◆ Autonomous Monitor
Reconstructs every order against the prevailing market, scores it on the best-ex policy, and assembles the regulator-ready evidence pack per trade. Flags the outliers that executed outside the policy band for an oversight agent to review.
Memory
Working The order timeline and the matched market data.
Semantic Best-execution policy, RTS 27/28 requirements, venue taxonomy.
Store Execution warehouse + benchmark store
Orchestration
MCP
Harness · Managed Agents … sandboxed code execution for TCA computation.
Tools
{ } Execution + market-data warehouse API ›_ TCA compute sandbox Code exec ⌕ Best-ex policy corpus Retrieval
Evals & guardrails
- Benchmark-reconstruction accuracy checked against a labelled sample.
- Outlier flags reviewed by an oversight agent-as-judge before they reach a client or regulator.
Frontier edge
- ▲Causal reasoning: separates genuine bad execution from coincident market moves, so an outlier flag means cause not just correlation.
- ▲Proactive surveillance: surfaces a drifting venue or algo before it shows up in the quarterly best-ex numbers.
- ▲Continual learning: refines its benchmark-reconstruction logic from each labelled correction, eval-gated before it changes scoring.
In numbers
190,000
Orders analysed / day
< 1 min
Best-ex evidence lag
Handoffs
Across ⇢ Compliance → Trade Surveillance for outlier review