RAND Forecasting Initiative (RFI)

SDK that provides access to the RAND Forecasting Initiative crowdsourced forecasting platform (formerly INFER, powered by Cultivate Labs). It covers policy-relevant forecasting questions with aggregated crowd probabilities and individual forecaster predictions with rationales.

Quick Start

python
1from sdk_rfi import Client
2
3client = Client()  # Uses RFI_EMAIL and RFI_PASSWORD env vars

Key Methods

Data Coverage

• Domain: Politics / policy forecasting
• Countries/Regions: Global (US policy focus)
• Time range: Questions from ~2020 to present
• Update frequency: Event-driven (new questions published, forecasts updated continuously)
• Key data: Forecasting questions on geopolitics, national security, science/technology policy, economics, biosecurity, nuclear risk, AI governance

Forecasting Patterns

• Crowd forecast as base rate: Use client.questions.list() to find questions matching your topic, then read the crowd probability from question.answers[i].probability as a calibrated starting point.
• Trend analysis via prediction history: Use client.prediction_sets.list(question_id=X) to see how individual forecasts changed over time. Plot created_at vs forecasted_probability to detect momentum.
• Expert rationale mining: Individual prediction sets include rationale text explaining the forecaster's reasoning. Use client.prediction_sets.list(question_id=X) and examine rationales for weak signals and arguments.
• Cross-reference with resolution: Use client.questions.list(status="closed") to find resolved questions. Compare crowd probability at various cutoff dates against actual outcomes to measure calibration.
• Comment-based signal detection: Use client.comments.list(commentable_id=X, commentable_type="Forecast::Question") to find discussion threads that may contain links to evidence or emerging developments.

Common Queries

python
1# Get all active forecasting questions
2questions = client.questions.list()
3for q in questions.questions:
4    print(f"{q.id}: {q.name}")
5    for a in (q.answers or []):
6        print(f"  {a.name}: {a.probability_formatted}")
7
8# Get individual forecasts and rationales for a question
9forecasts = client.prediction_sets.list(question_id=1234)
10for ps in forecasts.prediction_sets:
11    print(f"{ps.membership_username}: {ps.rationale}")
12    for pred in (ps.predictions or []):
13        print(f"  Answer {pred.answer_id}: {pred.forecasted_probability:.1%}")
14
15# Backtest: get data as it was available on a past date
16past_questions = client.questions.list(cutoff_date="2025-06-01")
17past_forecasts = client.prediction_sets.list(question_id=1234, cutoff_date="2025-06-01")

Full Method Reference

See references/methods.md for all 4 methods with complete parameter details.