A composition question with a carefully limited answer
France’s official bookstore directory records more than locations. It also assigns an assortment label to each entry, creating an opportunity to examine the kinds of book retail represented in the file. In the version retrieved on 30 October 2026, 2,114 of 3,237 records carry the label Généraliste. Another 984 carry one of nine named non-generalist assortment labels, while 139 carry Autres. Those groups represent 65.3%, 30.4% and 4.3% of the records respectively. These percentages describe the directory, not the complete population of physical bookstores.
That distinction is the foundation of this analysis. The Ministry explicitly describes the base as incomplete, and its public file does not contain the stable establishment identifiers or street addresses needed to resolve repeated names reliably. We therefore ask a narrower question: how does assortment composition appear across the directory’s geographic labels, and how much does that description change under transparent treatments of repeated records? The result is a category-distribution study, not a count of verified premises, a measure of consumer access, or a league table of places with better bookstores.
What a record can and cannot tell us
The public file supplies eight fields: bookstore name, assortment, reference label, commune, commune code, postal code, department and region. Every record has a stored assortment value, so there is no blank assortment cell to allocate or discard. Nevertheless, a populated category is not the same as a complete description of a shop. A generalist entry can sell comics; an entry classified under comics and manga can stock other books. The file provides neither shelf space nor a catalogue of titles.
The rows also do not establish a one-to-one correspondence with physical premises. There are repeated names within the same commune, and some of these groups contain different assortment labels. Separate branches, different departments of a business, repeated entries and other explanations cannot be distinguished from the released fields alone. Consequently, we retain the source row as the primary unit and keep identity-sensitive alternatives visible. We do not quietly merge names, count a brand once, or assume that different assortment labels prove either distinct shops or an error in the source.
Eleven labels, not a binary specialist market
The analysis preserves all eleven original labels. Alongside Généraliste and Autres, they are BD et Mangas, Religion, Jeunesse, Arts - Musées et Monuments, Mieux-être et spiritualité, VO, Imaginaire - Policier, SHS and Voyage. The supplementary term “named non-generalist” means only that a record carries one of those nine named labels rather than Généraliste or Autres. It is not an official certification of specialist status, and it does not turn every label into a literary genre.
This matters because the categories are heterogeneous. Some refer to a subject or audience, others to language or to an institutional setting. We retain their French source labels in the downloadable matrix rather than invent a uniform taxonomy. Autres is a recorded residual category with unspecified assortment; it is not missing data and cannot safely be assigned to either generalist or named non-generalist. The three-way summary is useful for navigation, but the eleven-category distribution is the substantive result. Readers interested in particular kinds of books should examine that detailed distribution rather than infer content from the broad partition alone.
Comics and manga are the largest named category
BD et Mangas accounts for 473 records: 14.6% of all directory rows and 48.1% of the 984 rows in the named non-generalist group. Religion contributes 148 records, or 15.0% of that group, and Jeunesse contributes 128, or 13.0%. Arts - Musées et Monuments adds 94 records, representing 9.6%. Together, these figures show that the named portion of the directory is not spread evenly across its available labels.
The remaining five named labels comprise Mieux-être et spiritualité with 47 records, VO with 41, Imaginaire - Policier with 25, SHS with 18, and Voyage with 10. Their shares of the named total are approximately 4.8%, 4.2%, 2.5%, 1.8% and 1.0%. These are conditional shares: the denominator excludes Généraliste and Autres. They must not be confused with shares of all records, and neither denominator measures sales. A small category in this register can still be important to readers, while a large recorded category does not establish a larger turnover, customer base or inventory.
Geography without selective winners
Every source geography label is retained. The file contains 21 distinct values in its Region field and 104 in its Departement field, including overseas territorial labels. These are therefore described as source-region and source-department labels, not as a newly asserted count of France’s administrative regions or departments. The downloadable matrix includes every category for every label, including zero category counts, under each of the three record treatments.
Across the raw file, named non-generalist records occur in 18 source-region labels and 99 source-department labels. BD et Mangas appears in 14 and 87 respectively. These statements concern presence within the observed file only. A zero means that no matching row appears in this release for that source label; it does not establish that a kind of bookstore is absent from the territory. The same caution applies to the size of a category: differences can reflect registration and classification practices as well as the underlying retail landscape. No population adjustment or geographic coverage correction is applied.
