Skip to content

Open scholarly books · Licence metadata · 2026 snapshot

What a book catalogue can tell us about reuse licences

A strict parser recognizes a Creative Commons family in 81.86% of book records in one official DOAB export. The remainder is not a count of unlicensed books: it includes external terms links, missing values and access-only statements.

Published
Share
Share this page
EmailCite & reuse this study
Key findings as text
Reusable imagesWhat a book catalogue can tell us about reuse licences — 81.86%: Book records with a strictly recognized CC-family statement. 85,108 of 103,963 records in the pinned CSV, not actual legal-permission prevalence.Download sharing image (JPG) ↓

Key findings

81.86%

Book records with a strictly recognized CC-family statement

85,108 of 103,963 records in the pinned CSV, not actual legal-permission prevalence.

7,270

Unresolved records with external terms or other URLs

Additional lookup required; not classified as unlicensed or closed.

7,552

Book records with no nonblank licence-field statement

Missing in the inspected representation, not proof of absent publisher terms.

306

Additional records recognized by case-folding sensitivity

Primary counts unchanged; 0.29 percentage points of the book denominator.

Access is not a complete description of reuse

A freely accessible book is not automatically a book whose contents can be reused in every way. Reading, copying, distributing modified material and using material commercially are different activities. The Directory of Open Access Books, or DOAB, explicitly distinguishes the accessibility of its listed books from the terms attached to their reuse. This study therefore asks a narrower, measurable question: how clearly does one official downloadable catalogue export represent book-content licence information? It measures statements in metadata, not the legal permissions ultimately available to a reader, researcher or publisher.

In the export retrieved on October 30, 2026, a conservative parser recognized an explicit Creative Commons family in 85,108 of 103,963 book records, or 81.86%. That is a result about recognizable metadata under stated rules. It is not an estimate of the percentage of books legally licensed under Creative Commons, and the remaining records must not be described as unlicensed or closed. Some contain links to external terms; others contain only an access statement, an empty field or a form the parser deliberately leaves unresolved. Those distinctions are the central result, rather than inconveniences to be removed from the denominator.

One catalogue representation and one record denominator

The primary source is a single complete response from the official DOAB CSV metadata endpoint. The downloaded response contained 114,618 rows, each with a distinct nonempty identifier and a distinct handle. Of those rows, 103,963 had the exact publication type book and 10,655 had the exact type chapter. The study excludes the chapter records from its main denominator rather than treating them as additional books. No book-type row carried a parent-book relationship in the inspected field, and no malformed row or duplicated identifier was found.

The resulting unit is a catalogue book record, not a unique intellectual work. Separate editions, versions or other bibliographic representations may exist without being duplicates in a record-based analysis. Similar titles and shared identifiers such as ISBNs are not used to merge records into an invented work hierarchy. Equally, the study does not claim that this CSV necessarily represents every entity available through every DOAB service. Its population is the eligible book records actually present in the pinned export. That boundary makes the count reproducible while avoiding a broader completeness claim that the retrieved documentation does not independently establish.

Metadata rights and book-content rights stay separate

DOAB makes its metadata feeds available under a CC0 dedication. That permission applies to the metadata service and supports reuse of the bibliographic information. It does not place every described book under CC0. The classification reads the export’s BITSTREAM_License field as a representation of content-related terms; it never substitutes the catalogue’s overall metadata policy when that field is empty, unclear or different. This is especially important because an automated workflow could otherwise convert missing licence evidence into a false permission claim.

DOAB is a discovery service rather than the host of the books themselves. Its records direct users to content on other platforms, including but not limited to the OAPEN Library. The study does not combine those platforms' catalogues, download their books or inspect their full texts. References to external download locations remain metadata, not an additional sample of content. The seven records with explicit public-domain designations in the content-related field are counted because those designations appear there, not because of DOAB’s CC0 metadata policy. Three identify CC0 and four identify the Public Domain Mark; those are kept apart from the six Creative Commons licence families.

What the strict classifier accepts

The classifier recognizes six familiar Creative Commons families: BY, BY-SA, BY-ND, BY-NC, BY-NC-SA and BY-NC-ND. It accepts supported structured URLs on the Creative Commons domain and narrowly specified textual labels. It retains distinctions between versioned references and family-only references. Supported legal-code and deed variants can identify the same family without being mistaken for additional licences. The implementation also retains selected older jurisdictional forms rather than forcing every historical reference into a current international version.

