# Commune-level data for place-based programmes and local grants: sources Assembled 2026-10-03/04. Every value comes from a file downloaded from the URL given below. Some values are derived from such files, and each derivation is described. The raw downloads were kept under `raw/`, with one subfolder per item. That folder is not committed. The xlsx, dbf and shp files were read directly, without conversion by hand. | File | Rows | Content | Years | Geography | |---|---|---|---|---| | `acv.csv` | 244 | Action cœur de ville communes (233 in the 2018 list, 11 later additions) | 2018 (+2023-2025) | COG 2025 | | `pvd.csv` | 1,642 | Petites villes de demain, with convention signature date | 2020-2024 | COG 2025 | | `ti.csv` | 34,790 | Territoires d'industrie: 2018-2023 generation (16,150, derived spatially) and 2023-2027 (18,640) | 2018, 2023 | COG 2025 | | `eligible.csv` | 34,746 | every metropolitan commune: legal population, Insee zonings, EPCI industry share, and whether it met each programme's published criteria | 2015-2016, zonings 2020-2022 | COG 2025 | | `did.csv` | 4,620 | event studies and synthetic DiD of the three programmes (`analysis/places_did.ml`) | 2006-2025 | COG 2025 | | `zfu.csv` | 152 | communes hosting a ZFU, by generation | 1997, 2004, 2006 | COG 2025 | | `ber.csv` | 402 | Bassins d'emploi à redynamiser (Vallée de la Meuse, Lavelanet) | 2007 | COG 2025 | | `frr.csv` | 24,390 | FRR socle/FRR+, FRR+, FRR bénéficiaire (from `zrr/zrr_status.csv`) | 2024, 2025 | COG 2025 | | `qpv.csv` | 1,632 | communes hosting a QPV, 2015 and 2024 lists | 2015, 2024 | COG 2025 (+COM) | | `pinel.csv` | 453,375 | ABC zone on 1 Jan and 31 Dec, Pinel eligibility, commune × year | 2014-2026 | COG 2025 | | `sitadel.csv` | 462,989 | housing units and floor area authorised and started, by type | 2013-2025 | year's COG + COG 2025 column | | `dvf.csv` | 308,997 | house and apartment sales and median price (€, €/m²) | 2014-2023 | published + COG 2025 column | | `sirene_creations.csv` | 521,136 | Sirene establishment creations | 2010-2025 | COG 2025 | | `dgf.csv` | 315,619 | DGF, forfaitaire, DSU, DSR (+fractions), DNP, DGF and Insee population, DSU index | 2018-2026 | year's COG + COG 2025 column | | `accounts.csv` | 569,817 | commune accounts: DGFiP 2011-2018 and OFGL 2018-2025, one row per source | 2011-2025 | see below | | `thresholds.md` | - | legal population thresholds with CGCT and other articles | - | - | The `year` column is the start year for programme lists, and the data year elsewhere. **Not obtained:** - Sitadel by commune for 2010-2012: the commune file starts in 2013. - DVF for 2024-2025. - DGF components by commune for 2011-2017: only the DGF total, as booked in the DGFiP accounts. - Commune-level Pinel B2 agréments. - The pre-2014 zoning. ## Programme lists (item 1) All downloaded 2026-10-03. Raw copies were kept in `raw/programmes/`. The ABC zoning chain and the parsing of the JO arrêté tables are described below. **Geography.** Every programme file is in **COG 2025**, the vintage already used for `zrr/zrr_status.csv` and the Urssaf file. Source codes are projected with the INSEE passage table `zrr/raw/passage_2003_2026.csv`, which was not downloaded again. `source_codes` keeps the codes as published and `source_cog` gives their vintage. `share_of_components` is the share of a COG 2025 commune's component communes (in the source vintage) that are in the list, counted per commune, not weighted by population. It is 1 everywhere except where the column says otherwise. Two exceptions keep their source codes because they are outside the INSEE COG: Saint-Martin (97801) and the Polynesian communes (987xx) in `qpv.csv`. The files list treated communes only. A commune that is not in a file was not in that programme, according to that source. ### acv.csv: Action cœur de ville (244 communes, 230 ACV ids) - Source: Caisse des Dépôts / ANCT, "Listes des villes Action coeur de ville", data.gouv dataset 5fa9901c66cff095591963e8, ODS dataset modified 2026-09-20. The file is https://opendata.caissedesdepots.fr/api/explore/v2.1/catalog/datasets/villes_action_coeurdeville/exports/xlsx (`acv_cdc.xlsx`). Codes are valid COG 2025 codes. - Ids ACV001 to ACV222 are numbered alphabetically by town, with one inversion (Romans filed under R), and 219 of them are present. **Inference:** these are the March 2018 list of 222 towns ("Lancée en mars 2018", dataset description). In that range, ids 163, 200 and 209 are missing from the current file. Ids ACV223 to ACV233 are later additions, all with 2023-2025 signature dates, and are tagged `cohort = later addition`. For these, `year` is the year of `date_signature_cdc`, and it is blank when that date is blank. - `date_signature_cdc` is the ODS field `date_signature`: "the date when the city charter or agreement was formally signed". It mixes 2018-2019 dates (first conventions) with 2023-2025 dates (most likely the ACV2 amendments) and has blanks, so it is **not** a clean 2018 start date. Use `year` for the cohort. - Binômes (two communes under one id) give 244 communes for 230 ids. - The separate ANCT "Entrées de ville" list (ACV2 extension, 47 communes, COG 2025) is a text CSV that WebFetch does not save, so it was not used: https://static.data.gouv.fr/resources/action-coeur-de-ville-les-entrees-de-ville/20250929-142434/liste-acv2-com2025-20250929.csv ### pvd.csv: Petites villes de demain (1,642 communes) - Source: CDC / ANCT, "Programme petites villes de demain - liste des villes bénéficiaires", data.gouv 601c8b8c3ebd3f3468c2435a, ODS modified 2026-10-02: https://opendata.caissedesdepots.fr/api/explore/v2.1/catalog/datasets/programme-petites-villes-de-demain-liste-des-villes-beneficiaires/exports/xlsx (`pvd_cdc.xlsx`, 7.9 MB, includes the polygons). It has 1,642 communes and 1,642 ids, all valid COG 2025 codes. - `date_signature` is the signature date of the commune's adhesion convention ("Signature date", ODS field description), and `year` is its year. The years are: 2020: 5, 2021: 1,559, 2022: 62, 2023: 7, 2024: 4, and 5 blanks. The selection announcements (late 2020 to 2021) are not in the file. - ANCT also publishes the list in COG 2026 as a text CSV, which was not used: https://static.data.gouv.fr/resources/programme-petites-villes-de-demain/20260819-140115/liste-pvd-com2026-20260819.csv ### ti.csv: Territoires d'industrie, two generations - **2023-2027 (year 2023, 18,640 COG 2025 communes, 183 TI).