The team wants to turn a quant strategy into a real product: backtests must be reproducible, definitions must align with the fund contract, and every day after launch must reconcile — traditionally months of work for a research team plus a middle-and-back office.
Have the CIO Agent and the engineering Agent take a quant strategy from idea to live trading: pin down a reproducible spec, finish a twelve-year historical backtest the same day, sweep parameter tiers systematically, align the full material set with the fund contract's compliance definitions, and take over daily operations after launch. Humans only raise hypotheses, ask questions, and make the call.
Writes the reproducible spec, research reports, and external materials, independently re-checks engineering output, and guards definition consistency.
Implements the backtest engine, runs parameter sweeps, builds the post-close data pipeline and dashboards.
Never looks at the original implementation; rebuilds results from the spec alone — any mismatch is a finding.
We're turning a quant strategy into a real product. Rules:
@cio writes the reproducible spec first: asset buckets, risk budgets, rebalancing rules — every definition pinned down.
@quant-eng implements the engine to the spec and finishes the twelve-year historical backtest the same day, verified item by item before review.
A systematic sweep across parameter tiers finds a better configuration; a human signs off on switching the flagship and freezing it.
The full material set aligns with the fund contract: risk-limit constraints and disclosures pass line by line, with documents updated alongside every definition change.
After launch, settlement files flow in daily; the Agent reconciles both ways to produce the daily NAV, dashboards update in sync, and missing artifacts alert automatically.
Settlement inflow, NAV roll-forward, dashboard updates; missing required artifacts alert automatically.
Rebalance days run from a pre-written playbook, with measured costs archived.
@cio drafts the monthly report; a human gives final review before it's archived and published.
Freeze the pre-launch checks into a gate checklist that every product change runs through automatically.
Build a contingency matrix for tail scenarios: which signal, which leg to cut, who decides, within what window.
Expanding the asset pool or layering enhancements goes through the same pre-registration process first.