Strict recognition is intentionally narrower than human interpretation of every possible string. A misspelled domain, an unexpected path, an unsupported legacy form or a malformed link is not silently repaired. Uppercase path variants are also left outside the primary strict result, with a separate sensitivity calculation discussed below. These decisions do not mean that the underlying book lacks a licence or that the intended label is unknowable. They mean that a reproducible, bounded parser should distinguish what it accepts directly from what would require an additional assumption. The downloadable vocabulary preserves every observed book-field value and its frequency so that alternative rules can be examined openly.

The complete metadata-state distribution

The largest mutually exclusive state contains the 85,108 records with one recognized explicit Creative Commons family and no other kind of assertion in the same field. Another 7,552 records have no nonblank statement, representing 7.26% of the book denominator. Access-only statements account for 3,211 records, or 3.09%. A further 137 records contain a reserved-rights phrase without a recognized Creative Commons family, and seven contain only an explicit public-domain designation. The remaining 7,948 records, or 7.65%, fall outside the accepted forms and remain unresolved under the parser.

The predeclared state design also allows multiple recognized families and a recognized family combined with a different kind of statement. Neither state is observed in this export under these rules; their zero counts remain visible in the evidence table. A reserved-rights phrase is not treated as proof of illegality, closed access or absence of exceptions. Likewise, its appearance alongside an open licence would require a mixed-statement category rather than automatic cancellation of either statement. The state distribution describes representation, not a ranking of publishers or a compliance judgement about the books they publish.

Most unresolved records point somewhere else

The unresolved group is not homogeneous. Of its 7,948 records, 7,270 contain an external terms or other URL. These include links to platform or publisher pages rather than a directly recognized Creative Commons identifier. Such pages may provide detailed and valid terms once consulted. The study does not follow those links at scale, interpret their current contents or assume that a platform-wide policy applies identically to every book. Consequently, this category is best understood as an additional lookup requirement within the chosen workflow, not as missing rights in the underlying publication.

Another 663 records have Creative-Commons-looking strings that the strict parser does not accept. This includes nonstandard or malformed forms and does not establish that every such reference is unusable in an ordinary browser. The remaining fifteen contain other text or a relative link. Separating these groups changes the practical meaning of the headline. A single unknown category would conceal the difference between absent information, an access-only statement, a custom terms link and a recognizable but nonaccepted syntax. Those situations call for different forms of metadata review and cannot responsibly be converted into one estimate of books without reuse permission.

Licence families describe conditions, not a simple openness score

Among the eligible book records, the recognized BY-NC-ND family appears in 36,636 records, or 35.24% of the full book denominator. BY appears in 34,432 records, or 33.12%. The remaining recognized families are BY-NC in 5,095 records, BY-NC-SA in 4,132, BY-SA in 3,878 and BY-ND in 935. These figures retain all eligible records in the denominator rather than reporting only the distribution among successfully recognized Creative Commons statements. A reader can therefore see the family distribution without losing the scale of the unresolved and missing groups.

The family labels encode different conditions. The Creative Commons explanations distinguish attribution, noncommercial use, share-alike requirements and restrictions concerning distribution of adapted material. Those terms should not be compressed into a binary claim that a book is either reusable or not reusable. Applicable exceptions, public-domain elements and other rights may also matter. This analysis makes no legal recommendation for a particular use and does not inspect whether individual third-party illustrations or extracts are covered by a book-level statement. The families are reported as metadata labels with distinct conditions, not as a complete answer to every potential reuse scenario.

Condition flags can overlap

For analytical convenience, the study also records whether a recognized family contains an NC, ND or SA element. NC appears in 45,863 book records, ND in 37,571 and SA in 8,010. These are non-additive flags. A BY-NC-ND record contributes to both NC and ND, while a BY-NC-SA record contributes to both NC and SA. Adding the flags would double-count records and would not produce a meaningful total number of restricted books. The accessible table and download identify the common denominator and retain the family counts separately.