Similar broad shares can hide different assortments
A broad generalist-versus-other comparison conceals meaningful differences inside the named group. For example, the Île-de-France rows contain 103 BD et Mangas entries and 48 Arts - Musées et Monuments entries. Bretagne contains 34 BD et Mangas entries and 11 Jeunesse entries; Grand Est contains 32 BD et Mangas entries and 17 Religion entries. These examples are descriptions of different category combinations, not claims that one region offers better retail choice.
They also illustrate why the full matrix matters. Two geographic groups could have the same named non-generalist share while the categories contributing to that share differ substantially. Conversely, a smaller named share could still contain a label that is relatively prominent within that subset. We therefore publish both the numerator for each label and its all-record denominator. Conditional within-named shares can be reproduced from these counts, but they should be clearly labelled when used. The observations do not establish differences in the number of available titles, opening hours, accessibility or the quality of advice offered by booksellers.
Three scenarios, with no preferred identity correction
The primary scenario uses all 3,237 source rows. The second retains 3,195 distinct full rows after removing 42 occurrences that repeat every public field. This is called the exact-unique-row sensitivity, not a deduplicated count of shops. Two real premises could still have identical released fields because the distinguishing address is absent; conversely, different public rows could describe the same premises. Exact-row filtering answers a question about repeated representations in the file, not establishment identity.
The third scenario excludes every row belonging to a repeated exact bookstore-name and commune-code group. There are 64 such groups containing 135 rows, leaving 3,102 rows outside them. This is deliberately a different exercise from collapsing each group to one row. It asks what the composition looks like without any of these ambiguous groups. It does not declare that the excluded businesses are duplicates or that the remaining rows have verified unique identities. None of the scenarios is a lower or upper bound on the number of real bookstores, and none repairs incomplete registration.
The national pattern changes little under these treatments
In the exact-unique-row scenario, the counts are 2,092 Généraliste, 966 named non-generalist and 137 Autres. Their shares are 65.48%, 30.23% and 4.29%. Compared with the raw records, the named non-generalist share decreases by about 0.16 percentage points. After excluding all repeated name-and-commune groups, the corresponding counts are 2,047, 923 and 132. Their shares are 65.99%, 29.75% and 4.26%, and the named share is about 0.64 percentage points below the primary estimate.
These changes support a limited statement: the broad national composition represented by this file is similar under these two particular sensitivities. They do not validate the original unit assumptions, prove that unidentified duplicates are rare, or demonstrate that the directory mirrors the national retail market. The scenarios only test observable repetition patterns. They cannot test missing bookstores, inconsistent category assignment across contributors, or records that refer to the same place under different names. Readers should treat numerical stability as a property of the calculation, not as a general certificate of source completeness.
La Réunion shows why national stability is not enough
La Réunion provides a useful counterexample to any blanket robustness claim. Its raw records contain 15 Généraliste entries and 14 named non-generalist entries, with no Autres entries. The named share is 48.28% of 29 records. The exact-unique-row scenario retains 15 generalist and 12 named records, making the named share 44.44% of 27. Excluding repeated name-and-commune groups leaves 15 generalist and 10 named records, or 40.00% of 25.
The named share therefore changes by minus 3.83 percentage points in the exact-row sensitivity and minus 8.28 points in the repeated-group exclusion. That does not tell us which scenario is closest to the physical retail landscape. It tells us that a geographic description with a modest denominator can be sensitive to a few repeated representations. The complete regional table makes this visible rather than burying it beneath the stable national total. Other small source-region groups also require caution even when their sensitivity values are zero, because zero observed repetition is not evidence of comprehensive or verified coverage.
Unspecified assortment and mixed groups remain visible
Autres contributes 139 raw records nationally and occurs in 13 source-region labels and 41 source-department labels. Its presence is analytically important because reallocating those rows would change the apparent balance between broad groups. We do not assume that “other” means specialist, generalist, unimportant or erroneous. It is the source’s recorded category, with insufficient detail for a more specific assignment in this analysis. Every scenario carries it as a separate category and retains it in the all-record denominator.
