Risk-On/Risk-Off Switching With the VaR Control Room

Architecture overview

Every agent emits VaR snapshots into Redis and SQLite. ErrorWatcher polls them, tags breaches, opens PagerDuty tickets, and sends Slack alerts. Ops can view the same feed from `/console`.

Switching rules

When VaR exceeds a desk-defined threshold, the Boss role issues `MODE=OBSERVE`. Executor will only hedge or flatten. Analyst queues prompts but does not execute until VaR cools. This automation triggered twice during that documented window and is recorded in the logs.

Governance checklist

Re-enabling trading requires: Ops acknowledgment, ticket closure, and prompt diff ID. These steps are enforced automatically so nobody can silently flip back to live mode.

Implementation

We provide the YAML policy plus sample alerts (SES, Slack, Discord). Drop them into your repo, restart `hyperagent-brain`, and you inherit the same VaR guardrails that protect our desks.

Reporting

VaR history flows into `/resources/live-pnl` so investors see performance and risk in one view. That transparency is what lets LPs replay the logs independently rather than take any number on faith.