Test the data and the failure modes before the strategy story.

Prerequisites

• A lawfully obtained historical dataset with a data dictionary, timezone, corporate-action policy, and license terms.
• A research hypothesis written before testing, including universe, signal timing, holding period, benchmark, and failure criteria.
• An isolated environment with no broker credentials, order routes, or production market-data secrets.
• A qualified human reviewer for the relevant market, regulation, tax, and risk context.

Preview the research kit

git clone --depth 1 --branch main https://github.com/aAAaqwq/AGI-Super-Team.git
cd AGI-Super-Team
./install.sh --source "$PWD" --destination /path/to/review-workspace quant-trader

Verify that the preview proposes the CQO, CDO, and CFO workspaces. Apply to a disposable destination only after inspecting required local content. Do not add exchange keys, brokerage credentials, or private account statements to these workspaces.

Use a reproducible research sequence

1. Profile the dataCDO checks missingness, duplicate timestamps, survivorship bias, point-in-time availability, corporate actions, and licensing constraints.
2. Freeze the hypothesisRecord the signal and failure criteria before exploring parameter combinations. Separate discovery data from evaluation data.
3. Backtest honestlyCQO includes fees, spread, slippage, latency, turnover, liquidity, and a walk-forward or held-out evaluation.
4. Stress the resultTest regime changes, parameter sensitivity, concentration, stale prices, capacity, and the effect of removing the best periods.
5. Review riskCFO challenges drawdown, leverage, financing, liquidity, operational controls, and whether the evidence supports any next step.

Minimum research receipt

dataset: identifier, license, retrieval date, immutable hash
universe: inclusion rules and point-in-time membership
signal: formula, timestamp, availability lag, parameters
execution: fees, spread, slippage, latency, position limits
evaluation: train/test split, benchmark, sensitivity, failures
result: returns and drawdowns with uncertainty and caveats
next_step: reject, revise, or human-approved paper evaluation

Keep code, configuration, and the immutable input hash together. A chart without the dataset contract and execution assumptions is not enough to reproduce the result.

Red flags that stop the study

Stop when the data license is unclear, timestamps allow future information, delisted assets disappear from history, transaction costs are omitted, results depend on one narrow parameter, or the only evidence is an in-sample equity curve. Do not repair a failed hypothesis by repeatedly changing it against the held-out period.

Limitations

Historical simulation cannot establish future returns. Market structure, liquidity, regulation, taxes, borrow availability, and execution quality can change. AI-generated analysis can contain mathematical, coding, or data errors. Independent reproduction and professional review are required before any decision involving real money.

Rollback and uninstall

Revoke any temporary data access, delete derived data when its license requires it, and preserve only permitted research receipts. Remove the newly created CQO, CDO, and CFO workspace directories listed in the preview. If credentials were added despite this guide, rotate them immediately rather than relying on file deletion.