** Source: CDC "Liste des Territoires d'industrie et des communes concernées", data.gouv 5fa9901d5a498251f21963ea, ODS modified 2025-09-30: https://opendata.caissedesdepots.fr/api/explore/v2.1/catalog/datasets/liste-des-territoires-dindustrie-et-des-communes-concernees/exports/xlsx?select=id_ti,lib_ti,insee_com,lib_com,epci_code,epci_name,dep_code (`ti_cdc_2018.xlsx`; the file name is historical and the content is the 2023 generation). It has 18,664 communes and 183 ids, with codes in COG 2023. The ODS description still says "149 territoires, 16,186 communes", but the data are the 183 TI of 2023-2027 (ids `ti-xxxx`). - **2018-2023 (year 2018, 16,150 COG 2025 communes, 149 TI). Derived, not published as a commune list.** WebFetch only offers the 2018-2023 list as a territory-level text CSV with no communes. The ANCT/MTE perimeter layer "TI - Territoires d'industrie périmètre au 2022-11-28" (data.gouv 660c007b6f059c84c900c513) was downloaded through its Atom link: https://atom.geo-ide.developpement-durable.gouv.fr/atomArchive/GetResource?id=e5bfd4e7-daa3-4989-a745-ac5468c8ee55&dataType=dataset (`ti2018_perimetre_com2022_atom.zip`, layer `liste_ti_perimetre-com2022-20221128`, 149 polygons, WGS84). Each COG 2025 commune centroid comes from https://public.opendatasoft.com/api/explore/v2.1/catalog/datasets/georef-france-commune/exports/xlsx?select=com_code,com_name,year,epci_code,geo_point_2d (`georef_commune_points.xlsx`, OpenDataSoft "georef-france-commune", year 2025, modified 2025-04-28). A commune is assigned to the TI polygon that contains its centroid (`method = centroid_in_perimeter_cog2022`). Every commune fell in at most one TI. - Check: TI were built from whole EPCI. `epci_share_inside` is the share of the commune's 2025 EPCI that is inside a TI. Of 558 EPCI touched, 18 are partly covered. Seven have exactly one commune inside: 240800920, 200030195, 200039865, 200067882, 200034379, 241200658 and 243100781. These are likely border false positives and are worth dropping or checking. Métropole du Grand Paris (40/130) and Aix-Marseille-Provence (19/92) are partial by design. 16,150 is close to the 16,186 communes that the CDC description gives for the 149 TI before the 2023 renewal. - `year` is the generation's start year. The dates are not verified here: first-generation launch late 2018, renewal 2023. ### eligible.csv: the communes that met each programme's criteria (34,746 communes) Built 2026-10-15 by `build/eligible.ml` (command in `build/dune`) for the controls of `did.csv`: communes that met a programme's published criteria but were not chosen. One row per metropolitan commune in COG 2025 (the Insee files' geography), with the inputs of each rule, so a different rule can be applied without downloading again. The programme lists stay in their own files; "eligible" does not say whether a commune was chosen. **Inputs**, downloaded 2026-10-15 from insee.fr, each in its own directory outside the tree (SHA-256 of the archive as downloaded): - Bassins de vie 2022 at 1 January 2025, https://www.insee.fr/fr/statistiques/fichier/6676988/BV2022_au_01-01-2025.zip (`b706e01bc15fcf1faa522772a417a38de73cf95af162f3cf5097b113b252d03a`), sheet `Composition_communale`: `TYPE_COMMUNE_BV2022` (20 pôle, 11 pôle partiel, 12 commune associée, 00 non-pôle). - Aires d'attraction des villes 2020 at 1 January 2025, https://www.insee.fr/fr/statistiques/fichier/4803954/AAV2020_au_01-01-2025.zip (`738f81b12832fe294bd24ddbfe3a31830c20d850143fc8b4393f18d4b8ee8181`): `AAV2020` and `CATEAAV2020` (11 commune-centre, 12 other commune of the main pôle, 13 secondary pôle, 20 couronne, 30 outside). - Unités urbaines 2020 at 1 January 2025, https://www.insee.fr/fr/statistiques/fichier/4802589/UU2020_au_01-01-2025.zip (`bc43d62e67c5dc6ae25a20fc3bd440dd37408ef8313b543aa1e9cb366be44840`): `STATUT_COM_UU` (C ville-centre, B banlieue, I ville isolée, H outside), and from sheet `UU2020` the size class `TUU2017` (4: 20,000-49,999 inhabitants … 7: 200,000-1,999,999; 8 Paris). - Historical legal populations, https://www.insee.fr/fr/statistiques/fichier/3698339/base-pop-historiques-1876-2023.xlsx (`07c992bc62c3136ab4db21647585d370666f6e56d40bad73bca43a6a5fc1d837`), sheet `pop_1876_2023`: `PMUN2015` (in force on 1 January 2018) and `PMUN2016` (in force on 1 January 2019). Paris, Lyon and Marseille appear by arrondissement and are summed. - EPCI at 1 January 2025, https://www.insee.fr/fr/statistiques/fichier/2510634/epci_au_01-01-2025.zip (`06510f6199943196b6f28c45ef65003243e73684063fd2ce21c1c35de5731696`): the commune's `EPCI`. - Jobs located in the commune and industrial jobs, 2015 census, from `../zrr/census.csv`. Each URL was fetched again on 2026-10-15 and gave the same checksum. The AU2010 and UU2010 zonings at the 2018 geography (`AU2010_au_01-01-2018_V2.zip`, `UU2010_au_01-01-2018.zip`) were also fetched but not used, since they are not in COG 2025. **The criteria, as published, and the rule applied.** - **Action cœur de ville.** The programme presentation of 14 December 2017, annexed to the instruction of 10 January 2018 (NOR TERR1800859C, BO MTES-MCT 2018/1, https://www.bulletin-officiel.developpement-durable.gouv.fr/documents/Bulletinofficiel-0030064/met_20180001_0000_0014.pdf), targets « villes de rayonnement régional qui ont un rôle de centralité pour leur bassin de vie », « hors d'une métropole », and says « Le plan ne prévoit ni liste de villes fixée au niveau central, ni seuil démographique minimal ». The instruction addresses « en priorité des villes « pôles d'attractivité », hors périmètre des métropoles »; its other criteria (political backing, a project, engineering capacity) are not observable. The Cour des comptes (S2022-1266, 29 September 2022, https://www.ccomptes.fr/system/files/2022-09/20220929-S2022-1266-1-programme-action-coeur-ville.pdf) reports that « le critère principal était celui du rôle de centralité », that a band of 20,000 to 100,000 inhabitants was an « orientation générale … pouvant être tempéré », that nearly 80 chosen towns are