Sources
Full description of each source
- Library Genesis catalogs, 25 snapshots, 2015–2025. The non-fiction ("updated" table) and fiction database dumps that Library Genesis published, as preserved by the Internet Archive on the dates below. Each catalog entry describes one file; entries describing the same edition are combined into one line, and each line carries the list of snapshots in which it appears, with the count of entries in each. The April 2019 snapshot contains the non-fiction catalog only. The February 2020 snapshot combines the non-fiction dump of February 13, 2020 with the fiction dump of February 8, 2020, which survives only as a Wayback Machine capture of
libgen.rs/dbdumps/; both are dated February 13, 2020 in reports. Only catalog metadata is used; no book files are held. - Pirate Library Mirror (PiLiMi) index of Z-Library, 2022. The metadata table of the PiLiMi mirror of Z-Library (11,783,153 records), in the metadata-only copy re-hosted on Hugging Face as
P1ayer-1/annas-archive-index. Each record names the PiLiMi release that carried the file: 2,179,884 in the first release (July 2022, the release the Bartz v. Anthropic order describes Anthropic downloading that month), 3,818,910 in the second (September 2022), and 5,784,359 not mirrored (most already in LibGen). This index carries no ISBNs, so PiLiMi records match registrations by title and name only. No files from the mirror are held. - Z-Library records added after PiLiMi, 2022–2025. Two later catalogs of Z-Library. (a) Anna’s Archive’s index of Z-Library records (“zlib3”), in the metadata-only copy on Hugging Face (
P1ayer-1/annas-zlib3-index): 2,630,955 records with Z-Library IDs 22,430,000 to 25,639,999, added to Z-Library from August 25, 2022 to August 6, 2023, each with the date added, title, author, publisher, ISBNs and the MD5 Z-Library reported for the file. (b) The Z-Library extract in the MajinBook catalogue (Mazières and Poibeau, Zenodo, CC-BY 4.0;zlibrary.jsonl.gz, 8,097,488 records), which its authors describe as Z-Library’s metadata as of January 2025, obtained through Anna’s Archive, limited to EPUB-type records (EPUB, MOBI, AZW and similar); it carries IDs, titles, authors, years, languages, publishers and ISBNs, but no dates or MD5s. Records already in PiLiMi are not repeated (a PiLiMi line notes whether the January 2025 extract still lists it); 2,352,562 records are known only from the extract, and their IDs place them after August 2023. Anna’s Archive is a search engine and mirror that combines other collections; the court record on Meta’s early-2024 download of it is summarized below. No Z-Library or Anna’s Archive files are held. - Internet Archive lending library. The catalog of the Internet Archive’s lending library (its
inlibrarycollection of scanned print books that signed-in users can borrow), read through the Archive’s public scrape API on the date given under Provenance: identifier, title, creator, ISBNs, publication year, publisher, language and the date the Archive added the item. The Internet Archive is a nonprofit library. In Hachette Book Group, Inc. v. Internet Archive, 115 F.4th 163 (2d Cir. 2024), the court held that the Archive’s scanning and lending of the 127 works at issue was not fair use. That AI developers obtained Internet Archive books is alleged in pending cases against Meta and NVIDIA; no court has found it. - Journal articles (DOI lists). For the free article check: the list of 62,835,101 DOIs Sci-Hub posted on March 19, 2017 (Internet Archive item
