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.
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