below 20,000, and that the CGET's work on the definition of villes moyennes produced a first list of « territoires cibles », which was not published. The CGET defined villes moyennes (2018, read as quoted by Warnant, Fondation Jean-Jaurès, 2020; the CGET site is gone) as « unité urbaine de plus de 20 000 habitants de type « grand pôle urbain » au sens de l'Insee, non incluse dans une des 22 aires urbaines englobant les métropoles institutionnelles ». The rule applies that definition on the 2020 zonings: the ville-centre (or ville isolée) of an unité urbaine of size class 4 to 7, whose communes held at least 10,000 jobs in 2015 (Insee's grand pôle threshold), outside the aires d'attraction of the 22 métropoles of 2018. It leaves out the prefectures the CGET added in six départements without such an unité urbaine. **Result:** 211 communes; 135 of the 218 metropolitan towns of the 2018 list (62%) meet it, the others mostly being villes-centres of unités urbaines of 10,000-19,999 (33) or communes of a binôme that are not the ville-centre; 76 eligible communes were not chosen, of which 23 joined Petites villes de demain. - **Petites villes de demain.** ANCT, « Comment candidater ? » (15 October 2020, https://anct.gouv.fr/actualites/comment-candidater): « Etre une petite ville de -20 000 habitants, avec des fonctions de centralité et présentant des signes de fragilité ». The Sénat (rapport n° 40, 2025-2026, https://www.senat.fr/rap/r25-040/r25-0407.html) adds that communes were to be « ne bénéficiant pas du programme « Action coeur de ville » ». The ANCT-INRAE study of centralities (October 2020), from which the prefects' pre-list of 2019 came, ranks centrality from the Base permanente des équipements but says it « ne constitue pas une sélection ou une éligibilité »; its commune file was not obtained, nor the fragility scores of the pre-list. The rule: legal population of 2016 below 20,000, the pôle or a pôle partiel of its bassin de vie 2022 or in the pôle of its aire d'attraction (CATEAAV2020 11-13), and not in the 2018 Action cœur de ville list. Fragility is not applied. **Result:** 2,717 communes; 1,092 of the 1,607 metropolitan PVD communes (68%) meet it, the others being mostly communes that are neither (515); 1,625 eligible communes were not chosen. - **Territoires d'industrie.** The Cour des comptes (November 2024, https://www.ccomptes.fr/sites/default/files/2024-11/20241121-Programme-Territoires-d-industrie.pdf) gives three criteria: « une part importante de l'industrie dans l'emploi local ; l'engagement de l'écosystème d'acteurs … ; et l'existence d'un potentiel important de développement », for an EPCI or group of EPCI proposed by the regions, with no threshold; every candidate of the 2023 generation was labelled. The criteria are loose and only the first is observable. The rule: in an EPCI (2025) where industry's share of the jobs located there in 2015 is at least its metropolitan share (11.79%). **Result:** 22,913 communes; 12,725 of the 16,096 metropolitan communes of the first generation (79%) meet it; 10,188 eligible communes were not in it, of which 2,461 joined in 2023. Columns: `insee_code`; `pop_2015`, `pop_2016` (legal populations, rounded); `bv2022_type`; `aav2020`, `aav2020_category`; `uu2020`, `uu2020_status`, `uu2020_size`; `epci`, `epci_industry_share` (2015, four decimals); `acv_eligible`, `pvd_eligible`, `ti_eligible` (`true` or `false`). Every commune has a population, a bassin de vie and an EPCI share. ### zfu.csv: Zones franches urbaines (100 ZFU, 149 source communes, 152 rows) - Source: ANCT, data.gouv 5a5784f9c751df1b5b1d530f, "Liste des communes inscrites (COG 2022)", last modified 2022-11-15: https://static.data.gouv.fr/resources/zones-franches-urbaines-zfu/20221115-155103/liste-zfu-com2022-20221115.csv It is a text CSV, so WebFetch did not save it. It was **transcribed from the WebFetch output** into `transcripts/liste-zfu-com2022-20221115.csv` (149 communes). The companion file https://static.data.gouv.fr/resources/zones-franches-urbaines-zfu/20221115-155347/liste-zfu-zfu2022-20221115.csv was transcribed independently and agrees. - Check against the official perimeter shapefile https://static.data.gouv.fr/resources/zones-franches-urbaines-zfu/20180117-110644/zfu.zip (`zfu_shp.zip`, 2016 layers): it has the same 100 ZFU codes, and the commune names per ZFU agree except for 3101ZZF (Douai), where the shapefile names 5 communes and the 2022 list names 9. The 2022 list was kept. Metadata are in `zfu_metadata.pdf`. - **Generation (inferred from the code pattern):** codes ending in digit+`ZF` or `ZFRAN` give 44 ZFU, `NZF` gives 41 and `ZZF` gives 15. This matches the 44/41/15 sizes of the 1997, 2004 and 2006 waves, which `year` uses. Légifrance confirms the legal bases: loi 2006-396 du 31 mars 2006 art. 26 and décret n° 2006-1623 du 19 décembre 2006 (3rd generation delimitation). CGI art. 44 octies A covers creations "entre le 1er janvier 2006 et le 31 décembre 2025", and art. 44 octies B (QPV) covers 2026-2030. The exact entry-into-force days (1 Jan 1997, 1 Jan 2004, 1 Aug 2006) were not verified. ### ber.csv: Bassins d'emploi à redynamiser (402 COG 2025 communes) - Legal list: décret n° 2007-228 du 20 février 2007, current version 1 July 2018 after décret n° 2018-550 du 29 juin 2018 (Légifrance https://www.legifrance.gouv.fr/loda/id/JORFTEXT000000465967 and https://www.legifrance.gouv.fr/loda/id/JORFTEXT000037131263). There are two bassins, "Zone d'emploi de la Vallée de la Meuse" (Ardennes) and Lavelanet (Ariège). The annex lists the communes with INSEE codes. - Communes: Insee zones d'emploi 1990, https://www.insee.fr/fr/statistiques/fichier/2114596/zones_emploi90.zip (`zones_emploi90.zip`, geography of 1 Jan 2009), zones 2101 (Vallée-de-la-Meuse, 362 communes) and 7319 (Lavelanet, 56 communes). **Check:** the 56 Lavelanet codes quoted from the Légifrance annex equal ZE1990 7319 exactly, and the 41 Meuse codes sampled from the annex (first and last entries) are all in ZE1990 2101 and are split across two ZE2010 zones. The BER therefore follow the 1990 zones. After projection to COG 2025 there are 346 + 56 communes. The full Meuse annex was not transcribed, so a commune listed in the annex but outside