USG MoU dashboard

Live app: https://usg-mou.streamlit.app/ · Standalone explorer: https://tafmeijer.github.io/USG_MOU/

Interactive dashboard tracking the America First Global Health Strategy bilateral health MoUs (34 countries, five-year cooperation plans 2026–2030): who funds what, how U.S. funding tapers while government co-financing ramps up, and the programmatic targets each agreement commits to.

All information here is publicly available — amounts from the KFF tracker, detailed tables transcribed from the 16 published full MoU texts hosted by Public Citizen, with Health Policy Watch and Think Global Health as secondary mirrors.

How each text became public

Public Citizen's table marks every agreement with the route by which its text reached the public, and the distinction is part of the story:

Marker	Meaning	Countries
*	The U.S. government made the text public itself, via its Case Act reporting page	Uganda only
*^	Published on the Case Act page and released under FOIA	Kenya, Mozambique, Nigeria, Ethiopia, Malawi
^	Public only because of Public Citizen's Freedom of Information Act requests	Rwanda, Liberia, Lesotho, Eswatini, Cameroon, Sierra Leone, Botswana, Madagascar, Côte d'Ivoire, Burundi

Ten of the sixteen public texts — and six of the seven countries added in September 2026 — exist in the public record only because Public Citizen sued the State Department for them. The August 2026 production (FL-2026-00021) is what that litigation yielded. Both the marker and a plain-language description are carried per country in data/countries.csv (Disclosure marker, How the text became public), and 18 of the 34 signed agreements still have no public text at all.

Pages

Page	What it shows
Home (Overview)	All 16 countries side by side, per investment area — yearly trajectories as a grey field with a clickable country list (ordered by 2026 size) that draws the chosen country in colour, plus full-term totals and a US$ ↔ % of combined toggle
Country investment view	One country's small multiples across every investment area, with a USG / Govt (existing + new) / Both toggle and the same $/% toggle; the country name links to the source PDF
Investment area view	The transpose: pick one investment area and see the global USG/government split for it, every country ranked by commitment, and one small multiple per country ordered largest to smallest
Country programmatic view (v0)	One country's indicator baselines & 2026–2030 targets (outcome / process / 7-1-7), each indicator as its own small chart
Full data tables	The complete budget and programmatic datasets across all countries — filterable, sortable, searchable, downloadable
Sources & methodology	Every link (trackers, mirrors, all 34 agreements), extraction method, and caveats

explorer.html is a standalone, dependency-free version of the trajectory explorer — open it directly in a browser, no Python needed. docs/index.html is the copy GitHub Pages serves; both are regenerated from data/budget_series.csv by analysis/rebuild_explorer_data.py.

Theming

The app ships light and dark themes (.streamlit/config.toml); Streamlit's default setting ("Use system setting") follows the browser/OS colour-scheme preference, and users can still override it in the app menu. Explicit chart colours track the active theme via mou_lib.palette() (st.context.theme), so donut strokes, centre labels, chip legends and the handful of dark-adjusted series colours switch with it.

Run locally

pip install -r requirements.txt
streamlit run Home.py

Publish

1. Push this folder to a public GitHub repository:

   git init
   git add .
   git commit -m "USG MoU dashboard"
   git branch -M main
   git remote add origin https://github.com/<you>/mou-dashboard.git
   git push -u origin main

2. Deploy free on Streamlit Community Cloud: New app → pick the repo → main file Home.py. Every push to main redeploys automatically.

Data

File	Contents
data/budget_series.csv	Aggregated, safely summable series (Country × Investment area × Year × Funder) — feeds all budget charts
data/budget_tidy.csv	Full line-item detail as printed in each MoU, with Row type flags (subtotals, appendix breakdowns, existing funding), source notes, and the MoUs' own printed footnotes transcribed verbatim (MoU footnote (verbatim) / MoU footnote location)
data/programmatic_tidy.csv	Every Section-1 indicator: baseline + yearly targets, units, value types, source notes, and the MoUs' own printed footnotes (verbatim + location)
data/countries.csv	All 34 agreements: announced amounts (KFF tracker), dates, program areas, source-PDF links, plus MoU USG (USD) / MoU co-financing (USD) / MoU basis note — what each of the 16 public texts prints in its own co-funding summary, which differs from the headline for 11 of them
data/sources.csv	Reference trackers and mirrors
data/strategic_areas.csv	Strategic-investment domains/areas named in every MoU's §2.6, with printed prices and page references
analysis/	Derived analyses: $/FTE imputation scripts and outputs, cadre-table harvests (Mozambique, Uganda), baseline valuations, analysis notes

Summing rules (see Sources & methodology page in the app): only Row type = "Line item" rows plus existing-government rows are summed; the MoUs' own subtotals, nested appendix breakdowns, FTE headcounts and domestic-expenditure pledges are excluded to avoid double counting. "Government" = new co-financing plus existing funding where tabulated (Nigeria, Ethiopia, Rwanda publish no existing split — their shares are understated).