scihub_doi_list_2017,doi.7z); the DOI column of LibGen’s scimag database as published in February 2020 (libgen-meta-20200213,scimag_id_doi1_doi2_isbn_md5_pmid_pmcid.tsv.gz); and the scimag database dump of August 6, 2021 (libgen-meta-20210806,libgen_scimag_dbbackup-2021-08-06.rar), which also gives the date each article was added. Each DOI is lowercased, stripped of anydoi:ordoi.org/prefix, and stored only as the first 8 bytes of its SHA-256 hash with a source mask; the article files were never fetched. Article lists for a person come from Crossref’s open metadata API (by ORCID iD where the publisher deposited it, or by author name). The free check shows list membership only. The paid Journal Article Registration Report adds two kinds of U.S. registration that may cover an article. Publisher side: the journal issue registered as a serial, from the Office’s serial registrations (the tabular filereg_serial_2026_01.csv, registrations 2006–2025, and the serial records of its MARC XML bulk file, registrations from 1978; 3,504,756 issue registrations across 174,707 serials). An article is matched to a registration through the journal’s ISSN (or its exact title when Crossref gives no ISSN) and the issue’s volume and number, or its volume or month where the registration gives only those; a serial registration covers the articles first published in that issue that the claimant fully owned when it filed (Circular 62B). Author side: a TX registration whose main title equals the article’s, with one of the article’s author surnames among the registration’s authors or claimants and a publication year within a year before or three years after the article’s; registrations carrying an ISBN (books) are left out. The § 412 timing condition is then compared with the dated article-copying events listed on the report, each graded. Many authors assigned their copyright to the publisher, so the publisher may be the owner who can sue. - Books3 file-name list. A list of 196,608 Books3 file names published by Peter Schoppert (psmedia/Books3Info,
filenames.txt). Books3 as released is reported to contain 196,640 books; the published list is 32 names shorter. No Books3 texts are held. - Books3 ISBN list. 129,635 unique ISBN-13s that Schoppert extracted from the Books3 texts (
Books2_ISBNs.txt, same repository). His README describes a pattern search over the texts that collected ISBN-13s only. - U.S. Copyright Office registrations, 1978–2025. The Office's public tabular data for non-dramatic literary works (TX) as published in January 2026 (
reg_non_dramatic_literary_work_2026_01.csv, 9.28 million rows, of which 8,034,119 carry a registration number: 6,047,480 published works (TX) and 1,986,639 unpublished works (TXu), registered from January 1, 1978 through December 30, 2025; the other rows carry no TX or TXu number and are not used). These are public U.S. government records; PayAuthors is not affiliated with, or endorsed by, the U.S. Copyright Office.
LibGen snapshot dates and Internet Archive items (table)
LibGen snapshot dates and Internet Archive items
| November 14, 2015 | libgen-meta-20151114 |
| December 26, 2015 | libgen-meta-20151226 |
| January 9, 2016 | libgen-meta-20160109 |
| June 17, 2016 | libgen-meta-2016-06-17 |
| November 4, 2016 | libgen-meta-20161104 |
| December 30, 2016 | libgen-meta-20161230 |
| May 12, 2017 | libgen-meta-20170512 |
| April 15, 2019 | libgen-meta-20190415 |
| February 13, 2020 | libgen-meta-20200213 (non-fiction); fiction dump of February 8, 2020: Wayback Machine capture of February 17, 2021 (web.archive.org/web/20210217010544/http://libgen.rs/dbdumps/fiction_2020-02-08.rar) |
| February 12, 2021 | libgen-meta-2021-02-12 |
| June 5, 2021 | libgen.rs-dbdumps-libgen_2021-06-05.rar-2021-06-05 and libgen.rs-dbdumps-fiction_2021-06-05.rar-2021-06-05 |
| March 7, 2022 | libgen-meta-20220307 |
| March 8, 2022 | libgen-meta-20220308 |
| June 19, 2022 | libgen-meta-20220619 |
| January 13, 2023 | libgen-meta-20230113 |
| April 29, 2023 | libgen-meta-20230429 |
| May 13, 2023 | libgen-meta-20230513 |
| January 6, 2024 | libgen-meta-20240106 |
| June 28, 2024 | libgen-meta-20240628 |
| October 2, 2024 | libgen-meta-20241002 |
| November 16, 2024 | libgen-meta-20241116 |
| December 17, 2024 | libgen-meta-20241217 |
| January 18, 2025 | libgen-meta-20250118 |
| March 8, 2025 | libgen-meta-20250308 |
| June 3, 2025 | libgen-20250603 |
Each report lists the SHA-256 checksum of every source file it was built from, and the dataset build it came from. The full provenance table (archive item, file name, size, the archive's own SHA-1, our SHA-256, and the checksum of each parsed catalog) is published with each dataset build at the end of this page.