ZE1990 2101 would be missed. - `year` = 2007 (the decree). The regime's end date was not checked. ### frr.csv: France Ruralités Revitalisation (from existing data) - Built from `zrr/zrr_status.csv`, which was not downloaded again. That file comes from the DGCL list of 10 July 2025, and its notes are in `zrr/SOURCES.md`. The statuses are: - FRR (socle or FRR+), start 2024-07-01: 17,769 communes, plus 20 partly classified (`classified = partial`). - FRR+, 2025-01-01: 4,456. - FRR bénéficiaire, the former ZRR communes that keep the effects from 2024-07-01 to 2027-12-31: 2,134, plus 11 partial. ### qpv.csv: communes hosting a QPV - **2015 list (year 2015; 799 COG 2025 communes, 1,436 QPV):** CDC "Quartiers prioritaires de la politique de la ville au 1er janvier 2023", modified 2023-09-20, field `list_com_2023`. The codes are concatenated with no separator and are split every 5 characters: https://opendata.caissedesdepots.fr/api/explore/v2.1/catalog/datasets/liste-des-quartiers-prioritaires-de-la-ville-2023/exports/xlsx?select=qp,lib_qp,list_com_2023,list_lib_com_2023,list_comarr_2023,dep_1,dep_2 ANCT's official list of 1,514 QP 2015 has no commune codes: https://static.data.gouv.fr/resources/quartiers-prioritaires-de-la-politique-de-la-ville-qpv/20250404-081558/liste-1514qp2015.xlsx Compared with it, the CDC file **lacks 78 QP: 76 in French Polynesia (987) and 2 in Saint-Martin (978)**. - **2024 list (year 2024; 833 communes incl. COM codes, 1,584 QPV, metropole + DOM + Saint-Martin + Polynesia):** ANCT perimeters (data.gouv 5a561801c751df42d7fca9b6), resource "Périmètre des QPV 2024 (SHP)", last modified 2026-01-16: https://static.data.gouv.fr/resources/quartiers-prioritaires-de-la-politique-de-la-ville-qpv/20260115-205035/qpv-2024-shp.zip Attributes are from `QP2024_France_Hexagonale_Outre_Mer_WGS84.dbf`, field `insee_com`, with a comma-separated list for the 149 multi-commune QPV. The CDC 2024 file (`quartiers-prioritaires-de-la-politique-de-la-ville-au-1er-janvier-2024`) leaves the commune blank for 77 multi-commune QPV, so it was only used as a check. - `n_qpv` and `qpv_codes` are the QPV that touch the commune. The data have no population inside the QPV. ### pinel.csv: zonage A/B/C and Pinel eligibility, commune × year 2014-2026 Zoning vintages: | Vintage | In force | Source | |---|---|---| | Z2026 | from 26 June 2026 | MTE official list "Zonage ABC en vigueur 26 juin 2026", data.gouv 656715871172d08f8f680063: https://static.data.gouv.fr/resources/liste-des-communes-selon-le-zonage-abc/20260703-091327/liste-ensemble-des-communes-zonage-abc-en-vigueur-26-juin-2026.xlsx (34,875 communes, COG 2025) | | Z2025 | 5 Sept 2025 to 25 June 2026 | Z2026 with Alfortville (94002) set to A, its post-2025 zone in the reclassification file. The arrêté du 23 juin 2026 moved it to A bis. | | Z2024 | July 2024 to Sept 2025 | Z2025 with the "Zone avant révision 2025" column of the 468 reclassified + 19 declassified files: https://static.data.gouv.fr/resources/liste-des-communes-selon-le-zonage-abc/20250908-122705/liste-des-468-communes-reclassees-zonage-abc-5-septembre-2025.xlsx and https://static.data.gouv.fr/resources/liste-des-communes-selon-le-zonage-abc/20250908-122608/liste-des-19-communes-declassees-zonage-abc-5-septembre-2025.xlsx (2025 "after" zones = Z2025 for all 487 except Alfortville) | | Z2023 | Oct 2023 to July 2024 | Z2024 with the arrêté du 5 juillet 2024 undone (JO 11 July 2024, NOR TREL2418376A): https://static.data.gouv.fr/resources/liste-des-communes-selon-le-zonage-abc/20240722-102354/joe-20240711-0164-0041.pdf. The PDF was parsed with `pdftotext -layout`, which gives 866 additions (48 A bis, 142 A, 676 B1; the notice says 865) and 698 removals (42 A, 145 B1, 511 B2). Removed tables give the previous zone; an addition with no removal was C. Names are matched to codes by department and name across all COG vintages. | | Z2022 | 21 Feb 2022 to Oct 2023 | Kadata "Zonage ABC des communes - Tension du marché des logements", data.gouv 649ea362894adb8c5be51c81: https://static.data.gouv.fr/resources/zonage-abc-des-communes-tension-du-marche-des-logements/20230630-094302/zonage-abc-communes-2022.xlsx (34,955 communes, COG 2022; a third-party file). It already includes the February 2022 changes (Chamonix A, Courchevel B1, ...). | | Z2014/Z2019 | 2014 to Feb 2022 | Z2022, except the communes changed in between, which are left **blank**: Poitiers and Angers (arrêté du 4 juillet 2019) and the 57 communes of the arrêté du 16 février 2022 (Légifrance JORFTEXT000045190946). Their names came from WebFetch summaries of the Légifrance page (`transcripts/abc_arrete_2022-02-16_names.txt`). The total of 57 equals the count in the 2024 arrêté's notice, and all 57 match a commune in the 2022 file with the same zone. | - **Validation:** Z2023 differs from the 2022 file for 211 communes. The 2024 arrêté's notice says the arrêté du 2 octobre 2023 reclassified 209 communes. The other two differences are Saint-Denis (93066, merged with Pierrefitte in 2025, zone A|Abis) and Saint-Vincent-de-Tyrosse (40284). The 2022 file gives 40284 B1, yet the 2024 arrêté adds it to B1 without removing it from any table. This is a source conflict: its Z2014-Z2023 values are blank and `source_conflict = 1`. - The notice of the 2024 arrêté (verbatim) is the source for the revision history: "l'arrêté du 4 juillet 2019 a reclassé les communes de Poitiers et Angers en zone B1, l'arrêté du 16 février 2022 a révisé le classement de 57 communes ... l'arrêté du 2 octobre 2023 ... a abouti au reclassement de 209 communes". The visas add that the arrêté du 30 septembre 2014 also modified the zoning. Its content was not obtained and it is not applied. - Columns: - `zone_abc_jan1` is the zone in force on 1 January (blank for 2014). `zone_abc_dec31` is the zone on 31 December. - `pinel_by_right_dec31` is 1 if A bis, A or B1, for 2014-2024 only. - `b2_agrement_possible` is 1 for B2 in 2014-2017. Service-public F31151: B2 could qualify "s'il y a un besoin particulier en logement locatif" with "la délivrance d'un agrément de l'État". The agréments themselves were not collected. - From 2018 only A bis/A/B1 qualify. CGI art. 199 novovicies, as summarised from Légifrance, has a