Global accessible-experience map
A stream is not healthy unless every promised accessibility experience is healthy. This is what each audience slice is actually receiving right now.
Accessibility promise registry
Versioned operational contracts. A promise cannot be silently changed once the event has started.
| Promise | Feature | Language | Territories | Tier | Version |
|---|
Error budget
Each accessibility SLO has an error budget for this event tier. The bar and the number carry the same information; colour is never the only signal.
Fault library
The digital twin can be broken in — documented ways. Pick one to run the closed loop against a real symptom.
Event readiness studio
The Accessibility Release Gate proves every promise works before the event is allowed to say “accessibility ready”. Hard assertions block certification.
Live event cockpit
Audience experience, operational telemetry, agent reasoning and control — kept visually and structurally separate so evidence is never confused with action.
Feature health
| Feature | Evaluated | Breaching | Abstained | Worst SLO |
|---|
Delivery environment
Delivery chain logs
| Time | Service | Level | Line |
|---|
Incident workspace
Evidence, hypothesis, policy, action and verified result are separate, labelled stages. The agent may propose; only policy plus a signed approval may act.
Evidence Retrieved through Grafana MCP
Hypothesis Ranked diagnosis
Change correlation
Policy & approval Governed action
Verified result Post-action assertions
| Assertion | Scope | Observed | Objective | Result |
|---|
Interrogate Ask the agent
Question the diagnosis. Every answer cites the evidence it rests on and the Grafana MCP tool that produced it. With the Gemini reasoning plane configured, a question the retrieved evidence cannot settle sends the agent back to Grafana MCP for more, on the same audited path as the investigation. This is a read path: it cannot change scope, policy, actions or verification.
Communications
Post-incident review
Audit trail
| # | Event | From | To | Actor |
|---|
Incident evidence replay
The exact affected interval: what was spoken, what was captioned, and the measured offset between them. This is the screen that turns a number into a decision.
Spoken dialogue (reference)
Rendered captions (delivered)
Agent and MCP observability
Raccord observes the agent it builds. Every Grafana MCP call and every agent step is timed, counted and traced.
Grafana MCP call chain
| # | Tool | Latency | Bytes | Result |
|---|
Capabilities resolved on this server
| Capability | Required | Resolved tool |
|---|
Agent steps
| Agent | Calls | Total time |
|---|
Benchmark laboratory
Every claim Raccord makes is measured over a seeded corpus of controlled accessibility incidents, with ablations.
No results loaded. Run raccord bench --scenarios 1000.