A checksum identifies the exact bytes of a file. By itself it does not establish a file's origin, completeness or authenticity.
Accuracy
Matching is automated and not perfect. Expect some missed records (other spellings, pen names the Office records do not tie to a legal name, entries with no author, registrations under a different form of the name) and some matches that belong to other people with similar names or to other editions. Every line should be checked against the author’s own records and, for registrations, the linked Copyright Office record.
The headline numbers
22.4M catalog records (22,370,394) is the sum of each source’s records, which overlap, so it counts records, not unique books: LibGen non-fiction 4,275,041 and fiction 2,515,792 (one line per edition, combining that edition’s catalog entries across the 25 snapshots); the PiLiMi index of Z-Library 5,998,794; later Z-Library records 4,979,534; the Internet Archive’s lending library 4,423,866; and Books3 file names with a recognizable author 177,367. 1.52M (1,522,018) is the number of distinct U.S. registrations that at least one catalog record is matched to as a candidate; 3,185,792 catalog records carry at least one candidate match. Each is a candidate to confirm against the Office record. The article check covers 85,435,752 distinct journal-article DOIs across the Sci-Hub list and the two scimag dumps. These figures update with each dataset build (now payauthors-2026-10-01-m64).
How a name is matched
The searched name is split into parts. A record matches when every part is found at the start of a word within one author name of the record, so “Ann” matches Ann and Anna but not Joanne, and initials are optional. In the default setting, “name parts close together,” the parts must fall within a short window of the same name, which excludes lists of other people’s names in the same field. “Name parts anywhere in the author field” drops that window and finds the parts anywhere in the author field, so it returns more records, including other people’s. Name order does not matter. Other spellings and pen names are separate searches.
LibGen catalog entries describing the same edition (same author field, title, ISBN, year, publisher and edition statement) are combined into one line, with a count of the entries; none are dropped. Books3 author and title are read out of the file name (“Title - Author.epub.txt”); where a name could not be split with confidence, both halves are shown. Some records cannot be found by name at all: about 138,000 non-fiction and 40,000 fiction LibGen entries carry no author, and about 19,000 Books3 file names carry no recognizable one.
Books found by ISBN when the record gives no author. A paid report also looks up the works Open Library lists under the name searched and checks their editions’ ISBNs against our catalogs, for books the name search did not find. A catalog copy found this way, whose record gives no author or another name and whose title agrees with the work, is added to the report marked as found by its ISBN and unconfirmed: it is left out of the counts until the report holder ticks it as their book. Open Library’s author records can mix people who share a name, which is why these lines start unticked. Each report reads up to six of our ISBN index files this way, for the books it is missing first.
Books found through a registration’s ISBN, listed under another name. A paid report also looks up, in our ISBN index, the ISBNs that tie its registrations to a book: the ISBNs printed on the Copyright Office record, the ISBN the Bartz settlement Works List gives for the registration, and the ISBN of a catalog line already matched to the registration. A catalog copy that carries one of these ISBNs and the same main title, but is listed under a misspelled or different name or none (an anthology under its editor’s name, spelled “Etchinson” for “Etchison”), is added to the work, marked “found by the ISBN on registration …”, and counted with it; the registration’s own checks still apply. Since revision 28 (Oct. 15, 2026), lines found by ISBN also carry the catalog dates of the copy. A catalog author field that is a surname with a role, “Monteleone (editor)”, is treated as the surname alone: shown, unconfirmed, and not counted unless included.
File IDs and MD5 checksums
Each LibGen catalog entry describes one file and carries a numeric ID and the catalog-reported MD5 checksum of that file: the library’s own record, which serves as the file’s identifier. We hold no files, so we have not recomputed any MD5. A report line that combines several entries lists each entry’s ID, file type and MD5, with the snapshots that list the entry where they differ from the line’s. LibGen numbers non-fiction and fiction separately. PiLiMi records show the Z-Library ID and MD5 from the PiLiMi index. The Books3 file list has names only, so Books3 lines have neither. All of these come from catalog metadata; we have not seen or checked the files themselves.