transitional rule for 2018 acquisitions with a permit filed by 31 December 2017. The scheme closed on 31 December 2024 (service-public). - Not covered: the 2009-2014 zoning used before Pinel (Scellier/Duflot) and the commune-level B2 agréments. ## Outcomes (item 2): sitadel.csv, dvf.csv, sirene_creations.csv Downloaded 2026-10-03/04 with WebFetch. The raw files are in `raw/outcomes/`. The DVF files were read from Parquet. All three outputs are mapped to COG 2025 with the Insee passage table. Codes are searched first in the data year's geography, and arrondissements are mapped to the city. ### sitadel.csv: housing authorised and started, commune × year, 2013-2025 - **Source:** SDES, DiDo dataset 6513ee3a3b05e5cd969c270f, "Logements autorisés et commencés, nombre et surfaces, séries annuelles (données non estimées)". Datafile 9c90a880-4ba0-49b4-b99d-d7dd6c810dd0 is "communal annual data", with "données en date réelle". Millésime 2026-09 was published 2026-09-29 and the dataset was last modified 2026-09-27. Temporal coverage is 2013-01-01 to 2025-12-31. - One xlsx per year: `https://data.statistiques.developpement-durable.gouv.fr/dido/api/v1/datafiles/9c90a880-4ba0-49b4-b99d-d7dd6c810dd0/xlsx?millesime=2026-09&withColumnName=true&withColumnDescription=false&withColumnUnit=false&ANNEE=eq:` for 2013 to 2025, saved as `sitadel_com_.xlsx` and converted to `.csv`. - Requests for `ANNEE=eq:2012` and `ANNEE=lt:2013` return HTTP 400. - **Gaps:** - **No commune series for 2010-2012.** The datafile description says "date réelle à partir de 2011", but the published file starts in 2013. The SDES monthly series that go back to 2000 (dataset 6513eed03b05e5cd969c2710) are estimated and only at national, regional and departmental level. - **`log_com_*` (starts) is empty for 2025 in the source.** - Real-date figures for the latest years are revised upward as late declarations arrive. National starts are 246k in 2023 and 211k in 2024, so treat 2023-2025 as provisional. - **Columns:** `insee_code` is as published, in the year's geography. `insee_code_cog2025` is blank for 38 rows: communes later split, plus 5 unknown codes. `log_aut_*` and `log_com_*` are housing units authorised and started, `sdp_*_total` is floor area in m², and the suffixes are `total`, `ip` (individuel pur), `ig` (individuel groupé), `coll` (collectif) and `res` (résidence). - Paris, Lyon and Marseille arrondissement rows (377) are dropped because the source also has whole-city rows. The type split sums exactly to the total in every row. - The file has 462,989 rows. National totals from the file: - authorised: 2013: 367,502; 2017: 460,062; 2021: 442,527; 2022: 450,363; 2024: 310,318; 2025: 358,902 - started: 2013: 311,119; 2017: 386,677; 2021: 351,208; 2024: 210,551 ### dvf.csv: median price per m² of dwellings sold alone, commune × year, 2016-2021 Rebuilt 2026-10-15 (replacing an earlier extract of this name that was never committed). Source: DGFiP, "Demandes de valeurs foncières", the yearly raw files 2016-2021 on data.gouv.fr (`https://static.data.gouv.fr/resources/demandes-de-valeurs-foncieres//valeursfoncieres-.txt[.zip]`, editions 20201026 for 2016-2018, 20240408 for 2019, 20250406 for 2020 and 20260405 for 2021, as downloaded by the `house` project into `pipeline/data/raw/dvf/raw/`), Licence Ouverte. Built by `dune exec data/places/build/dvf.exe -- data/places/dvf.csv valeursfoncieres-2016.txt ... ValeursFoncieres-2021.txt` (the 2020 and 2021 files unzipped first). - A sale is the run of consecutive lines with the same date, price, département and commune. It is kept when it is a "Vente" of exactly one dwelling (one distinct house or flat; outbuildings and land allowed); its price per m² is the price over the dwelling's built area, kept between €200 and €30,000. The median is over a commune's kept sales in the year. Arrondissements of Paris, Lyon and Marseille count in their city. - Columns: `insee_code` (as published, arrondissements merged), `year`, `sales`, `median_m2`. The analyses ignore a year with fewer than 5 sales. - DVF excludes Alsace-Moselle (57, 67, 68) and Mayotte. ### dvf_etalab.csv: the same price, from Etalab's statistics, 2021-2025 Source: Etalab, "Statistiques DVF" (`https://data-pipeline-open.s3.sbg.io.cloud.ovh.net/dvf/stats_dvf.csv`, fetched by `house` on 17 July 2026), monthly medians of price per m² of houses and flats (`med_prix_m2_apt_maison`) per commune; `house` collapses them to one row per commune and year, the sales-weighted mean of the monthly medians (`lib/fetch/dvf.ml`, file `pipeline/data/raw/dvf_prices.csv`). The commune rows of that file were copied unchanged, in its order, under the header `insee_code,year,sales,median_m2`: `{ echo insee_code,year,sales,median_m2; awk -F';' '$1=="commune"{print $2","$3","$4","$5}' dvf_prices.csv; } > data/places/dvf_etalab.csv`. **Not spliced to dvf.csv:** for the 13,904 communes with at least 10 sales in 2021 in both, the log of Etalab's 2021 median over the DVF files' has mean +0.010 and standard deviation 0.084, so a commune's series would jump by several per cent at the join. The analyses use each source on its own. ### lovac.csv: vacant private dwellings, commune × year, 2019-2025 Source: Cerema for the DGALN, "Logements vacants du parc privé par commune, département, région, France, de 2020 à 2026 (LOVAC)", data.gouv.fr dataset 61816c6e23197bb34835228e, file `https://static.data.gouv.fr/resources/logements-vacants-du-parc-prive-par-commune-departement-region-france/20260625-163627/lovac-opendata-communes26.csv` (updated 25 June 2026), from DGFiP's 1767 bis Com file crossed with Cerema's fichiers fonciers, Licence Ouverte. Built by `dune exec data/places/build/lovac.exe -- lovac-opendata-communes26.csv data/places/lovac.csv`. Millésime 20YY counts the dwellings on 1 January 20YY-1, so `year` runs 2019-2025 (the source's metadata date `pp_vacant_23` to 1 January 2023, but its `pp_vacant_plus_2ans_23`, from the same 1767 bis Com 2023, to 1 January 2022; the tool reads both as 2022). `vacant` and `vacant_2y` (vacant for more than two years) are empty where the source writes "s": counts under 11 are secret. Codes are COG 2026. The source notes changes in its inputs in 2023 (the end of the residence tax, GMBI