The family flags are also defined as non-additive in the general design because a future record could explicitly represent more than one family. In this particular CSV result, only one recognized family occurs per classified record, so the six family counts happen to sum to the recognized-family state. That arithmetic coincidence is not a reason to discard the more cautious design. Multiple licences can reflect legitimate alternatives or different bitstreams; the study does not assume that multiple families create a legal contradiction. It preserves statements before interpretation and avoids assigning a single supposed most restrictive licence to every record.

Specificity matters even after a family is recognized

A family label is not the same as a fully specified licence reference. Of the 85,108 recognized-family records, 85,079 include a supported versioned URI, twenty-seven have a family-only URI and two have an accepted text-only designation. These categories clarify what the parser actually recovered. A family-only reference can communicate a recognizable set of broad conditions while leaving the version unspecified in the inspected field. It should not be upgraded to a particular version simply because one version is common elsewhere in the catalogue.

The same caution applies to spelling and syntax. A sensitivity exercise that lowercases the source string would recognize 306 additional records, equivalent to approximately 0.29 percentage points of the full book denominator. The primary count remains unchanged. This exercise shows that one limited normalization rule affects a relatively small but measurable group; it does not validate all repaired links or justify correcting every malformed string. The distinction between direct recognition and normalization is useful for anyone building catalogue integrations, because different pipelines can produce different counts even when they begin with exactly the same source file.

Multiple statements and a checked representation difference

Thirty-two book records contain multiple semicolon-separated statements that resolve to the same recognized family. No record is classified as containing multiple recognized versions under the primary CSV rules. These observations describe the inspected field, not the full complexity of the underlying catalogue. Repeated links to a deed and its legal code are not automatically separate permissions. Conversely, a single CSV value cannot establish that no additional statement appears in another interface or in the source publication.

One deliberately checked example illustrates that limitation. The CSV record with handle 20.500.12854/26303 contains a BY-NC 4.0 URI. An official xoai response for the same handle contains BY-NC 3.0 and 4.0 references associated with the same download location in separate metadata structures. The xoai response is dated October 30, 2026, with a record datestamp in July 2025; the CSV retrieval has no independently confirmed generation timestamp. The representations differ in this checked case, but their timing and implementation are insufficient to establish a particular export-loss mechanism. Nor can one example estimate the prevalence of such differences. The defensible conclusion is that the CSV cannot be assumed exhaustive for every rights assertion.

Reproducibility without republishing the catalogue

The complete source response was retrieved once and preserved privately in its original byte form. Its size is 230,316,814 bytes, and its SHA256 fingerprint is recorded with the source URL and retrieval timestamp. The timestamp identifies when this analysis obtained the response, not the publication year of the books or a provider-certified generation time. The study is a cross-sectional snapshot and does not infer growth, improvement or deterioration from earlier catalogue counts. The initial fifty-five-row feasibility pilot was an ordered convenience prefix and contributes no population estimate to the results.

The compact evidence package includes aggregate vocabulary counts, calculated tables, source references and reproducible Python code. It does not distribute the entire catalogue, abstracts, thumbnails or full book texts. Default reproduction uses the derived value-frequency input to reconstruct the classifications and totals. Optional verification against an independently obtained official source file requires the recorded hash to match before comparing the pinned results. A later live export may legitimately differ; it should be treated as a new snapshot rather than silently substituted into this one. This separation allows transparent arithmetic without turning an analytical download into a bulk republication of source content.

What the validation establishes

The parsing checks include twenty-two synthetic cases covering missing values, ordinary licences, multiple statements, mixed reserved-rights statements, domain spoofing and malformed Creative Commons paths. A deterministic sixty-record sample reconciles original CSV fields with the derived vocabulary mapping. Separately, all 211 distinct book-field values reconstruct the state and family totals through their published frequencies. The identifier, type and row-shape checks also establish that the stated record denominator is internally consistent for this export. These are computational validation steps, not a claim of human legal review or a gold-standard assessment of publishers.

The remaining uncertainty is principally about representation and interpretation rather than sampling error. This is a complete analysis of the eligible records in one retrieved file, so the figures do not receive invented confidence intervals. Errors or omissions in source metadata, changes in external terms, differences between interfaces and limitations of the parser can still affect what the counts mean. No publisher-type comparison is attempted because the study has not established a defensible official classification for that purpose. Names, imprints and institutional branding are not used as proxies for commercial status, mission or licensing practice.