More on file IDs, torrents, pieces, EPUB/MOBI and added dates
An MD5 identifies exact file contents, so a catalog-reported MD5 can be compared with a defendant’s file lists, hashes or logs produced in discovery, and a produced copy can be hashed to see whether it is the file the catalog describes. A match there would tie a work to a specific file. A catalog listing alone does not show that anyone obtained that file.
Torrents. LibGen distributes its files in torrents of 1,000 consecutive IDs, each named for the first ID in its block: non-fiction r_000, r_1000, r_2000 and so on, fiction f_0, f_1000 and so on. We checked this against the Internet Archive’s crawls of both torrent directories (November 20, 2020: 2,817 non-fiction and 2,223 fiction torrent names, every one a multiple of 1,000) and against a court filing that places two fiction files, identified by MD5, in fiction torrent “1000.” Our catalog lists both, as fiction IDs 1,286 and 1,291 (class plaintiffs’ Rule 56.1 statement ¶¶249, 251, In re OpenAI Copyright Litigation, S.D.N.Y. 1:25-md-03143, Dkt. 1987; an allegation). A LibGen file line names a torrent only when that torrent’s own file list holds the file. Where we hold the torrent for a file’s ID block and its file list does not include the file, the line says so and names no torrent; where we hold no copy of the block’s torrent, the line gives the block’s torrent name and says it was not checked. Each torrent named carries the creation date written in the torrent file or, when the file carries none, the date of the earliest public copy we have of it. A torrent made after a download ended is marked as made after that download, and it is never offered as the torrent that download used. PiLiMi lines show the PiLiMi torrent named in the PiLiMi index.
Torrent pieces. A torrent file lists every file it carries, with its size, and divides the combined data into pieces of a fixed length. Each piece is checked as a whole, but it is sent in small blocks, which a downloader can collect from several peers. From the torrent files’ own metadata (file names, sizes and piece length; no book text), each file line shows the file’s size, the piece length, how many pieces the file spans, and the largest share of the file that lies inside one piece. When a file lies inside one piece, every verified copy of that piece holds the whole file; whether one uploader sent all of it has to come from that uploader’s records. LibGen torrents come from the Internet Archive’s November 20, 2020 crawls and from 1,920 later block torrents captured one by one by the Wayback Machine at libgen.rs and libgen.is; PiLiMi torrents come from the PiLiMi torrent archive (pilimi-zlib-meta). Each LibGen file is located in its torrent by MD5 and each PiLiMi file by Z-Library ID; a file its block’s torrent does not list is shown without torrent data. Piece data covers 99.6% of the non-fiction and 98.9% of the fiction files in the June 5, 2021 catalogs, and 95.2% and 84.0% of those in the June 3, 2025 catalogs, because some later torrents were never captured. For Works List copies in LibGen blocks those captures missed, the evidence uses the copy of LibGen’s own torrent file kept by Anna’s Archive: wherever both exist, Anna’s Archive’s copy is identical to the one the Internet Archive or the Wayback Machine kept (6,959 of 6,960 torrents have the same info hash; the one that differs, f_2996000, is an earlier version of that block’s torrent made before the block was full, and such partial torrents are not used). No source has a torrent for the non-fiction blocks holding IDs 81,000 to 82,999, so files there are shown by catalog ID and MD5 only. When a Works List entry has several copies, the report shows a copy in one of Anthropic’s download windows first, preferring one whose torrent existed by the end of that download, and among those one tied by an ISBN (the Works List ISBN, or an ISBN the Copyright Office record lists) before one tied only by title and author, and it says how the copy is tied; a copy found only outside Anthropic’s download windows is labeled that way and is never offered