Evaluating Ukraine’s
Reconstruction Projects
A proposed experiment on presentation effects and AI-assisted evidence preparation.
The question
When the facts of a civilian reconstruction project stay fixed, does presentation change human reviewers’ priority assessments? Can an AI-assisted, human-verified brief reduce that sensitivity at a lower preparation cost than a simple template workflow?
Why it matters
Reviewers have limited attention, and communities differ in their ability to prepare documents. If wording changes judgments about the same evidence, a common review format may help. The study informs a practical choice: narrative descriptions, a fixed template, or AI-assisted preparation.
Data & feasibility
Public DREAM records supply descriptions, objectives, costs and document references. The stored 12 August pilot contains 100 project records, 98 with an exact community-code match and 85 with forecast costs. These are recently updated records, not a representative set of grant applications. A documented public API was checked successfully in September.
Fact checking, reviewer recruitment and experimental ratings remain to be done. The optional community digitalization index has an unresolved measurement period.
The experiment
- Select a defined civilian project category. Two people verify each source-linked dossier, including missing facts.
- Create plain and polished versions with the same claims, qualifications and comparable length. Randomly assign versions; each reviewer sees one version per project.
- Preregister a primary 0–100 priority-for-further-review score. Estimate within-project differences with uncertainty accounting for shared projects and reviewers.
- Independently prepare standard briefs from each version using a frozen AI workflow, then verify them. Compare presentation sensitivity, errors and total preparation time against a no-AI template workflow.
AI reviewers are a separate comparison. They cannot establish how human reviewers or actual funders behave.
Contribution & boundaries
Prior studies already examine AI-assisted proposals and AI reviewer sensitivity. The proposed contribution combines reconstruction records, expert human reviews, fact-level verification and a non-AI baseline. It does not claim to be the first presentation experiment.
No experiment has been run. A score difference would not prove an effect on funding or project quality. An identical fact card removes wording differences by construction; AI value must be shown through faithful preparation and its cost.
Request to the professor
Help choose a sector, agree a review rubric and secure access to qualified reviewers. Begin with roughly 30 manually checked projects as a feasibility pilot; use its variance, workload and recruitment results to design a powered, preregistered main study. Deliver an auditable dataset, a tested brief format and evidence on whether AI adds value. A precise null result or a better simple template is also useful.