Research methods

The repository implements research and model-risk boundaries for experiment identity, trial registration, point-in-time controls, uncertainty analysis and AI observations without granting research or AI order authority; independent model validation remains outstanding.

Pending approval

Identifier: CLM-012
Scope: Repository research-control implementation only.
Evidence date: 2026-08-09
Owner role: model-validator
Review / expiry date: 2026-10-31
Residual risk: No alpha, profitability, capacity or predictive-validity claim is approved. Data quality, dependence, regimes, costs and strategy-specific validation remain unresolved.

Evidence sources

Limitations

Do not infer investment suitability or model approval from the presence of research-control code.
Do not allow research or AI to size, approve, route or modify an order or risk limit.

Explanatory research governance

The following methods are explanatory context governed by CLM-012; they are not an approved performance statement.

Research code is isolated from production code and cannot access credentials, risk-limit storage or execution networks.
Research models emit intents, not orders. No AI-generated order authority is permitted.
Experiment manifests record hypothesis, data, parameters, environment, date and results.
Results are reported with commit, command, environment, coverage limits and date.

Uncertainty and model risk

Model risk is treated as a first-class concern. No performance claim (alpha, Sharpe, capacity, profitability or predictive validity) may be published without a registered experiment, point-in-time data, cost/capacity analysis, uncertainty quantification and independent validation.

Backtests are not forecasts or guarantees.
Simulator results are clearly distinguished from observed results.
Strategy parameters are not exposed in public exports.
Model validation requires an independent model-validator approval.

Claim classification

Claims are classified as legal requirement, supervisory expectation, voluntary standard, internal policy, capability or performance. Each classification has distinct approval owners and publication rules. Performance claims are prohibited until independently validated.