as the file Anthropic took. A file counts as listed in a catalog when that catalog lists the same file (MD5) under any version of its record. A copy tied to a Works List entry only by title, through a catalog record that names no author, is not used when another catalog record of the same file names a different author; nor is a copy tied only through an ISBN the Works List gives to several different titles, when the file’s own catalog record is registered under another of them. The same holds when another record of that file (LibGen.li, or book-trade and library records for the ISBN printed in it) names a different author. Where such a record names the Works List author, the copy is shown as tied by title and author; where no record names any author, it is shown as tied by title only (the catalog record names no author), and never ahead of a stronger copy. Where a Works List entry has no copy of the registered edition in Anthropic’s download windows, we also look for another edition of the same title and author. The format, edition, publisher and date shown next to an ISBN come from Open Library’s public editions data (Sept. 30, 2026); they describe that ISBN, not the copy in the library, and are shown only where Open Library has them. Candidate editions are found as leads. Each copy shown is a catalog record naming the same title and author, checked by hand; a further check against public book listings is under way, and a copy that fails it is removed. Such a copy is labeled as another edition, tied by an ISBN of that edition rather than the registered one, and is listed after any copy of the registered edition. Every PiLiMi release 1 file has piece data; release 2 was distributed as whole archive files, so its pieces do not follow book files and none is shown. Of the files located, 79% of fiction files, 26% of non-fiction files and 74% of PiLiMi release 1 files lie inside a single piece, and 97%, 61% and 93% have at least half of their data inside one piece. No torrent was joined.
EPUB and MOBI. File lines mark EPUB and MOBI files, because OpenAI’s 2018 download script is alleged to have taken only those two types (same filing, ¶147). The mark is information only; it does not change the statutory-damages timing mark.
Added dates. Each LibGen entry carries the date LibGen says the file was added, shown as “added to the catalog … (LibGen’s own record).” Between May 12, 2017 and February 12, 2021 the only fiction catalog that survives is the February 8, 2020 dump (a Wayback Machine capture of libgen.rs/dbdumps/fiction_2020-02-08.rar, 801,097,717 bytes, whose SHA-1 matches the digest the Wayback Machine recorded for the capture). Directory listings captured in 2019 show weekly fiction backups, but none of those files was captured. For fiction entries between those dates, the added date and the February 2020 dump are the only records of when an entry appeared. Entries deleted before a catalog we hold do not appear at all. As a check on the added dates: LibGen assigns IDs in order, and in our February 12, 2021 fiction catalog the added dates pass October 31, 2019 at about ID 2,190,000, below which the catalog holds 2,189,999 entries; the class plaintiffs’ filing puts LibGen fiction at 2,182,642 files in October 2019 (¶¶137, 236). In the February 13, 2020 non-fiction catalog the added dates pass January 31, 2020 at about ID 2,470,000, with 2,469,999 entries below it; the filing puts non-fiction at about 2.46 million as of January 2020 (¶137). Both agree within about half a percent (the ID cut-off is estimated to the nearest 10,000). The February 8, 2020 fiction dump allows a direct count: 2,185,033 of its 2,206,073 entries carry an added date on or before October 31, 2019, 0.1% above the filing’s 2,182,642, and 2,205,667 carry one on or before January 31, 2020.
How registrations are matched
Five tiers, strongest first, summarized under At a glance. Tiers 2 to 5 are heuristic: a match means a registration exists for that title and name, not that the catalog copy is that exact edition. Every match is a candidate to confirm against the linked Office record.