A second kind of uncertainty concerns repeated name-and-commune groups with more than one assortment label. Eleven groups, containing 26 records, meet that description. They are reported separately from exact repeated rows because their ambiguity is different. The group might cover genuinely different premises or activities; the public fields cannot decide. It would therefore be inappropriate to select one category as the correct classification, or to turn the group into a multi-category establishment without additional evidence. Our primary distribution remains a distribution of stored row labels, and the third scenario removes whole repeated groups rather than resolving them by assumption.
Dates, coverage and reproducibility
The resource title identifies a March 2026-labelled CSV, and the catalogue records its publication and resource update on 20 April 2026. We retrieved the current file on 30 October 2026. The metadata also records a later technical Last-Modified header, dated 19 October. Those dates serve different purposes: the filename is not proof that every record was observed in March, and the current bytes should not be assumed identical to the first April release. The accompanying SHA-256 fingerprint identifies the precise file analysed.
Reproduction uses the official CSV and standard-library Python. It checks the fingerprint before calculating any output, preserves the original category and geography values, and writes only aggregate tables. A changed source causes the script to stop rather than silently produce a different study. The public matrix contains 4,158 rows: eleven categories across the national total, 21 source-region labels and 104 source-department labels, under three scenarios. These are aggregate analytical cells, not a redistributed bookstore directory. Counts reconcile to their scenario-specific denominators, and shares total 100% within each nonempty distribution before rounding.
What this analysis adds, and where it stops
The contribution is a reproducible view of assortment composition in one official release, with every geography retained and visible sensitivity to observable record ambiguity. It is not another locator map and does not claim to be the first use of bookstore assortment information. Its purpose is to make comparisons interpretable: readers can see which labels drive a broad share, how large the denominator is, and whether excluding repeated representations materially changes the result.
The scope stops well short of demand, profitability, survival, ownership or causal explanation. Reference labels do not identify every independent shop, and absent labels do not establish chain membership. No population file was joined, so the analysis cannot compare town sizes, per-capita provision or access outside large urban centres. The Ministry’s scope also excludes cultural-superstore chains, and incompleteness remains unresolved. The defensible conclusion is narrower but useful: generalist entries dominate this directory, comics and manga are the largest label (48.1%) among its named non-generalist records, and geographic composition should be read alongside denominators and identity-sensitive alternatives rather than converted into claims about the entire physical bookselling market.
All eleven recorded assortment labels
3,237 directory records · Raw rows · All source labels, in table order
Full chart and source
Scroll horizontally to inspect the chart. ↔
Chart and values
Scroll horizontally to see all columns.
| Source assortment label | Records | All-record share (%) |
|---|---|---|
| Arts - Musées et Monuments | 94 | 2.90 |
| Autres | 139 | 4.29 |
| BD et Mangas | 473 | 14.61 |
| Généraliste | 2,114 | 65.31 |
| Imaginaire - Policier | 25 | 0.77 |
| Jeunesse | 128 | 3.95 |
| Mieux-être et spiritualité | 47 | 1.45 |
| Religion | 148 | 4.57 |
| SHS | 18 | 0.56 |
| VO | 41 | 1.27 |
| Voyage | 10 | 0.31 |
Current official directory records, raw-row scenario; named non-generalist is an analytical grouping, not a certified specialist status. Autres is an unspecified recorded category.
Regional composition: sensitivity to repeated records
21 source Region labels · Named non-generalist share · Not verified shops
Full chart and source
Scroll horizontally to inspect the chart. ↔
Chart and values
Scroll horizontally to see all columns.
