A trading desk that gets smarter after every decision it makes.
Loch Raven Capital: Our own fund, run end to end by AI agents, and the live proof of Continuous Decision Architecture.
Paper portfolio the agents manage end to end, launched June 2026
Of the S&P 500 since launch, graded every trading day
Decisions journaled, graded, and fed back as lessons
Most automation follows fixed rules and quietly goes stale. Forecasts drift, thresholds age, and nobody notices until the numbers hurt. Real systems should measure their own results and improve without waiting for a human to catch the decay.
We made the improvement loop the product. The fund journals every decision, grades it against what actually happened the next day, and tunes its own parameters from the scorecard. It is our Finance and Information Systems training fused into one system, and it is graded in public terms: against the S&P 500, every trading day.
What runs inside the organization
- Four strategy sleeves research, size, and execute decisions with AI agents end to end, inside hard risk guardrails.
- Every decision is journaled with its reasoning, then graded against what actually happened, including what passing would have earned.
- Parameters tune themselves from the scorecard, so the system grows careful exactly where it has been wrong.
Decide
The system makes or drafts the call inside guardrails you set: risk caps, circuit breakers, human-visible journals.
Record
Every decision is journaled with the reasoning behind it, not just the outcome. Nothing important lives in someone’s memory.
Grade
Each call is scored against what actually happened, including the counterfactual: what would have happened if you had passed.
Adjust
Thresholds tune themselves from the scorecard. The system becomes more careful exactly where it has been wrong.
What a build like this delivers
- The Decide, Record, Grade, Adjust loop installed on a decision your business makes weekly
- A decision journal your team can audit
- A scorecard showing hit rate over time
- Self-tuning thresholds with human-set guardrails
- 3-5 weeks for a first learning loop on your data
- Timeline
- $2K-$5K
- Pilot sprint
- Standard
- Complexity
Every engagement starts with a fixed-scope pilot sprint. Larger builds are scoped after the pilot proves itself.
What this means for you
You are not buying a trading bot. You are seeing the operating loop we install on the decisions your organization repeats every week: budgets, forecasts, pricing, staffing. The same loop, pointed at your numbers.
See if this fits your team