ds-subagent-driven-analysis

Execute a data analysis plan by dispatching a fresh subagent per independent task. This approach enforces methodology validation and reproducibility checks for each step, ensuring conclusions match outputs while maintaining runtime observability without polluting notebook outputs.

Isolate SQL extract steps and notebook sections by dispatching fresh subagents for each independent task scope. This preserves plan context and prevents stale workflow states from impacting subsequent analysis steps within the same session.

Review escalation rules trigger a methodology compliance reviewer for standard or strict tasks, followed by a reproducibility reviewer. This dual-escalation workflow confirms reruns, parameter clarity, and notebook self-containment for A/B experiment validations and robustness checks.

Control review depth by escalating to methodology and reproducibility reviewers only when tasks impact final conclusions or trigger validation risk signals. This budgeted approach applies focused review effort to standard, strict, and risk-sensitive analysis steps.

Prevent progress spam in saved notebook outputs by using non-polluting progress signals. This maintains runtime observability during subagent task execution while ensuring the final notebook remains clean and self-contained for reproducibility verification.

Use subagent-driven analysis when your plan contains mostly independent tasks like SQL extracts, metric validations, or memo-style interpretations. It prevents stale context reuse by assigning fresh workers per task while enforcing methodology-before-reproducibility review escalations.