Provenance & caveats

Read this before using the data. All figures were transcribed from the sixteen signed MoU texts in the public domain — released by the U.S. government via its Case Act reporting page, or under Public Citizen's FOIA requests, and mirrored by Public Citizen, Health Policy Watch and Think Global Health — and from the KFF tracker. The MoUs are government-to-government cooperation documents that were not originally drafted for publication; no official consolidated dataset of them exists, and this repository is an independent research reconstruction. In particular:

1. Amounts are plans, not obligations. By the MoUs' own terms they are not international agreements and all activities are subject to the availability of funds. Nothing here represents appropriations, disbursements, or current implementation status, and the published scans may have been amended or superseded since.
2. Figures are transcribed as printed, including source-document errors (misprinted totals, conflicting appendices, internally inconsistent tables). These are preserved uncorrected and flagged in the Source note column.
3. Derived values are clearly separated from printed ones. Rows flagged Imputed (derived - not printed in MoU) / Imputed baseline (pre-MoU - derived) and series with Basis other than "Printed in MoU" are estimates produced by this project's documented methodology (FTE commitments priced at unit rates derived from the same documents) — they are never printed values, they carry stated confidence ranges, and every chart lets you toggle them off. See the app's Sources & methodology page and analysis/ for full derivations.
4. Tables were machine-transcribed from the source PDFs and independently verified against them. Compiled August 2026. Not an official product of any government or organisation, and no affiliation with any party to the MoUs is implied. Content is provided as-is for research and transparency purposes; corrections are welcome via issues or pull requests.

License

Code and derived data are released under the MIT License. The underlying MoU texts remain the work of their authors; transcription here is for research and transparency purposes.

Changelog

September 2026 — the FOIA release

Public Citizen published the State Department's FOIA production (FL-2026-00021) on 30 August 2026. Public full texts went from 9 to 16 of 34.

Added: Lesotho, Eswatini, Sierra Leone, Botswana, Madagascar, Malawi, Burundi — full budget appendices, Section-1 indicator tables and §2.6 strategic areas.

Re-sourced: Rwanda, Liberia, Cameroon and Côte d'Ivoire now point at the official release rather than a third-party mirror. Rwanda, Liberia and Cameroon are numerically identical to the earlier copies (Rwanda was diffed word by word — no change). Côte d'Ivoire gained three Appendix 1 rows that were illegible in the earlier scan: Frontline Lab Workers $ ($2.1M), Frontline Healthcare Workers $ ($31.7M) and Management & Operations $ ($29.2M), lifting its itemised U.S. total from $423.6M to $486.7M against a printed $487.2M.

South Sudan is still excluded — the only available text is a pre-signature April 2026 draft; the agreement was signed on 25 June 2026 and its final text is not public.

Pattern breaks worth knowing about:

• Botswana runs 2026–2028, not 2026–2030. Its headline is not comparable to the others without adjusting for term length.
• Botswana's U.S. line items do not reconcile to its own printed totals in any year (−$3.6M over the term) — the only published MoU where the U.S. side fails to add up.
• Burundi and Côte d'Ivoire print the 6% management-and-operations carve-out as its own Appendix 1 line; elsewhere it is a silent gap (Lesotho, Madagascar and Botswana all sit exactly 6% below their headline).
• The Appendix 1 government tables are mislabelled. They are headed "total new planned financial support" but carry the §2.x.3 Total Government Funding column — new plus existing. True of the original nine as well.
• Three new texts disagree with the KFF tracker: Malawi ($744.8M/$55.0M vs $792M/$143.8M), Eswatini ($192.7M vs $205M) and Botswana ($99.6M vs $106M, exactly 6%).

Build scripts. data/budget_tidy.csv and data/programmatic_tidy.csv are the hand-transcribed source of truth; everything else is derived. After any change under data/, regenerate and check in the same commit:

python3 analysis/fte_rate_imputation_all.py && python3 analysis/apply_imputation_to_dashboard.py && python3 analysis/rebuild_budget_series.py && python3 analysis/rebuild_explorer_data.py && python3 analysis/validate_data.py

The first two recompute the imputed and pre-MoU-baseline government rows; the next two regenerate data/budget_series.csv and the JSON snapshot embedded in explorer.html / docs/index.html. analysis/validate_data.py then re-checks the invariants the dataset rests on — aggregates against their parts, series against tidy, countries.csv arithmetic, the explorer snapshot, the totals quoted on the methodology page, and the programmatic Direction / Qualifier columns. Skip the imputation steps if only printed figures changed. analysis/extract_aug2026_release.py covers the indicator rows and provenance refresh for the August 2026 release.