# Register of French public spending and its evaluations: sources

Retrieved 2026-10-03 with WebFetch. Every downloaded file was kept under
`raw/` (and `raw/rdd/` for the spending reviews) next to `register/`; that
directory is not in the repository. "pdf p." is the page index in the PDF
file; "p." alone is the printed page number.

Every table is rebuilt by an OCaml executable in `build/`, run from the
repository root. `REGISTER_SRC` (default `data`) names the directory that
holds `register/` and `raw/`; `parse_plfss` and `write_part_b` take it as
their argument instead.

- `dune exec data/register/build/parse_plfss.exe` writes
  `social_exemptions.csv` from `raw/plfss2026_annexe4.txt`.
- `dune exec data/register/build/build_guillaume.exe` writes
  `guillaume2011.csv` from `raw/ifrap_evaluation_croisee.pdf`.
- `dune exec data/register/build/build_part_a.exe` writes
  `evaluations_part_a.csv`, checking each quote against `raw/*.pdf`.
- `dune exec data/register/build/write_part_b.exe` writes
  `evaluations_part_b.csv`.
- `dune exec data/register/build/merge_and_cover.exe` merges both parts into
  `evaluations.csv` and prints the coverage arithmetic.
- `dune exec data/register/build/register.exe` writes `register.csv` from
  `cofog2.csv`, `social_exemptions.csv`, `../tax-expenditures.csv` and
  `../literature/studies.csv`.

The PDF tools need poppler's `pdftotext` and `pdfinfo`.

## cofog2.csv: general-government spending by function

- Eurostat `gov_10a_exp`, France, S13 (general government), TE (total
  expenditure), MIO_EUR.
  - Values for 2014-2024:
    `https://ec.europa.eu/eurostat/api/dissemination/sdmx/2.1/data/gov_10a_exp/A.MIO_EUR.S13..TE.FR?format=SDMX-CSV&startPeriod=2014&endPeriod=2024`,
    saved as `raw/eurostat_gov_10a_exp_FR_S13_TE.csv`. Dataset last update
    16/09/26.
  - Labels:
    `https://ec.europa.eu/eurostat/api/dissemination/sdmx/3.0/data/dataflow/ESTAT/gov_10a_exp/1.0/A.MIO_EUR.S13.*.TE.FR?c[TIME_PERIOD]=2024&format=csvdata&formatVersion=2.0&labels=both`,
    saved as `raw/eurostat_gov_10a_exp_FR_2024_labels.csv`. Its 2024 values
    match the first file cell for cell.
- The file has 69 level-2 codes, the 10 level-1 divisions and TOTAL, marked
  in the `level` column. All 2023-2024 values carry flag `p` (provisional).
- Checks:
  - The level-2 codes sum to 1,671,794.1 M€ against a TOTAL of 1,671,793.8.
  - Every division equals the sum of its level-2 codes to within 0.1.

## social_exemptions.csv: social-contribution exemptions, 2024

- Source: Direction de la Sécurité sociale, *PLFSS 2026, Annexe 4*,
  14 October 2025,
  https://www.securite-sociale.fr/files/live/sites/SSFR/files/medias/PLFSS/2026/PLFSS2026-Annexe4-20251014-183718-59-11_avec%20couverture.pdf
  (`raw/plfss2026_annexe4.pdf`, 59 pages).
- Tables parsed:
  - Tableau 13: allègements généraux and compensated targeted exemptions,
    ROBSS. Printed p. 44.
  - Tableau 14: non-compensated exemptions, ROBSS. Printed p. 45.
  - Tableau 21: schemes outside ROBSS, such as complementary pensions and
    unemployment insurance. Printed p. 50.
  - Tableau 23: Covid exemptions. Printed p. 52.
  - Tableau 9: exemptions d'assiette. Printed p. 30. The value used is the
    net equivalent contribution in Md€ ×1000, and 2024 is provisional.
- Columns:
  - `level` separates group subtotals from items and "dont" sub-items.
  - Sum only the `item` rows within a table to avoid double counting.
- Checks:
  - Tableau 13: items sum to 71,722 = 64,928 + 6,794.
  - Tableau 21: items sum to 14,169 = 12,671 + 1,498.
  - Tableau 9: items sum to 15.0 Md€ against a printed total of 14.9. The
    difference is rounding.
- Totals for 2024:
  - ROBSS: 74,319 M€ (Tableau 13). Tableau 6 gives 74.7 Md€ and Tableau 15
    gives 74,271.
  - Outside ROBSS: 14,169 M€.
  - Exemptions d'assiette: 14,900 M€ net.
- Limit: the annex lists every exoneration but only the main exemptions
  d'assiette. Other social niches, such as reduced CSG rates, are in the
  LACSS annex, which was not downloaded.

