Autonomous Agents, Tool-Calling Definitions & Workflow Chains
Quality 97/100

Parallel Sectioning with Deterministic Aggregation

Split a task into independent sections evaluated concurrently, then aggregate results in code rather than in a model call.

A fan-out plan with one prompt focus per section, the deterministic aggregation rule, and a conflict-resolution policy including vote thresholds where used.

Template

Role

You are designing a concurrent evaluation stage for a workflow.

Task

For {{task_description}}, define the fan-out over {{section_list}}. Give each section its own prompt focus from {{per_section_prompt_focus}} so no section is asked to weigh another's concern. Where the same section is evaluated repeatedly for confidence, apply the voting threshold {{voting_threshold}}. Specify the aggregation step as deterministic code using {{aggregation_rule}}, and resolve disagreement with {{conflict_policy}}. State explicitly which sections are independent and therefore safe to run concurrently.

Context

Models attend better when each call carries one consideration; concurrency then buys latency, and code-side aggregation keeps the combination step auditable.

Inputs

  • {{task_description}}
  • {{section_list}}
  • {{per_section_prompt_focus}}
  • {{aggregation_rule}}
  • {{voting_threshold}}
  • {{conflict_policy}}

Constraints

  • No section prompt may depend on another section's output
  • Aggregation must be deterministic and reproducible
  • Vote thresholds must be stated numerically
  • Conflicts must never be resolved by silent last-write-wins

Output Format

Markdown: section table (section, prompt focus, independence), aggregation pseudocode, conflict policy.

Quality Criteria

  • Independence claim is defensible for every section
  • Aggregation rule computable without a model
  • Threshold tuned to the false-positive tolerance
  • Conflict policy covers ties
advanced
aggregation
guardrail_split
multi-step-workflow-chains
parallelisation
sectioning
voting