| Source Region label | N | Généraliste | Named non-generalist | Autres | Exact-row delta pp | Repeated-group exclusion delta pp | Mixed-assortment groups |
|---|---|---|---|---|---|---|---|
| AUVERGNE RHONE ALPES | 433 | 299 | 120 | 14 | −0.14 | −0.66 | 3 |
| BOURGOGNE FRANCHE COMTE | 131 | 90 | 39 | 2 | +0.00 | −1.09 | 0 |
| BRETAGNE | 219 | 146 | 63 | 10 | −0.33 | −0.66 | 0 |
| CENTRE VAL DE LOIRE | 99 | 70 | 28 | 1 | +0.29 | +0.58 | 0 |
| CORSE | 15 | 13 | 1 | 1 | +0.00 | +0.00 | 0 |
| GRAND EST | 198 | 126 | 68 | 4 | +0.19 | +0.39 | 0 |
| GUADELOUPE | 9 | 6 | 3 | 0 | +0.00 | +0.00 | 0 |
| GUYANE | 7 | 6 | 1 | 0 | +0.00 | +0.00 | 0 |
| HAUTS DE FRANCE | 170 | 112 | 45 | 13 | +0.00 | +0.00 | 0 |
| ILE DE FRANCE | 735 | 422 | 260 | 53 | −0.24 | −0.98 | 2 |
| LA REUNION | 29 | 15 | 14 | 0 | −3.83 | −8.28 | 0 |
| MARTINIQUE | 6 | 1 | 5 | 0 | +0.00 | +0.00 | 0 |
| MAYOTTE | 3 | 3 | 0 | 0 | +0.00 | +0.00 | 0 |
| NORMANDIE | 151 | 104 | 45 | 2 | −0.47 | −1.53 | 0 |
| NOUVELLE AQUITAINE | 284 | 207 | 72 | 5 | +0.10 | +0.11 | 1 |
| OCCITANIE | 316 | 222 | 80 | 14 | +0.00 | −0.89 | 2 |
| PAYS DE LA LOIRE | 150 | 89 | 55 | 6 | −1.29 | −1.70 | 0 |
| PROVENCE ALPES COTE D AZUR | 278 | 180 | 84 | 14 | +0.44 | +0.70 | 3 |
| Saint-Barthélémy | 2 | 1 | 1 | 0 | +0.00 | +0.00 | 0 |
| Saint-Martin | 1 | 1 | 0 | 0 | +0.00 | +0.00 | 0 |
| Saint-Pierre-et-Miquelon | 1 | 1 | 0 | 0 | +0.00 | +0.00 | 0 |
N/G/Named/Other are raw record counts. Deltas are named non-generalist percentage points relative to raw: exact unique full rows; exclusion of all repeated name/commune groups. Neither is verified entity resolution or a bound. Overseas territory labels preserved.
Data & downloads
Scroll horizontally to see all columns.
| Source assortment label | Records | All-record share (%) |
|---|---|---|
| Arts - Musées et Monuments | 94 | 2.90 |
| Autres | 139 | 4.29 |
| BD et Mangas | 473 | 14.61 |
| Généraliste | 2,114 | 65.31 |
| Imaginaire - Policier | 25 | 0.77 |
| Jeunesse | 128 | 3.95 |
| Mieux-être et spiritualité | 47 | 1.45 |
| Religion | 148 | 4.57 |
| SHS | 18 | 0.56 |
| VO | 41 | 1.27 |
| Voyage | 10 | 0.31 |
Current official directory records, raw-row scenario; named non-generalist is an analytical grouping, not a certified specialist status. Autres is an unspecified recorded category.
Scroll horizontally to see all columns.
| Source Region label | N | Généraliste | Named non-generalist | Autres | Exact-row delta pp | Repeated-group exclusion delta pp | Mixed-assortment groups |
|---|---|---|---|---|---|---|---|
| AUVERGNE RHONE ALPES | 433 | 299 | 120 | 14 | −0.14 | −0.66 | 3 |
| BOURGOGNE FRANCHE COMTE | 131 | 90 | 39 | 2 | +0.00 | −1.09 | 0 |
| BRETAGNE | 219 | 146 | 63 | 10 | −0.33 | −0.66 | 0 |
| CENTRE VAL DE LOIRE | 99 | 70 | 28 | 1 | +0.29 | +0.58 | 0 |
| CORSE | 15 | 13 | 1 | 1 | +0.00 | +0.00 | 0 |
| GRAND EST | 198 | 126 | 68 | 4 | +0.19 | +0.39 | 0 |
| GUADELOUPE | 9 | 6 | 3 | 0 | +0.00 | +0.00 | 0 |
| GUYANE | 7 | 6 | 1 | 0 | +0.00 | +0.00 | 0 |
| HAUTS DE FRANCE | 170 | 112 | 45 | 13 | +0.00 | +0.00 | 0 |
| ILE DE FRANCE | 735 | 422 | 260 | 53 | −0.24 | −0.98 | 2 |
| LA REUNION | 29 | 15 | 14 | 0 | −3.83 | −8.28 | 0 |
| MARTINIQUE | 6 | 1 | 5 | 0 | +0.00 | +0.00 | 0 |
| MAYOTTE | 3 | 3 | 0 | 0 | +0.00 | +0.00 | 0 |
| NORMANDIE | 151 | 104 | 45 | 2 | −0.47 | −1.53 | 0 |
| NOUVELLE AQUITAINE | 284 | 207 | 72 | 5 | +0.10 | +0.11 | 1 |
| OCCITANIE | 316 | 222 | 80 | 14 | +0.00 | −0.89 | 2 |
| PAYS DE LA LOIRE | 150 | 89 | 55 | 6 | −1.29 | −1.70 | 0 |
| PROVENCE ALPES COTE D AZUR | 278 | 180 | 84 | 14 | +0.44 | +0.70 | 3 |
| Saint-Barthélémy | 2 | 1 | 1 | 0 | +0.00 | +0.00 | 0 |