declarations) and 2025. ### bpe.csv: shops and services by bassin de vie, 2015, 2020 and 2025 Source: Insee, base permanente des équipements, evolution file (Melodi `DS_BPE_EVOLUTION`, edition 2025, `https://api.insee.fr/melodi/file/DS_BPE_EVOLUTION/DS_BPE_EVOLUTION_2025_CSV_FR`), Licence Ouverte. Insee publishes it by bassin de vie 2022 and wider areas, not by commune, because commune counts are not comparable across BPE editions; it keeps the facility types counted alike in the three years. A bassin's code is its pôle's commune code. Built by `dune exec data/places/build/bpe.exe -- DS_BPE_EVOLUTION_2025_data.csv data/places/bpe.csv`: `shops` sums domain B (commerce), `services` domains A (services to individuals) and D (health and social action), over the types with no missing value in any bassin and year (44 types with a missing value are left out). ### sirene_creations.csv: establishments created, commune (COG 2025) × year, 2010-2025 - **Source:** OpenDataSoft `economicref-france-sirene-v3`, a copy of the Insee Sirene stock of establishments. It includes closed establishments ("Actif" and "Fermé") and creation dates up to 2026, so it is a 2026 extract. Its exact extraction date is not published. - The aggregate is computed by the server: `https://public.opendatasoft.com/api/explore/v2.1/catalog/datasets/economicref-france-sirene-v3/exports/csv?select=count(*) as n&where=datecreationetablissement>=date'2010-01-01' and datecreationetablissement_g1|g2.xlsx`. Each is `https://data.ofgl.fr/api/explore/v2.1/catalog/datasets/dotations-communes/exports/xlsx?select=exercice,code_insee,categorie,variable,valeur&where=exercice=date'' and variable in ()` (URL-encoded). The g1 list is 'Montant Dotation DGF', 'Population DGF', 'Population INSEE', 'Montant Dotation forfaitaire', 'Montant Dotation DSU', 'Montant Dotation DSR'. The g2 list is 'Montant global réparti - DSR Bourg-centre', 'Montant global réparti - DSR Péréquation', 'Montant global réparti - DSR Cible', 'Montant Dotation DNP', 'Potentiel financier par hab.', "Valeur de l'indice synthétique de classement de la commune à la DSU", 'Rang de classement à la DSU des communes métropolitaines de plus de 10 000 hab.', 'Rang de classement à la DSU des communes métropolitaines de 5 000 à 9 999 hab.'. The exact URLs are generated in the Bash history of this session, for example 2024 g1: https://data.ofgl.fr/api/explore/v2.1/catalog/datasets/dotations-communes/exports/xlsx?select=exercice%2Ccode_insee%2Ccategorie%2Cvariable%2Cvaleur&where=exercice%3Ddate%272024%27+and+variable+in+%28%27Montant+Dotation+DGF%27%2C%27Population+DGF%27%2C%27Population+INSEE%27%2C%27Montant+Dotation+forfaitaire%27%2C%27Montant+Dotation+DSU%27%2C%27Montant+Dotation+DSR%27%29 **Transformation.** The long format (one row per commune, year and variable) is pivoted to one row per `insee_code` × `year`. No value is computed; every cell is the DGCL value as exported. Units are euros, inhabitants, € per inhabitant, or index and rank. **Columns.** - `insee_code`: DGCL `code_insee`, in the geography of 1 January of the allocation year. Row counts match the COG of that year plus 86 Pacific and Saint-Pierre-et-Miquelon communes (codes 975xx/98xxx, which get the DACOM). Examples: 2024 has 35,021 = 34,935 + 86, and 2018 has 35,443 = 35,357 + 86. - `insee_code_cog2025`: the code mapped to COG 2025 through the Insee annual passage table (`zrr/raw/passage_2003_2026.csv`, from https://www.insee.fr/fr/statistiques/fichier/7671867/table_passage_annuelle_2026.zip). It is blank when the code is absent from the table (98xxx). It is also blank when a commune was later split; `cog2025_note` then lists the successors (20 rows). Amounts are **not** re-aggregated to COG 2025. - `pop_dgf` (Population DGF, CGCT L2334-2) and `pop_insee` (Insee total population used by DGCL for the year). - `dgf_total` (Montant Dotation DGF: all DGF to the commune, including forfaitaire, DSU, DSR, DNP, DACOM and dotation commune nouvelle), `dotation_forfaitaire`, `dsu`, `dsr`, `dsr_bourg_centre`, `dsr_perequation` and `dsr_cible` (each "Montant global réparti", exit guarantees included), `dnp` (dotation nationale de péréquation). - `potentiel_financier_par_hab`. - `dsu_indice_synthetique`, `dsu_rang_10000_plus`, `dsu_rang_5000_9999`: 2020-2026 only (not in the 2018-2019 reduced files). In 2023 the two rank variables are filled for about 34,800 communes, while other years cover about 1,050 and 1,270 ranked communes; this is as published. **Blank cells.** A commune has no row in the source for a component it does not receive (for example, 2024 has 860 DSU rows and 33,145 DSR rows). Such cells are left blank, not set to 0. For amounts, a blank effectively means no allocation. **Rows.** 315,619 (2018: 35,443; 2019: 35,056; 2020: 35,054; 2021: 35,051; 2022: 35,041; 2023: 35,031; 2024: 35,021; 2025: 34,961; 2026: 34,961). **Check.** Summed over communes, `dsr_bourg_centre + dsr_perequation + dsr_cible` equals `dsr` (2024: 751.2 + 862.1 + 465.5 = 2,078.7 M€). National totals for 2024: DGF 12,576 M€, forfaitaire 6,749, DSU 2,619, DSR 2,079, DNP 741. **Gap: DGF components for 2011-2017 were not obtained.** The DGCL consultation site was unreachable. The DGCL "critères de répartition" CSVs for 2016-2019 on data.gouv.fr (dataset 5e174be4634f414cd8a244f1, https://www.data.gouv.fr/fr/datasets/5e174be4634f414cd8a244f1/) contain only criteria (populations, potentiel financier, DSU/DSR-cible indices and ranks), not amounts. They are also served as text/csv, which this toolchain could not save. For 2011-2017, `accounts.csv` holds the DGF *as booked* by each commune (DGFiP `dgf`), with no split into forfaitaire, DSU and DSR. Not used: the DataSpot/Le Télégramme file "DGF des communes 2014-2016" on data.gouv.fr (https://static.data.gouv.fr/resources/la-dotation-globale-de-fonctionnement-dgf-des-communes-2014-2016/20160602-151224/DotationGlobaleFonctionnement2014_2016.xlsx), which is over the 10 MB fetch limit and is a press-made dataset. Also saved, for reference only: LexImpact, "Montants des dotations de fonctionnement aux communes", national parameters (OpenFisca YAML) 2012-2025, https://static.data.gouv.fr/resources/montants-des-dotations-de-fonctionnement-aux-communes/20260205-184044/openfisca-france-dotations-locales-extract-20260205.zip (data.gouv dataset 6984e42c4abb23c7072e5f41, modified 2026-02-05) → `raw/grants/leximpact_openfisca_dotations_locales_extract_20260205.zip`. It holds national envelopes, not commune values. ### accounts.csv: commune budget accounts (principal budget), 2011-2025 Two sources, flagged in `source`. **2018 appears in both** so the two can be spliced. #### OFGL, 2018-2025 (`source = OFGL`) - Dataset `ofgl-base-communes`, "Comptes des communes 2018-2025"; OFGL computes the aggregates from DGFiP balances. Modified 2026-07-29T07:17:27Z. OFGL notes that 97 communes lacked 2025 accounts at that update. Page: https://data.ofgl.fr/explore/dataset/ofgl-base-communes/ - Exports: two per year, `raw/grants/ofgl_comptes/ofgl_comptes__A|B.xlsx`, of the form `https://data.ofgl.fr/api/explore/v2.1/catalog/datasets/ofgl-base-communes/exports/xlsx?select=exer,insee,com_code,agregat,montant,euros_par_habitant,ptot&where=exer=date'' and type_de_budget='Budget principal' and agregat in (...)`. A covers 'Dépenses totales', 'Dépenses totales hors remb', 'Dépenses de fonctionnement', "Dépenses d'investissement", "Dépenses d'investissement hors remb". B covers "Dépenses d'équipement", 'Frais de personnel', 'Recettes de fonctionnement', 'Dotation globale de fonctionnement', 'Encours de dette'. - Only the **budget principal** is kept; annex budgets (water, sanitation...) are excluded. OFGL also publishes a consolidated main-plus-annex dataset, `ofgl-base-communes-consolidee`, which was not used. - `insee_code` = OFGL `insee` (the commune code in that year's accounts); `insee_code_cog2025` = OFGL `com_code` ("Code Insee 2025 Commune"). - `population` = OFGL `ptot` (Insee total population on 1 January), the denominator of `*_eur_hab`. `*_eur` = `montant` in euros; `*_eur_hab` = `euros_par_habitant` as published. - Definitions follow OFGL's aggregates (real flows, excluding *ordre* entries): https://data.ofgl.fr/explore/dataset/methodologie-ofgl-definitions-agregats-financiers/ (not downloaded). `dep_totales` = dépenses de fonctionnement + dépenses d'investissement (including debt repayment); `dep_totales_hors_remb` excludes debt repayment. #### DGFiP "comptes individuels des communes", 2011-2018 (`source = DGFiP`) - Publisher DGFiP, Licence Ouverte v2.0, on data.economie.gouv.fr: - `comptes-individuels-des-communes-fichier-global-2011-2015` (modified 2025-10-07T15:59:36Z, 183,432 records), https://data.economie.gouv.fr/explore/dataset/comptes-individuels-des-communes-fichier-global-2011-2015/ - `...-2016` (modified 2025-10-02T14:15:49Z, 35,857), `...-2017` (2025-10-06T18:00:30Z, 35,409), `...-2018` (2025-10-06T19:01:04Z, 35,355) - Exports: one per year, `raw/grants/dgfip_comptes/dgfip_comptes_.xlsx`, as `https://data.economie.gouv.fr/api/explore/v2.1/catalog/datasets//exports/xlsx?select=an,dep,icom,inom,pop1,prod,fprod,charge,fcharge,perso,fperso,equip,fequip,depinv,fdepinv,remb,fremb,dgf,fdgf,dette,fdette&where=an=""`. - Variable dictionary: `Maquette_commune.ods`, attachment of the 2016 dataset, https://data.economie.gouv.fr/api/explore/v2.1/catalog/datasets/comptes-individuels-des-communes-fichier-global-2016/attachments/maquette_commune_ods → `raw/grants/dgfip_maquette_commune.ods`. It says: "Population légale en vigueur au 1er janvier de l'exercice : pop1 - Budget principal seul". Amounts are "En milliers d'Euros", and the `f*` columns are "Euros par habitant". `prod` = "TOTAL DES PRODUITS DE FONCTIONNEMENT", `charge` = "TOTAL DES CHARGES DE FONCTIONNEMENT", `perso` = "Charges de personnel", `depinv` = "TOTAL DES EMPLOIS D'INVESTISSEMENT", `equip` = "Dépenses d'équipement", `remb` = "Remboursement d'emprunts et dettes assimilées", `dgf` = "Dotation globale de fonctionnement", `dette` = "Encours total de la dette au 31 décembre N". - Transformations: thousand-euro amounts × 1000 → `*_eur`; `f*` → `*_eur_hab` unchanged; `pop1` → `population` (legal population). `dgfip_charges_plus_emplois_invest_eur` = (`charge` + `depinv`) × 1000 is the only computed column. The INSEE code is `dep[1:3] + icom` for metropolitan departments (`067`+`100` → `67100`, `02A` → `2A`). For overseas departments (DGFiP dep `101`-`106`) it is `97` + `icom`; for example dep `102` / icom `302` = Cayenne 97302 (the DGFiP dep numbering of DOM changes across years, but `icom` always carries the Insee digits). DGFiP dep `143`, `161`-`163` (101 Pacific communes in 2011) are dropped. `insee_code_cog2025` comes from the Insee passage table; it is blank for 1-8 codes per year that are not in the table. - Column names shared with OFGL: `frais_personnel`, `dep_equipement`, `recettes_fonctionnement`, `dgf`, `encours_dette`. DGFiP-only columns: `charges_fonctionnement`, `emplois_investissement`, `remboursement_emprunts`, `dgfip_charges_plus_emplois_invest_eur`. **Comparability (2018 overlap, national sums).** Frais de personnel: DGFiP 37.63 bn€, OFGL 37.63. DGF: 11.68 and 11.68. Équipement: DGFiP 22.26, OFGL 19.96. Fonctionnement: DGFiP charges 74.77 vs OFGL dépenses réelles 67.58. The DGFiP totals include *ordre* entries (amortisation and the like), so **do not chain DGFiP `charges_fonctionnement` to OFGL `dep_fonctionnement` without calibrating on 2018.** Staff spending and DGF chain directly. **Rows.** 569,817 (DGFiP: 2011 36,682; 2012 36,682; 2013 36,639; 2014 36,675; 2015 36,653; 2016 35,857; 2017 35,409; 2018 35,355. OFGL: 2018 35,356; 2019 34,966; 2020 34,968; 2021 34,965; 2022 34,955; 2023 34,945; 2024 34,932; 2025 34,778). Not in OFGL: 2012-2017, which OFGL does not publish on its portal (its base starts in 2018). ### thresholds.md Légifrance articles read on 2026-10-03; each URL and version date is in the file. Search URL pattern used to find the IDs: `https://www.legifrance.gouv.fr/search/code?tab_selection=code&searchField=NUM_ARTICLE&query=
&searchType=ALL&typePagination=DEFAULT&pageSize=10&page=1`. Older versions were read by appending `/` to an article URL. ## Results: did.csv, event studies and synthetic difference-in-differences Written 2026-10-15 by `analysis/places_did.ml` (`places_did.exe data/places`, about 1.5 minutes on a loaded machine), beside the matched estimates of `results.csv`, with the estimators of `lib/did.mli`. Seed 2018; 999 département-cluster bootstrap draws per event study; synthetic DiD standard errors by the leave-one-département-out jackknife (200 placebo draws are also computed and logged, see below). **What the staggering buys here, and what it does not.