Configuration

The pipeline is driven by a single YAML file passed to load_config().

Minimal Example

storage_url: ./data/measure

ALLOCATE:
  budget: 100000
  rule: minimax_regret

INITIATIVES:
  - initiative_id: my-initiative
    cost_to_scale: 15000
    measure_config: configs/my-initiative.yaml

Top-Level Parameters

Key	Type	Default	Description
`max_workers`	int	4	Thread pool size for parallel fan-out stages
`storage_url`	str	required	Directory where MEASURE writes job results

`ALLOCATE` Block

Key	Type	Default	Description
`budget`	float	required	Total budget constraint for portfolio selection
`rule`	str	required	Decision rule: `minimax_regret` or `bayesian`

`INITIATIVES` List

Each entry configures one initiative:

Key	Type	Default	Description
`initiative_id`	str	required	Unique identifier (alphanumeric, hyphens, underscores)
`cost_to_scale`	float	required	Budget required to deploy this initiative to production
`measure_config`	str	`""`	Path to the initiative’s measure YAML (resolved relative to this file)
`evaluate_strategy`	str	`"score"`	Confidence strategy: `"score"` (deterministic) or `"review"` (LLM-powered)

Path Resolution

measure_config paths are resolved relative to the orchestrator config file, so a layout like this works without absolute paths:

config.yaml
configs/
  my-initiative.yaml
  another-initiative.yaml