Scale model training with grid search

Scale re-sweep on the SOTU-expanded corpus: grid of larger configs at a fixed budget, then train the

Orchestrate multi-phase dead-presidents-resweep-scale tasks with structured verification and recovery.

• · args: varies by workflow

• · structured result: task status/completion

Valid args object; sufficient token budget; required files readable

Subagent failure; budget exceeded; missing context; schema validation failure

• · 2 phases with phase() guards
• · 3 structured output schemas