Purpose
Assess whether the evidence behind a proposed intervention is strong enough — and transferable enough — to justify funding or scaling in a specific context. Produces an evidence tier rating, a contextual fit and feasibility assessment, and a clear fund / pilot / decline recommendation, so program officers can allocate limited dollars toward what actually works in the setting where it will run.
When to Use
Invoke this skill when the user asks to:
- Decide whether to fund, scale, or discontinue a public health or social program
- Assess how strong the evidence is behind an intervention or grant proposal
- Compare two or more competing interventions for the same funding pool
- Judge whether evidence from another population or setting will transfer to theirs
- Vet a grantee's claim that their program is "evidence-based" or "proven"
- Build a defensible rationale for a funding decision to present to a board
Inputs Required
Before starting, collect:
- Intervention description: What the program does, its core components, and the causal logic (why it should work)
- Target outcome: The specific health or social outcome the intervention aims to change
- Target population and setting: Who and where — demographics, geography, delivery context (clinic, school, community, digital)
- Available evidence: Studies, evaluations, or data cited in support — include study design and sample size if known
- Implementation context: Who would deliver it, existing infrastructure, and budget available
- Decision at hand: Fund a pilot / scale an existing program / choose between options / renew or cut
- Constraints: Non-negotiables such as timeline, equity requirements, or reporting mandates
If study designs or sample sizes are not provided, infer the likely evidence tier from the description and flag the uncertainty explicitly.
Instructions
Step 1: Classify the Evidence Level (Internal Validity)
Rate the strength of the causal claim using a public-health evidence hierarchy. Assign the intervention to a tier based on the strongest available study, then note the depth of the evidence base behind that tier.
- Tier 1 — Well-supported / Causal: One or more well-conducted RCTs, or a systematic review/meta-analysis of RCTs, showing effects on the target outcome. Look for replication across settings.
- Tier 2 — Supported / Quasi-experimental: Strong quasi-experimental designs (difference-in-differences, regression discontinuity, matched comparison, interrupted time series) with a credible counterfactual.
- Tier 3 — Promising / Observational: Correlational, pre-post without comparison group, cohort, or cross-sectional studies. Directional signal but confounding is uncontrolled.
- Tier 4 — Theory-based / Emerging: Logic model or theory of change grounded in established mechanisms, but no direct outcome evaluation of this intervention.
Reference the tiering logic used by established frameworks (e.g., the Prevention Research Center / CDC evidence continuum, the Cochrane risk-of-bias approach, Title IV-E and ESSA evidence tiers, and the What Works Clearinghouse standards). For the assigned tier, note: number of studies, effect sizes if reported, and any risk-of-bias concerns (attrition, selection, publication bias, single-site dependence).
Step 2: Assess Contextual Fit (External Validity)
Evidence strength is not portable by default. Score how well the evidence transfers to the user's population and setting.
- Population match: How similar is the studied population to the target population (age, risk profile, cultural context, baseline conditions)?
- Setting match: Was the evidence generated in a comparable delivery context (research clinic vs. real-world community, high- vs. low-resource)?
- Mechanism plausibility: Is the causal mechanism likely to operate the same way in the new context, or does it depend on conditions that won't be present?
- Effect size realism: Real-world effects typically shrink relative to efficacy trials — estimate expected attenuation.
Rate contextual fit as Strong / Moderate / Weak with one sentence per dimension.
Step 3: Assess Implementation Feasibility
Even strong, transferable evidence fails without capacity to deliver with fidelity.
- Fidelity requirements: How complex is the intervention, and how tightly must it be delivered to reproduce results? Flag interventions where effects depend on specialized staff or precise dosage.
- Delivery capacity: Does the grantee/setting have the staff, training, and infrastructure needed?
- Cost realism: Is the available budget consistent with the cost-per-participant implied by the evidence?
- Scalability: Does the model hold up at the intended scale, or was it only tested small?
- Sustainability: What happens to outcomes when grant funding ends?
Rate feasibility as High / Medium / Low with the single largest implementation risk named explicitly.
Step 4: Composite Rating and Recommendation
Combine the three dimensions into a decision. Do not average blindly — weight by the decision at hand.
- Produce a composite label: Fund with confidence / Fund as monitored pilot / Fund only with conditions / Decline.
- State the reasoning: which dimension drives the recommendation and why.
- If declining or conditioning, specify what evidence or capacity would change the answer.
Step 5: Evidence Gaps and Questions to Resolve
List the specific unknowns that should be answered before or during funding — framed as questions the program officer can put to the grantee or a built-in evaluation requirement (e.g., "commit to a matched-comparison evaluation in year one").
Output Format
- Structure: Scored assessment across three dimensions, then a composite recommendation, then gaps to resolve
- Length: 700–1,000 words
- Format: Structured document with a summary scorecard table (Evidence Tier / Contextual Fit / Feasibility / Composite) followed by prose sections
- Sections: Summary Scorecard → Evidence Level Analysis → Contextual Fit → Implementation Feasibility → Recommendation → Questions to Resolve Before Funding
- Tone: Rigorous and impartial — like an evaluation officer who respects the mission but will not overstate weak evidence. Direct about limitations.
Quality Checklist
Before delivering output, verify:
- [ ] The evidence tier is assigned to a specific level (1–4) and names the study design(s) that justify it
- [ ] At least one named evidence framework is referenced in the tiering rationale
- [ ] Contextual fit is scored across all four dimensions (population, setting, mechanism, effect size)
- [ ] Feasibility names the single largest implementation risk explicitly
- [ ] The recommendation is a clear decision label — not a hedge — and states which dimension drives it
- [ ] Questions to resolve are specific and answerable, not generic ("more research needed")
Notes
- The most common error in funding decisions is treating "evidence-based" as binary; the tier and the contextual fit together determine whether evidence actually applies here.
- A Tier 1 intervention with weak contextual fit is often a worse bet than a Tier 2 intervention designed for the exact population — say so when the analysis supports it.
- When comparing multiple interventions, apply the same three-dimension scoring to each and present a side-by-side scorecard so the tradeoffs are visible.
- Watch for single-site and single-team evidence: strong effects that have never been replicated by an independent team are a known scaling risk.
How to use this skill
This file is a Claude Code skill in Anthropic's standard format — markdown with YAML frontmatter at the top.
Installing in Claude Code:
~/.claude/skills/<folder-name>/SKILL.md— auto-loads in every session anywhere on your machine (recommended for skills you'll use across multiple projects)<your-project>/.claude/skills/<folder-name>/SKILL.md— auto-loads only when working in that project (recommended for project-specific skills)
Restart Claude Code or open a new session to register.
Using in Claude.ai, ChatGPT, Gemini, or other AI tools:
- Drag this file into a new chat, or copy and paste its contents
- Send your task or request as your next message — the AI will apply the skill where it fits
For Claude Code skill details: https://docs.anthropic.com/en/docs/claude-code/skills