How the panel is assembled

Our project follows the effect of the reorganisation of United States global health assistance on HIV services in LMICs. Measuring the effect of the reorganisation on HIV services requires data that represent two things, namely (1) what donors report spending, and (2) what the HIV services report doing. Funding data alone are not enough, and programme data alone are not enough, because funding data are timely but say nothing about delivery, while programme data describe delivery but are not available in real time and are unevenly distributed geographically.

Principles

We follow four rules that govern what enters the panel (/what data are collected).

• Structured sources only. Figures come from machine-readable feeds and from published tables that can be read deterministically. Prose is never parsed for numbers. Scanned documents are out of scope, because reading scanned documents reliably would require optical character recognition and manual checking that the project cannot guarantee at scale.
• No language models anywhere in the pipeline. Every figure is traceable to a field in a file or an entry in a published table. An honest record that a country does not publish in a usable form is more useful than a plausible number without a traceable origin. Language models are not used for data extraction to minimise data contamination and to maximise the validity and reliability of our project.
• Data origin on every row. Every observation carries the name of the source, the URL the observation was read from, the indicator code the source uses and the date of retrieval.
• Uncertainty is kept. Modelled estimates carry lower and upper bounds, and the panel stores and draws the bounds. The point estimates for the 95-95-95 targets, the UNAIDS targets for the shares of people living with HIV who are diagnosed, on treatment and virally suppressed, carry intervals wide enough that a single number misrepresents the estimate.

What is read, and from where

We designed the pipeline to build the panel with two layers. The aggregator layer holds the figures published by international and donor bodies that compile figures across many countries, covers almost every country, and provides the near-global base of the panel. Data in the aggregator layer are sourced primarily from UNAIDS AIDSinfo, the WHO Global Health Observatory, the World Bank, the OECD Creditor Reporting System, and United States self-reported disbursements from ForeignAssistance.gov and programme results from PEPFAR Panorama. The country layer is the national programme record, read directly where a country publishes the record in a usable form.

Going to source, country by country

A country is read at source when our project takes the figures for the country directly from a national publisher, for example the national HIV programme, the health ministry, the national statistics office or an official national HIV model, and not only from the international and donor bodies that compile figures across many countries. The question we encountered at the pilot stage was whether national programme data can realistically be read at source, country by country. The answer varies from country to country, and the variation between countries is a finding of our project. All nine Southeast Asian countries run a national HIV programme database. Data exist in every Southeast Asian country, but several countries do not allow foreign or non-institutional access, and access to the national data takes three forms, from fully open to closed.

• An open structured feed, as in Thailand, where the national programme's data mart answers queries directly.
• An open born-digital report whose tables can be read without interpretation, as in Cambodia and the Philippines.
• A closed database that releases only scanned documents or the national submission to the international aggregators. For a country with a closed database, the realistic route is an institutional data-sharing agreement, not public collection.

We decided whether a national report is worth reading at all by a single test. A national report adds information only if the report carries data the aggregators do not already receive, for example sub-national detail, more recent periods, figures for key populations (groups at higher risk of HIV, for example sex workers and people who inject drugs), or programme counts. The Philippine quarterly surveillance report, for example, passes the test. The Philippine report is regional, quarterly, and more detailed than the annual report the Philippines submits to UNAIDS. Malaysia's national report does not pass the test, because the Malaysian report is the annual submission of Malaysia to UNAIDS, so reading the Malaysian report would return figures already in the panel through UNAIDS.

Southeast Asia

In the pilot stage (i.e., exclusively extracting data from the Southeast Asia region), the national treatment cascades of Thailand, Cambodia and the Philippines, meaning the numbers of people diagnosed, on treatment and virally suppressed, are read at source. Thailand is read through a live programme interface, and Cambodia and the Philippines through published report tables. Indonesia is covered by a community-led monitoring interface, which reports what service users say about a small number of facilities.