** The programmes barely stagger: Action cœur de ville is one cohort (2018, plus 11 later additions), Petites villes de demain is in effect one (1,559 of 1,637 conventions signed in 2021) and Territoires d'industrie two (2018, and 2023 for the communes that joined with the second generation). Callaway and Sant'Anna's estimator does not fix a bias from staggering that is barely there; what it adds is an effect for every year before and after, each measured from the year before the cohort's start, so pre-trends are shown year by year rather than by one placebo, and the not-yet-treated comparison (which differs from the never-treated one only through the later cohorts). Synthetic DiD reweights controls and years instead of pairing communes. The eligible controls compare the chosen towns with towns that met the same published criteria (`eligible.csv`). **Designs.** For each programme, outcome and control set: - `cs-never`, `cs-not-yet`: ATT(g, t) for each cohort g and year t, compared within the region, from 2006 (the first Urssaf year; IRCOM series are cut there too) to 2025 (2024 for IRCOM), aggregated by event time (rows with `cohort = all`, weighted by treated communes) and over every post-treatment cell (the row with no `event_time`, which carries `pre_trend_p`: the joint Wald test that every event-time coefficient from the first year to e = -2 is zero, with the bootstrap covariance; 0.0010 is the smallest value 999 draws can give). Event times are kept from 2006 - s to 2025 - s, s the main cohort's start, so that a cohort of a few communes does not alone carry an event time; the cells (rows with a `cohort` year) keep them all. The `cs-not-yet` first-period coefficient (event time 0) is de Chaisemartin and D'Haultfœuille's DID_M. - `sdid`: synthetic DiD of the main cohort (ACV 2018, PVD 2021, TI 2018) on the communes observed every year of the window (the counts left out are in the log); `cohort` is that year, the row with no `event_time` the average effect, the others each year's effect against the time-weighted base. - Outcomes as `results.csv` measures them: `jobs` and `establishments` as log(1 + x), `households` and `income` (per household) as logs; for ACV and TI also the census pair `population` and `workplace_jobs` (2015 to 2021, one long difference, `event_time` 3, `year` 2021), which has no pre-period and so no pre-trend test. - Control sets: `far`, communes in none of the three programmes and more than 10 km from a commune of the programme evaluated (as `zrr.ml` keeps its controls away from the zone); `eligible`, communes that met the programme's criteria in `eligible.csv` and were not chosen, and for ACV and PVD are in neither of the two. Members without a signature date are left out. Treatment is absorbing: the first TI generation stays treated after 2023. **Standard errors of synthetic DiD.** The placebo standard error of the paper (controls drawn as fake treated units) assumes the treated units are as noisy as the controls. Here they are towns and the controls mostly small communes: for jobs, the placebo gives 0.0285 for ACV against far controls and the département jackknife 0.0070; for PVD, 0.0117 and 0.0069 (far) and 0.0158 and 0.0049 (eligible). The placebo is not defined where controls are fewer than treated units (ACV eligible, TI). The jackknife also respects the clustering the event-study bootstrap uses, so it is the one reported; it holds the weights fixed, as `synthdid` does. With pools of thousands of small communes, the noise level that sets the ridge penalty is high and the unit weights stay close to uniform (12,123 of 12,140 far controls weighted for ACV jobs): the reweighting works mostly through the time weights. **Reading (jobs).** Overall ATT, 95% interval, pre-trend p: | Programme | Controls | cs-never | cs-not-yet | sdid | matched (2025, `results.csv`) | |---|---|---|---|---|---| | ACV | far | -0.040 [-0.051, -0.027], p 0.001 | -0.040 [-0.051, -0.027] | -0.034 [-0.048, -0.020] | -0.039 [-0.086, +0.008] | | ACV | eligible | -0.018 [-0.039, -0.002], p 0.001 | -0.031 [-0.059, -0.008] | -0.007 [-0.026, +0.012] | | | PVD | far | -0.036 [-0.052, -0.023], p 0.001 | -0.036 [-0.052, -0.023] | -0.032 [-0.046, -0.019] | +0.000 [-0.017, +0.017] | | PVD | eligible | -0.020 [-0.030, -0.012], p 0.001 | -0.020 [-0.030, -0.012] | -0.015 [-0.024, -0.005] | | | TI | far | -0.007 [-0.025, +0.013], p 0.21 | -0.008 [-0.021, +0.005] | -0.013 [-0.031, +0.004] | +0.004 [-0.021, +0.029] | | TI | eligible | -0.001 [-0.015, +0.012], p 0.52 | -0.003 [-0.015, +0.008] | +0.001 [-0.014, +0.016] | | - **Territoires d'industrie:** no effect on jobs by any estimator or control set, and no pre-trend detected (p 0.17 to 0.61): the matched null holds. - **Action cœur de ville and Petites villes de demain:** the event studies reject parallel pre-trends with every control set (p = 0.001), so their negative averages cannot be read as effects. Against eligible towns the chosen ones were already losing ground before the start, by about 0.8% a year (ACV +0.094 at e = -12 to +0.009 at e = -2; PVD the same), and the post-treatment coefficients continue that slope or flatten it (ACV -0.010 at e = 0 to -0.012 at e = 7). Against far controls the pre-period is not a trend but a rise and fall (ACV -0.04 at e = -9 to -0.008 at e = -2; PVD +0.024 at e = -4 to +0.008 at e = -2). Synthetic DiD, which fits the pre-period by construction, finds nothing for ACV against eligible towns (-0.007) and -1.5% for PVD, a drop that appears at once in the signature year (-0.005 in 2021, -0.029 in 2025 against eligible; -0.024 in 2021 against far), as a selection on a 2020-2021 shock would and a programme of engineering support would not. The matched estimates, which pair on jobs in three earlier years, find no effect for either. The reading kept is that of no measurable effect on private jobs, with the caveat that these designs cannot separate a small effect from the towns' own decline.