## guillaume2011.csv: the 2011 Guillaume committee scores (PARTIAL)

- **Gap: the measure-by-measure annex was not obtained.** The report places
  the individual scores in Annexe IV. It is at
  https://www.economie.gouv.fr/files/rapport-comite-evaluation-depenses-fiscales-et-niches-sociales.pdf
  (356 pages). economie.gouv.fr returns HTTP 403 to the fetcher for every
  URL, the homepage included. The leparticulier.lefigaro.fr copy is blocked,
  the budget.gouv.fr copy cited by Lexbase returns 404, the vie-publique
  pages return empty, the IGF archive starts after 2011, and SlideShare
  shows only 20 of the 356 pages.
- What the file holds instead:
  1. **Senate summary tables** for all measures:
     - Tax expenditures: 470 measures, of which 339 scored; scores 0/1/2/3 =
       125/99/46/69, for 11,764/28,236/11,393/9,305 M€.
     - Social niches: 68 measures, of which 46 scored.
     - Source: Sénat, rapport d'information n° 64 (2011-2012),
       https://www.senat.fr/rap/r11-064/r11-06422.html. This is a web page
       read through WebFetch. Each table's rows were checked to add up to
       its totals.
     - The 46 social niches scored differ from the "52" in the brief.
  2. **IFRAP's tabulation by budget mission** for 13 missions: counts and
     amounts by score. Source:
     https://www.ifrap.org/sites/default/files/documents/import/evaluation-croisee-niches-fiscales.pdf
     (`raw/ifrap_evaluation_croisee.pdf`), parsed with pdftotext. Missions
     that appear in several of its tables were checked to be identical.
- To finish: download the economie.gouv.fr PDF in a browser and put it in
  `raw/`. Annexe IV can then be parsed.

## evaluations.csv: evaluations by line or measure

- **Part A** (32 rows, built by `build_part_a.exe`):
  - Each quote taken from a PDF is checked against the PDF text by
    `Evlib.locate`, which also finds its page.
  - Some quotes come from web pages, through WebFetch (marked "not
    byte-verified" in `location`). These are the Cour des comptes platform
    summaries, the Insee web summary and the RePEc abstract.
  - Tax-expenditure amounts come from the *Évaluation des voies et moyens,
    tome II, PLF 2026* (https://www.budget.gouv.fr/documentation/file-download/30586),
    2024 actuals, through the repo's parsed `data/tax-expenditures.csv`. The
    builder asserts that each amount matches. The 2024 total of 89,406 M€
    matches the Cour des comptes' "89,40 Md€" (NEB 2026, synthèse, footnote
    1).
  - Social amounts come from PLFSS 2026 Annexe 4. Spending amounts come from
    cofog2.csv.
  - When a whole COFOG line is used as the amount but only part of it was
    evaluated, the `cofog_or_measure_code` column says "(component)".
- **Part B** (22 rows, `evaluations_part_b.csv`, notes in
  `part_b_notes.txt`):
  - It covers the IGF/IGAS revues de dépenses for 2023-2026 (18 reports, all
    downloaded except social housing) and the HCSP July 2026 report.
  - The HCSP report has no per-scheme evaluation annex. Its annexes are
    descriptive, and Part 3 proposes a framework only.
- **`causal_yes_no`**: "yes" means the evaluation (or the study it relies
  on) uses a counterfactual design: DiD, RDD, experiment, or matching with a
  comparison group.
- Main documents:
  - Cour des comptes, NEB *Dépenses fiscales* 2023 (pub. April 2024) and
    2025 (pub. April 2026).
  - Cour des comptes, *Le pacte Dutreil*, Nov. 2025.
  - France Stratégie:
    - CICE synthesis, 2020.
    - CNEPI CIR, 2021.
    - Bozio-Wasmer mission, 2024.
    - Capital-tax reform committee, final report, Oct. 2023.
    - France Relance committee, synthesis, Jan./Mar. 2024.
  - IPP rapport n°38 (CICE switch to contribution cuts), 2022.
  - Insee Analyses n°1 (2011) and n°19 (2014).
  - Drees ER 922 (2015).
  - NBER WP 23848 (Benzarti & Carloni, 2017).
  - CEPREMAP Docweb 1316 (ZFU, 2013).
  - CAE Focus 113 (2025).
- Not done:
  - The CPO reports.
  - The "évaluations" sections of the Projets annuels de performance.
  - Causal studies of the overtime exemptions (codes 120146 and the PLFSS
    lines). No downloadable copy was found, so none is claimed.

## ccomptes_platform_causal*.csv: the Cour des comptes evaluation platform

- Source:
  https://www.ccomptes.fr/fr/plateformes-citoyennes/plateforme-evaluations-politique-publique/explorer-evaluations
- On 2026-10-03 the platform listed 1,114 evaluations. Of these, 422 are
  tagged "Approche descriptive et causale". Theme counts for all evaluations
  and for the causal ones are in `ccomptes_platform_causal_themes.csv`.
  Themes overlap.
- The item list (`ccomptes_platform_causal.csv`) was harvested from the
  listing pages through WebFetch. **It is incomplete: 20 of the 422 rows**
  (pages 0-1). ccomptes.fr then refused connections and the harvesting agent
  hit a usage limit. Titles come through WebFetch's summariser and may not be
  character-exact.
- Keyword searches on the causal filter were also used to find studies:
  "heures supplémentaires" (15 hits), "TVA" (3), "domicile" (16), "RSA" (23),
  "hôpital" (7), and the "Logement et urbanisme" theme (14).

## Verification of quotes

- Part A: every quote marked with a pdf page was matched by `Evlib.locate`.
- Part B: every quote was re-matched against the PDF text, fragment by
  fragment around "[...]". All matched except the social-housing row, whose
  source is the IGF web page because the PDF was not downloaded.

## Coverage estimate

`merge_and_cover.exe` prints the arithmetic. See the final report.