# Instructor guide: DIRD claim-provenance and forecast audit

Course: DIRD critical-reading seminar

Suggested time: 75–100 minutes

## Learning target

Researchers reconstruct provenance, separate source from inference, and score dated forecasts without personality-based deference or hindsight leakage.

## Prepare

- Freeze a document and evidence cutoff.
- Define direct, inferred, contextual, and contradictory edges.
- Prepare one missing-primary-source case.

## Facilitation moves

- Ask what the source actually demonstrates.
- Require dates on forecasts and outcomes.
- Assign an adversarial reviewer to attack the highest-leverage dependency.

## Misconception checks

- **A government-commissioned document establishes every claim it discusses.** Commissioning establishes provenance and purpose, not independent experimental confirmation.
- **A correct later interpretation validates an earlier vague forecast.** Forecast scoring requires preregistered specificity and a fixed evidence cutoff.

## Accessibility and participation

- Provide the graph as a sortable edge table.
- Use edge labels and shapes, not color alone.
- Allow oral provenance defense with a written citation map.

## Evidence of learning

- A complete provenance chain
- A cutoff-safe forecast score
- An adversarially revised evidence tier

## Extension

Publish a machine-readable adversarial dossier with disputed edges, source snapshots, and versioned confidence changes.

## Evidence boundary

Assess the learner's reasoning only within the declared model and recorded observations. Do not upgrade a simulation result into a claim about an unmodeled physical system.