A dashboard for asking a question the official numbers don't really answer: for every euro France collects in tax, how much actually does anything useful?
Government communications tend to report spending the way a household reports its grocery bill ("we spent €X on education"). That tells you nothing about what the money produced, what fraction was eaten by the apparatus moving it around, or whether other countries get more out of the same euro. Fiscoscope tries to compute that.
The dashboard is built for people who want to look at public finance numbers without someone explaining first why they're good or bad. Mostly: journalists, researchers, and anyone who's tired of budget debates that never cite a denominator.
It is available here
It reports ratios, not totals. Total spending tells you almost nothing on its own; ratios at least let you compare across years and across countries.
Wherever the data goes back that far, series start in 1995, not just the last election cycle. A lot of trends only show up over decades.
Peers are Germany, the UK, Italy, Spain, the Netherlands, Sweden, and the OECD average. France-only numbers are usually misleading.
The whole pipeline runs on a cron schedule. Nobody is editing spreadsheets by hand, which means the numbers can't drift from whatever INSEE and the OECD publish.
No institutional backing. The code and the methodology are in this repo.
| KPI | What it measures |
|---|---|
| Administrative Overhead Rate | Public sector wage bill as % of total public expenditure |
| Friction Ratio | Share of revenue absorbed by the State's core support and sovereign functions (general services, defence, public order) — the running cost of the apparatus, not a measure of waste (France-only) |
| Productive Spend Ratio | Share of expenditure flowing to productive COFOG functions (infrastructure, R&D, environment) |
| Social Protection / Investment Ratio | Total social protection spending (COFOG GF10 — pensions, unemployment, family, etc.) vs. gross fixed capital formation, as a multiple |
| Monthly Budget Execution | Live cumulative revenues, spending, and balance for the current year |
| Fiscal Deficit Trend | Balance and debt as % of GDP, France vs. peers |
| Tax Expenditure Cost | Cost of niches fiscales (GTED estimate; lower than the official PLF figure) as % of revenue |
| Spend vs. Outcome (health) | Public health spend (% of GDP) and life expectancy at birth, shown as two parallel series (education/PISA not yet available) |
Full KPI specifications and methodology are in docs/PRD.md §5.
┌────────────────────────────────────────────────────┐
│ VPS (Hetzner) │
│ ┌──────────────────────────────────────────────┐ │
│ │ Python pipeline (systemd timer) │ │
│ │ fetch → cache → compute KPIs → write JSON │ │
│ └──────────────────────────────────────────────┘ │
│ ┌──────────────────────────────────────────────┐ │
│ │ FastAPI app (uvicorn, always-on) │ │
│ │ read-only, CORS-restricted, rate-limited │ │
│ └──────────────────────────────────────────────┘ │
└────────────────────────────────────────────────────┘
│ HTTPS (Let's Encrypt via <ip>.sslip.io)
▼
┌────────────────────────────────────────────────────┐
│ Cloudflare Pages — Vite + React + Recharts │
└────────────────────────────────────────────────────┘
Two systemd timers drive the pipeline: a monthly one for budget execution, Urssaf, and France Travail, and an annual one for INSEE COFOG, OECD, and the GTED tax-expenditure data.
The FastAPI app serves data/output/*.json over read-only HTTP. No auth, because the data is public anyway. Abuse is bounded by slowapi rate limiting and a CORS allowlist, which is enough for a small read-only API.
The frontend is a static site on Cloudflare Pages that fetches from the VPS API at page load.
All sources are public. None require an API key.
- INSEE BDM for national accounts time series (COFOG, wage bill, GDP, CPI), via the SDMX API. idBanks are resolved at runtime from INSEE's official mapping file.
- data.economie.gouv.fr for monthly budget execution (
situations mensuelles budgétaires) and annual final execution (PLRG). - OECD Data Explorer for COFOG and fiscal indicators on peer countries.
- open.urssaf.fr for the private sector wage bill, used in the public/private ratio.
- data.gouv.fr / France Travail for monthly unemployment-insurance allocataires.
- GTED (Global Tax Expenditures Database) for the annual cost of tax expenditures — a CC-BY academic compilation of France's PLF Voies et Moyens tome II (not the official annex directly, and narrower in scope than it).
git clone https://github.com/Arnobgr/french-efficiency-dashboard.git
cd french-efficiency-dashboard/backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
# Validate INSEE idBank resolution first:
python -m fetchers.insee_idbank_resolver
# Full pipeline run (writes data/output/*.json):
python run_pipeline.py --mode full
# Serve the output over HTTP:
ALLOWED_ORIGINS="http://localhost:5173" \
uvicorn api:app --host 127.0.0.1 --port 8000 --proxy-headersRequirements: Python 3.11+. No secrets needed because all data sources are public.
| Mode | What it runs | Cadence |
|---|---|---|
--mode monthly |
Budget execution, Urssaf, France Travail | 1st of each month |
--mode annual |
INSEE COFOG, OECD, PLRG, structural KPIs | 1st February (and June, after PLF) |
--mode full |
Both | Initial setup or manual refresh |
GET /healthz— liveness checkGET /api/meta— pipeline run metadataGET /api/kpis— list of available KPI namesGET /api/kpi/{name}— one KPI's JSON
french-efficiency-dashboard/
├── backend/
│ ├── fetchers/ # One module per data source
│ ├── processors/ # One module per KPI
│ ├── data/
│ │ ├── raw/ # Cached API responses (not committed)
│ │ └── output/ # Final JSON served by api.py
│ ├── api.py # FastAPI read-only server
│ ├── config.py # Constants and env-var loading
│ └── run_pipeline.py # Entry point
├── frontend/ # Phase 2 — Vite + React + Recharts
├── docs/
│ ├── PRD.md # Full specification
│ ├── runtime-discoveries.md # Methodology decisions per KPI
│ └── known-gaps.md # Documented limitations
├── CLAUDE.md
└── README.md
The one place where this project actually editorializes is COFOG bucketing: sorting the 10 COFOG functions into productive, redistributive, and administrative. No official source does this for you, and the choice is debatable. The bucketing lives in backend/processors/cofog.py and is repeated in each KPI's output methodology field so anyone reading the JSON sees what was assumed.
Education (GF09) and Health (GF07) both mix investment and transfers, so any bucketing of them is approximate. The output flags it. INSEE rebased the national accounts in 2020, which can create a structural break in long series; the processors detect and flag that too.
For per-KPI methodology decisions and the reasoning behind them, see docs/runtime-discoveries.md. For things we tried and couldn't make work, see docs/known-gaps.md.
PRs and issues welcome. A few things that make review easier:
- Target the
devbranch. - Keep changes scoped. One KPI or one fetcher per PR is ideal.
- Processors must be deterministic given the same raw input.
- Update
docs/runtime-discoveries.mdwhen a methodology choice is made or revised. - For changes that affect KPI computation, include a before/after of the relevant JSON output.
MIT.
Built on top of the open data published by INSEE, the Direction du Budget, the OECD, the Urssaf, France Travail, and data.gouv.fr.