The full rule, the § 412 timing mark and its limits
Records are matched to Copyright Office TX registrations as follows, strongest first: (1) the same ISBN on both sides, kept only when the registration also shares an author or claimant surname, pen name or publisher name, or most of the title, because one ISBN is often printed on several editions, formats or an anthology; (2) the same full title plus an author or claimant surname; (3) the same main title (before a colon, slash or dash), or the book’s own title after a series name or number, plus surname, with safeguards against series names and disagreeing subtitles; (4) the same title words with stop words ignored, plus surname, only when nothing else matched; (5) the same full title plus the catalog’s publisher named on a company registration (a work made for hire); (6) since dataset v6.4, a record with no registration whose author is a pen name that Office records tie to a legal name (“pseud.” or “writing as” registrations; house names shared by several writers excluded) is matched again with the legal name added, and such matches are labeled and held for a person to confirm. Titles are compared after routine normalization (case, punctuation, leading articles and edition notes), and common first names and role words are not treated as surnames. A match is set aside when the registration is for a translation, a study guide or an adaptation of the work (a play or graphic novel, say), or when volume, issue or edition numbers disagree. A record lists up to eight registrations of the work, chosen by how close their dates are to the record’s. Measured accuracy is below. Tiers 2 to 5 are heuristic: a match means a registration exists for that title and name, not that the catalog copy is that exact edition. Every match is a candidate to confirm against the linked Office record. TX registrations from 1978 through 2025 are covered, as published by the Office in January 2026; sound recordings such as audiobooks and renewals are not. Statutory-damages timing mark. For each matched registration the report compares the registration's effective date with each documented download date on the timeline (Anthropic: Books3 early 2021, LibGen June 2021, PiLiMi July 2022; Meta: October 2022 and after) and with the publication date in the registration. The screen sorts each comparison into one of four cases. (1) Known start: a court record, company statement or complaint gives when the download period began. (2) Catalog lower bound: the library’s own record of when that copy was added (LibGen’s or Z-Library’s record date; for Books3, its October 2020 release). This is catalog metadata as the library reported it, not independent proof of when anyone acquired the file. For the download date we use the later of (1) and (2); for an event known only by an outer bound (a paper, model release or complaint date), only (2) is available. A registration dated before that date passes the screen. (3) Publication route: a registration made within three months of a publication that came before that earliest date also passes (§ 412(2)). (4) Insufficient dates: a registration dated after the start of the download but before its end, or a comparison with no earliest date at all, is shown as “dates overlap or download date unknown” and never counted. The screen is deliberately conservative: it may leave out works for which § 412 would still allow statutory damages, for example a registration made within three months of publication where no earliest copying date is known. The condition comes from 17 U.S.C. § 412, which has exceptions this screen does not assess. The journal-article report uses the same four cases, with LibGen scimag’s record date as the catalog lower bound. For counts, copies, formats and editions of one book count as one work, timed from its earliest copy in the dataset (a translation included). A copy in another language with a foreign title or ISBN is a translation: it is listed with the registration of the original but never counted as the book’s presence in a dataset, because a translation is not the registered text, and a book found in a dataset only in translation is named apart. A copy the library tags in another language that carries the English title and an English ISBN (or none) is counted as the book with a wrong language tag and, for each company, from that company’s earliest known download of the work (see the rules we apply). It is a date screen, not a finding that the registration covers the listed edition, that the work was in the download, or that damages are available. Nor does it address whether a claim is timely: copyright claims must generally be brought within three years after they accrue (17 U.S.C. § 507(b)). In suits by authors who opted out of the Bartz v. Anthropic settlement, Anthropic argues (motion to dismiss, September 24, 2026) that claims based on its 2021–2022 downloads are time-barred and that the class action did not suspend the period for all of them; the court has not ruled. Limitations, the discovery rule, class-action tolling and the Bartz release should be checked with counsel for each work. Plaintiffs have begun to define classes by the same timing: the textbook-author classes proposed in Sullivan v. Meta (N.D. Cal., filed July 2, 2026) and Sullivan v. OpenAI Foundation (S.D.N.Y., filed August 14, 2026) cover only works registered within five years of first publication and before the defendant's copying, or within three months of first publication. Neither class has been certified.
Downloads of part of a library. Some records show a company took only part of a library: DeepSeek’s paper describes 860,000 English and 180,000 Chinese e-books from Anna’s Archive; Anthropic took the PiLiMi copies of books it did not already have from LibGen, at least two million (Bartz v. Anthropic, Dkt. 244 at 2:25–3:5); Meta’s verified interrogatory answers describe a partial torrent download of LibGen’s non-fiction collection in October 2022 and a direct download in spring 2023; and which files Meta torrented from Anna’s Archive is disputed (the court ordered Meta to produce its torrent servers’ command histories, Kadrey v. Meta, Dkt. 831, Sept. 11, 2026). A line that passes the date screen against one of these downloads says so: the library was taken in part, and whether this copy was among the files taken is not shown. Since October 13, 2026 (analysis revision 26).