Eastern and Southern Africa

The Eastern and Southern African countries were assessed on 3 August 2026 by an automated check of national data hosts. The automated check recorded whether every host answered, what every host returned and whether a DHIS2 data interface responded, DHIS2 being the health information platform used by many national health ministries. An automated check can show that a host blocks access, but a check of a guessed host name that does not exist shows nothing about whether national figures are available. South Africa's Thembisa model is the only national source in the region that is open, structured and gives figures that international reporting does not. Angola, Burundi, Comoros, Eritrea, Madagascar, Mauritius, Somalia and South Sudan were not checked and are not yet assessed.

Western and Central Africa

Western and Central Africa was assessed in four survey rounds, and the fourth round ended on 1 October 2026. The fourth round graded every candidate national source against the panel, so a national report whose years the panel already holds counts as no gain. Nigeria, Cameroon, Niger, Guinea and Chad are read at source, through six national sources. The spending assessments of DR Congo and Togo would add four country-years from 2019 onward and are under review.

South Asia

South Asia was assessed from 14 to 25 October 2026, by the method of the Western and Central African assessment. Every candidate national source was graded against the panel, and a national report whose years the panel already holds counts as no gain. Bangladesh, India, the Maldives, Nepal and Sri Lanka are read at source, through six national sources.

East Asia and the Pacific

East Asia and the Pacific was assessed from 14 to 27 October 2026, by the method of the Western and Central African assessment. China, Mongolia, North Korea, Papua New Guinea, the Solomon Islands and Vanuatu are read at source. The Fijian connector is built, and the Fijian figures enter the panel once the gov.fj government domain gives a network address again.

Eastern Europe and Central Asia

Eastern Europe and Central Asia was assessed from 20 to 22 October 2026, by the method of the Western and Central African assessment. Armenia, Belarus, Georgia, Kazakhstan, Kyrgyzstan, Montenegro, Tajikistan and Ukraine are read at source, through nine national sources.

Latin America and the Caribbean

Latin America and the Caribbean was assessed from 5 to 28 October 2026, by the method of the Western and Central African assessment. Argentina, Belize, Bolivia, Haiti, Nicaragua, Paraguay and Peru are read at source, through eight country sources.

Middle East and North Africa

The Middle East and North Africa was assessed on 28 October 2026, by the method of the Western and Central African assessment. Lebanon is read at source, through the statistical bulletins of the Ministry of Public Health.

Reporting periods

Funding and programme data are filed in periods, and a period that is still open for data collection is likely subject to a figure drop when the data collected are only provisional but not full. For example, United States disbursements for fiscal year 2027 are still being reported, and PEPFAR results for 2025 cover only one of the four quarters of 2025. Points from an incomplete period are drawn with an open red marker, are named in the note under every figure, and are never compared with complete years.

Comparing sources

Where two sources report the same country, indicator and year, the panel keeps the value from every source and compares the values. A comparison between two sources is classified before the comparison is read, because agreement between two sources that publish the same underlying estimate is arithmetic, not corroboration. The classification method, the thresholds, and the classification of every source pair with the basis for the classification are set out on the reconciliation page.

What the panel does not support

The figures in the panel describe reported funding and reported service activity. The figures alone do not establish that a change in funding caused a change in services. Reporting lags differ by source and by country, so a decline that appears in a first series and not in a second series may reflect a difference in filing and not a difference in delivery. Programme results are not national totals. PEPFAR reports what PEPFAR-supported sites did, which is part of the national HIV response of a country and not the whole national response. Community-led monitoring covers a small number of selected facilities and is not a representative sample.

Reproducing the panel

We design the pipeline as a Python package with a separate connector, a short program that reads a single source, for every source, and a single assembly step. The pipeline writes the panel to CSV, Parquet and SQLite, runs the reconciliation, and then we build the website from the CSV, Parquet and SQLite files alone. The panel on the website was last assembled on 29 October 2026.