| Saint-Martin | 1 | 1 | 0 | 0 | +0.00 | +0.00 | 0 |
| Saint-Pierre-et-Miquelon | 1 | 1 | 0 | 0 | +0.00 | +0.00 | 0 |
N/G/Named/Other are raw record counts. Deltas are named non-generalist percentage points relative to raw: exact unique full rows; exclusion of all repeated name/commune groups. Neither is verified entity resolution or a bound. Overseas territory labels preserved.
Evidence & method
How we calculated it
- Source: Ministère de la Culture, Base des librairies en France, March 2026-labelled CSV; resource updated 20 April 2026, retrieved 30 October 2026. Hash-frozen current bytes, not an assumed historical census.
- Primary unit: public directory row. All eleven assortment labels, all 21 source Region labels and all 104 source Departement labels retained. Named non-generalist groups nine labels; Autres remains a separate recorded unspecified category.
- Three separate scenarios: all source rows, exact unique full rows, and exclusion of all repeated exact name/commune-code groups. No scenario establishes unique premises or bounds physical shop counts.
- Category share = 100 × category records / scenario-specific geography records. No population join, confidence intervals, ownership inference or causal interpretation.
What the data cannot tell us
- Non-exhaustive directory; no coverage correction. Ministry scope excludes cultural-superstore chains and includes group-owned bookstores.
- No SIRET or street address in public file; 42 repeated full-row occurrences and 64 repeated name/commune groups containing 135 rows. Eleven groups / 26 rows carry differing assortment labels.
- Autres is not blank. Source labels do not measure inventory, genre availability, demand, sales or profitability. No town-size or population-adjusted comparisons.
- March is a filename label; API temporal coverage and resource update refer to April, with a later October technical header. No change over time is inferred. Small geographic denominators and repetition sensitivity preclude blanket robustness claims.
Original sources
- Official dataset and definition
The Ministry describes the directory as non-exhaustive. Its definitions and inclusion rules do not independently verify one row as one physical shop.
Accessed October 30, 2026
- Official metadata API
Official metadata was checked on 30 October 2026. The resource is labelled March 2026 and its recorded update is 20 April 2026; these dates do not establish an immutable historical stock.
Accessed October 30, 2026
- Official source CSV
Current official CSV bytes were retrieved on 30 October 2026; SHA-256 748b616d4ce974e87035f529d584cc7334a5d39713f254698d79f9242767eea3. Record-level observation dates remain unverified.
Accessed October 30, 2026
- Source licence
Licence Ouverte 2.0 attribution and update-date requirements are retained; no endorsement or invented Creative Commons terms.
Accessed October 30, 2026
Cite & reuse this study
Alisson Bento. Beyond the shop count: assortment recorded in France’s bookstore directory. Secondary analysis of Ministère de la Culture, Base des librairies en France, March 2026-labelled resource updated 20 April 2026, current file retrieved 30 October 2026. https://alissonbento.com/research/france-bookstore-directory-assortment-2026/
Primary data: Ministère de la Culture, Base des librairies en France, resource updated 20 April 2026, under Licence Ouverte / Open Licence 2.0 (https://www.data.gouv.fr/pages/legal/licences/etalab-2-0). Calculations and English analysis: Alisson Bento. Derived category aggregation and record sensitivities; no official endorsement. The raw directory is not redistributed.
Corrections or editorial questions