety gate has actually fired across recent decisions — not a theoretical limit, a lived one.

Loading…

Buyer intelligence methodology

Buyer intelligence adjusts routing per-buyer based on that buyer's own real historical behavior — but it is inherently private to each buyer.

Why no numbers appear here: showing any real buyer's actual latency tolerance, spend pattern, or usage history on a public page would expose that buyer's private behavior. This dashboard is public and read-only by design, so buyer intelligence is documented conceptually here rather than displayed per-buyer. A specific buyer's own real context is visible to them via their own API responses.

How it works

latency_tolerance and budget_sensitivity are derived from a buyer's own real chain history — not a stated preference, since none exists in this system, but an honest read of what that buyer's calls have actually cost and taken. A buyer whose real history shows fast, cheap calls gets latency and cost weighted more heavily in their own routing; a buyer with a longer real track record of slower or costlier calls does not get penalized for it.

If a buyer's real sample size is below … chains, this re-weighting is suppressed entirely and falls back to neutral — a real safety bound, not just a reported number.