Principle:Ucbepic Docetl Baseline Pipeline Authoring
| Knowledge Sources | |
|---|---|
| Domains | Data_Engineering, Optimization |
| Last Updated | 2026-02-08 01:40 GMT |
Overview
A pipeline design principle that establishes an initial unoptimized pipeline as the baseline for subsequent automated optimization.
Description
Baseline Pipeline Authoring is the process of creating a first-draft pipeline with operations marked for optimization. The baseline serves as the starting point that optimization algorithms will improve upon. Key practices include:
- Writing clear prompts for each LLM operation
- Marking operations with optimize: True to signal the optimizer
- Establishing expected output schemas
- Running the baseline to get initial accuracy and cost metrics
The baseline provides the reference point against which optimized variants are measured.
Usage
Use this principle at the start of the optimization workflow. Before running any optimizer, you need a working baseline pipeline that produces correct (if suboptimal) results. This baseline will be automatically rewritten by the V1 or MOAR optimizer.
Theoretical Basis
Baseline establishment follows an iterative improvement model:
- Initial Design: Write operations with simple prompts
- Execution: Run baseline to establish cost and accuracy metrics
- Annotation: Mark operations eligible for optimization
- Optimization: Feed baseline to automated optimization search