Intervention evaluation worksheet

A lightweight, evidence-first worksheet for deciding what to test or support. Version 1.1 (2026-10-13). It is a prompt for disciplined reasoning, not a scoring model: do not add the fields into a single number unless the weights and value judgments are explicit.

1. Decision

• Decision owner / date:
• Decision to make: (fund, run, scale, stop, or learn more)
• Alternatives, including doing nothing:
• Time horizon and constraints:
• Welfare scope and moral assumptions: Whose welfare counts, how are interests compared across groups/species, and what welfare conception or metric (if any) is being used? State exclusions and value judgments rather than hiding them.

2. Theory of change

• Who is affected? Specify population, location, and baseline conditions.
• What changes for them, and through what mechanism?
• Key assumptions / causal links:
• What would falsify this explanation?

3. Evidence and uncertainty

For each important claim, record the source, date, population, method, and limitations. Separate direct observations from extrapolation and opinion. Name the study design/source type (e.g., randomized, quasi-experimental, observational, systematic review, expert judgment); use labels as descriptions, not a universal quality ranking. Strength depends on the question, execution, and external validity. Distinguish primary data from secondary synthesis.

| Claim | Source (link, type, access date) | Applicability | Confidence and why | Likely direction of error / why | Missing evidence | |---|---|---|---|---| | | | | | |

• What is the strongest counterevidence or plausible alternative explanation?
• Are outcomes measured in a way that captures who benefits and who bears costs? Who is missing from the evidence or consultation, especially people with relevant lived or stakeholder expertise?

4. Expected effects

Describe plausible outcomes for each affected group, including magnitude, probability, duration, and reversibility. Use ranges/scenarios where point estimates would imply false precision.

For each scenario, make the counterfactual explicit: what likely happens to each group without this intervention, including other actions or market responses? Ask explicitly whether the intervention could replace or reduce a complementary program (funding, staff, access, or political attention), and who would lose as a result.

Group	Benefit	Harm / opportunity cost	Likelihood / range	Evidence

Account for distribution (not just totals), displacement/substitution, unintended effects, and effects on nonhuman sentient beings. Avoid double-counting correlated benefits. State the welfare scope and moral assumptions from section 1; do not silently omit affected groups.

5. Safeguards and learning

• Potential severe or irreversible harms:
• Reversibility / exit costs: What would stopping or reversing the intervention cost, and who bears that cost?
• Delivery feasibility: Capacity, staffing, costs, dependencies, timeline, and main implementation failure modes.
• Relevant expertise / consultation: Whose lived or stakeholder expertise is needed, and who has not been consulted or is excluded?
• Affected but unvoiced: Who is affected but cannot represent their interests in this process (e.g., nonhuman animals, future people, distant populations)?
• Consent, privacy, dignity, and fairness considerations:
• Stop conditions / escalation plan:
• Smallest useful test: What can be learned before scaling?
• Outcome measures and disaggregation:
• Review date and who will review results:

6. Recommendation

State the choice, strongest reason, key uncertainty, what evidence would change the choice, and a calibrated confidence range (or explain why one is not meaningful). Record what evidence would have changed the decision before reviewing outcome data, then date the entry. Distinguish empirical uncertainty from value judgments. Include dissenting views and document assumptions so another person can reproduce the reasoning.

Use notes

• Prefer primary evidence and independent replication when feasible; check whether the studied context matches the intended one.
• A lack of evidence is not evidence of no effect. Conversely, plausible mechanisms are not proof of impact.
• Do not use this worksheet as a substitute for domain expertise, stakeholder consultation, or safety review.
• Preserve source links and access dates; update conclusions when evidence or context changes.
• For repeated use, record the worksheet version and date; maintain a short changelog when changing the template. Where practical, date and preserve initial predictions before outcomes are known.

Changelog

• 1.1 (2026-10-13): Explicitly prompt for displacement of complementary programs, in addition to other counterfactuals.
• 1.0 (2026-10-13): Initial reviewed worksheet.