A practical conclusion about metadata, not permission

The study’s central finding is that one official book-catalogue representation supports direct, strict Creative Commons family recognition for 81.86% of its eligible book records, while the remainder comprises several substantively different metadata situations. The largest unresolved component points to external pages rather than offering a directly accepted family identifier. That result is more informative than a generic title count or a binary open-versus-closed classification. It describes the work needed to move from discovery metadata to a carefully specified reuse decision without pretending that the catalogue alone can complete that decision.

For catalogue users, the immediate lesson is to preserve provenance and uncertainty. Keep the original field, state the parser rule, distinguish missing information from custom terms, and avoid converting a readable download into assumed adaptation rights. For interpretation of an individual book, the relevant licence and its scope still need to be checked at the appropriate source. Alisson Bento’s analysis contributes a reproducible account of metadata representation, with its denominators and exceptions visible. It does not replace the publisher’s applicable terms, evaluate the legality of a proposed use or attribute a single rights policy to the scholarly book ecosystem as a whole.

Licence statements represented in book records

DOAB CSV retrieved 30 October 2026 · 103,963 book records · Strict recognition of metadata, not an assessment of actual reuse permission

Full chart and source
Mutually exclusive metadata states among 103,963 book records. Strictly recognized CC family: 85,108 of 103,963 (81.86%). One recognized CC family only: 85,108 (81.86%); No nonblank statement: 7,552 (7.26%); Access statement only: 3,211 (3.09%); Reserved phrase; no recognized CC family: 137 (0.13%); Other / unresolved by strict parser: 7,948 (7.65%); Explicit public-domain designation: 7 (0.01%); Multiple recognized CC families: 0 (0.00%); One CC family plus other assertion: 0 (0.00%). Unresolved metadata is not evidence of unlicensed or closed books. No legal-permission inference. CC0 metadata terms do not determine book-content terms. Counts are catalogue records, not works.

Scroll horizontally to inspect the chart. ↔

Chart and values

Scroll horizontally to see all columns.

CSV-represented book-record states
Metadata stateRecordsPercent of book records
One recognized CC family only85,10881.86
No nonblank statement7,5527.26
Access statement only3,2113.09
Reserved phrase; no recognized CC1370.13
Other / unresolved by strict parser7,9487.65
Explicit public-domain designation70.01
Multiple recognized CC families00.00
One CC family plus other assertion00.00

Mutually exclusive states; 103,963 eligible book records. Zero states retained. Unresolved does not mean unlicensed or closed. Rounding can prevent percentages summing to exactly 100.

All eight states retained. External terms or other URLs account for 7,270 of 7,948 unresolved records. CC0 metadata terms do not determine book-content terms. Counts are catalogue records, not works. DOAB metadata · CC0 1.0 · Original analysis: Alisson Bento

Download chart (SVG) ↓ · Download chart (PNG) ↓

Data & downloads

Scroll horizontally to see all columns.

CSV-represented book-record states
Metadata stateRecordsPercent of book records
One recognized CC family only85,10881.86
No nonblank statement7,5527.26
Access statement only3,2113.09
Reserved phrase; no recognized CC1370.13
Other / unresolved by strict parser7,9487.65
Explicit public-domain designation70.01
Multiple recognized CC families00.00
One CC family plus other assertion00.00

Mutually exclusive states; 103,963 eligible book records. Zero states retained. Unresolved does not mean unlicensed or closed. Rounding can prevent percentages summing to exactly 100.

Scroll horizontally to see all columns.

Recognized CC-family flags
FamilyBook recordsPercent of book records
CC BY-NC-ND36,63635.24
CC BY34,43233.12
CC BY-NC5,0954.90
CC BY-NC-SA4,1323.97
CC BY-SA3,8783.73
CC BY-ND9350.90

Common denominator: 103,963 book records. Flags are non-additive by design; no multiple-family record detected in this CSV. Not a measure of legal permission for every use.

Scroll horizontally to see all columns.

