Transparent Humanitarian Funding UI: Data-Grounded LLM-Compatible Explanations of OCHA FTS Funding Flows

Abstract

Humanitarian funding dashboards expose complex relationships among donors, recipients, sectors, reporting statuses, and response-plan gaps. This paper evaluates a transparent funding-interface design in which data-grounded, LLM-compatible explanations are generated for OCHA Financial Tracking Service (FTS) flow records and paired with accessible legends for funding gaps and reporting certainty. The study conducts a complete reproducible experiment on downloaded FTS/HDX CSV data for the Occupied Palestinian Territory. The primary incoming-flow file contains 394 records and 37 fields, with $794.41M in reported incoming flows, 57 source organizations, 73 destination organizations, and 20 normalized cluster labels. The linked requirements file identifies the 2026 OPT Flash Appeal (FPSE26) with $4.06B in requirements, $459.73M in reported funding, and an official 11.0% funded value, leaving a measured $3.60B gap. Four generation conditions were evaluated on every flow: a minimal label, a generic narrative baseline, a gap-aware explanation, and a constrained audit explanation. Metrics included entity coverage, numeric grounding, unsupported-claim rate, traceability, legend completeness, actionability, reading ease, length, and an aggregate UI-fit score. The proposed constrained explanation achieved a mean UI-fit score of 98.3, compared with 28.2 for minimal labels, 13.7 for generic narratives, and 52.8 for gap-aware explanations. Results show that transparent funding explanations improve semantic completeness when they explicitly bind donor, recipient, sector, amount, status, boundary scope, and plan-level gap context. The included code, processed data, tables, and figures reproduce all reported findings.