Usage

The orchestrator runs the full MEASURE → EVALUATE → ALLOCATE → SCALE pipeline from a single config file.

Quickstart

Step 1 — Write a pipeline config:

# pipeline.yaml
storage_url: ./data/measure

ALLOCATE:
  budget: 100000
  rule: minimax_regret

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

  - initiative_id: initiative-b
    cost_to_scale: 40000
    measure_config: configs/initiative-b.yaml

Step 2 — Run the pipeline:

from impact_engine_orchestrator import run_pipeline

results = run_pipeline("pipeline.yaml")

Step 3 — Read the results:

# Per-initiative pilot measurements
for pilot in results["pilot_results"]:
    print(pilot["initiative_id"], pilot["effect_estimate"], pilot["p_value"])

# Portfolio allocation decision
allocation = results["allocate_result"]
print(allocation["selected_initiatives"])
print(allocation["budget_allocated"])

# Outcome reports (predicted vs. actual at scale)
for report in results["outcome_reports"]:
    print(report["initiative_id"], report["predicted_return"], report["actual_return"])

Output Structure

run_pipeline() returns a dict with five keys:

Key	Type	Description
`pilot_results`	list[dict]	MEASURE results for all initiatives (pilot phase)
`evaluate_results`	list[dict]	EVALUATE confidence scores for all initiatives
`allocate_result`	dict	ALLOCATE decision: selected initiatives, budget, predicted returns
`scale_results`	list[dict]	MEASURE results for selected initiatives (scale phase)
`outcome_reports`	list[dict]	Predicted vs. actual returns for selected initiatives

See the API Reference for the full field list per stage.

Evaluate Strategy

Each initiative can use a different confidence strategy:

INITIATIVES:
  - initiative_id: initiative-a
    evaluate_strategy: score    # fast deterministic scoring (default)
    measure_config: configs/initiative-a.yaml

  - initiative_id: initiative-b
    evaluate_strategy: review   # LLM-powered review via evaluate backend
    measure_config: configs/initiative-b.yaml

See Configuration for the full parameter reference.