Unresolved metadata and recognition diagnostics
DiagnosticRecords
External terms or other URL7,270
CC-looking nonaccepted string663
Other text or relative link15
Additional recognition after lowercasing (sensitivity)306
Versioned CC URI present85,079
Family-only CC URI present27
Accepted text-only CC designation2
Multiple same-family statements32
Multiple recognized versions in CSV0
Recognized NC flag45,863
Recognized ND flag37,571
Recognized SA flag8,010

Counts refer to eligible book records. Diagnostic rows overlap and must not be summed. Public-domain designations: three CC0 and four PDM.

Evidence & method

How we calculated it

  1. Single official metadata-only CSV response, SHA256-pinned; exact book type, distinct UUID and handle, no work deduplication.
  2. Classify content-related BITSTREAM_License only; no inheritance of metadata CC0. Preserve missing/access-only/reserved/custom-link distinctions.
  3. Conservative structured URI and anchored-label parser; semicolon assertions; mutually exclusive states and non-additive family/condition flags.
  4. Twenty-two synthetic tests, sixty seeded record/value reconciliations and complete 211-value frequency reconstruction; checks concern computation, not legal truth.
  5. Report case-folding sensitivity separately; one xoai representation check is illustrative and not a population sample.

What the data cannot tell us

  • Snapshot limited to book records represented in one export; no verified provider generation timestamp or complete-work census.
  • External terms links and unsupported strings remain unresolved; this is not an estimate of unlicensed, closed or unlawfully distributed books.
  • CSV may not capture every assertion in other interfaces; one checked difference does not establish its cause or prevalence. Multiple versions may be legitimate alternatives.
  • No full texts, third-party component rights, link availability or legal ownership reviewed. No publisher-type comparison, causal claim, trend or invented sampling interval.

Original sources

  1. DOAB official CSV metadata export

    DOAB official CSV metadata export: Single response retrieved 2026-10-30T08:37:57.351428Z; original SHA256 cddd2660be7cbb14cc88eeb854e31e1346767508d69dd74c1d46e5aba9e4c5e6; 103,963 eligible book records. Source generation timestamp not independently confirmed.

    Accessed October 30, 2026

  2. DOAB metadata documentation

    DOAB metadata documentation: Documents official bulk formats, OAI harvesting and CC0 metadata terms; selective exports optimized for up to 500 titles.

    Accessed October 30, 2026

  3. DOAB frequently asked questions

    DOAB frequently asked questions: Distinguishes metadata rights from book rights, discovery service from content repository, and books from chapters.

    Accessed October 30, 2026

  4. Creative Commons licence overview

    Creative Commons licence overview: Official descriptions of six licence families and their conditions; not a legal determination for any book.

    Accessed October 30, 2026

  5. Creative Commons BY-NC-ND 4.0 deed

    Creative Commons BY-NC-ND 4.0 deed: Conditions and notices, including distribution of adapted material and exceptions; deed is not the legal code.

    Accessed October 30, 2026

  6. Creative Commons CC0 1.0 deed

    Creative Commons CC0 1.0 deed: Public-domain dedication; catalogue metadata CC0 is never assigned to book content.

    Accessed October 30, 2026

  7. Creative Commons Public Domain Mark 1.0 deed

    Creative Commons Public Domain Mark 1.0 deed: A public-domain designation, distinct from a licence; limitations concerning jurisdictions and other rights remain.

    Accessed October 30, 2026

  8. DOAB checked xoai record

    DOAB checked xoai record: Response 2026-10-30T08:33:44Z; record datestamp 2025-07-30T08:59:03Z. One checked representation differs from CSV; no export mechanism or prevalence inferred.

    Accessed October 30, 2026

Cite & reuse this study

Alisson Bento. What a book catalogue can tell us about reuse licences. Analysis of DOAB CSV book-record licence metadata retrieved October 30, 2026. https://alissonbento.com/research/doab-book-licence-metadata-2026/

DOAB source metadata are available under CC0 1.0. Book-content licences vary and are not replaced by that metadata dedication. Calculations and explanation: Alisson Bento. No licence for the complete article is implied; no publisher endorsement or legal advice.

Corrections or editorial questions

Pass it on.

What a book catalogue can tell us about reuse licences | Alisson Bentoalissonbento.com

WhatsAppFacebookXTelegramRedditThreadsEmailSMS