Record-match marks
Every registration match in a report carries one of these marks.
- Record matched: the match passed every check against the Copyright Office record. The catalog title equals the registered title or its main title; a surname of the catalog author appears among the registration’s authors, pen names, claimants or statement of responsibility (or the match was made on the ISBN); no volume, part, issue or edition number conflicts; neither side is a translation, guide, summary or adaptation; and the catalog edition is dated from the registration year to two years after it. A match also counts when the catalog ISBN is printed on the Office record itself and the title agrees, whatever the year. Since October 4, 2026, a match found only by main title, by title words or by publisher is never marked on these checks alone: it waits for the review below (reports first issued earlier keep the earlier rule until a person reviews them). The same holds for a short catalog title when the registration adds a subtitle, which often marks another book in a series. Since October 6, 2026, the same holds for a registration that covers only something added to an existing text: an abridgment or a performance, an introduction, illustrations, notes or editing claimed as the authorship, or work made for hire by someone other than the catalog author. Full-record check. A match found by main title that waits for review is first checked again, one by one, against its full Copyright Office record rather than the summary index: the complete title and subtitle, the authors and claimants, the edition statement, any earlier registration, the material the registration covers and excludes, and the dates. It is confirmed only if nothing on the full record conflicts. If the registration covers only new matter, another language, a work about the book, a different volume or a later edition, or if the full record is not available, the match stays Not yet checked; a catalog copy dated more than two years after first publication is marked Record matched: other edition. A random sample of matches confirmed this way is checked again, one by one, in a separate software review, and a person then checks a sample of those. The samples and their results are kept with the dataset build. A catalog author that names an organization (a college, bureau or board) does not match on a shared word.
- Record matched: other edition: the registration is for the same work in another edition, version or printing.
- Registration for a later edition: not counted: the registration is for an edition published two or more years after the catalog copy. A registration for a revised edition covers that edition’s new material, which the earlier copy does not have, so it is listed for you to check but not counted. A catalog ISBN printed on the Office record, or a person’s review, overrides this.
- Registration covers the collection: not counted: for a piece the author contributed to someone else’s book, such as an essay in an anthology or a chapter in an edited volume, the matched registration is usually the collection’s, filed by its editor or publisher. When the Copyright Office record names the author neither as an author nor as a claimant, nothing on it shows that it covers the author’s own piece, so the registration stays in the report as evidence but is left out of the counts and the registration-date screen. A registration of the piece itself, or a record that names the author, still counts. The same applies to any book added from the author’s own book list whose registration names someone else as author and claimant, such as a publisher registering the book as a work made for hire: it is labeled “Registration found; it does not name the searched author”. Since October 11, 2026.
- Not yet checked: found by our matching but not yet confirmed. Matches that do not pass every check are reviewed one by one against the same record in a software review, and some batches of matches that did pass are reviewed as well. Matches the automatic rules can’t settle get a second, software review; a person checks samples, and published accuracy figures come only from a person’s checks. Each reviewed batch is either reviewed twice in full or has a random sample checked again in a second, separate review, and the mark changes when the review is done; a match the review cannot confirm stays marked Not yet checked, even if it passed the automatic checks. A report’s counts and date screen use confirmed matches only. Preliminary and reviewed reports. A paid report is issued at once as a preliminary report, on the automatic checks and the software review. A person at PayAuthors then checks every registration match on a line that passes, or would pass, the date screen against the full Copyright Office record, including the claimant’s address and the rights-and-permissions contact where the Office lists them, and the report is marked Reviewed at the same link. A person’s decision overrides the software’s. The review is completed within 7 days after purchase, or within 7 days after a new version with changed figures if we issue one before the review. Every decision, whether automatic, by the software or by a person, is stored with its basis, so any one can be inspected. When a buyer opens a report, its matches not yet checked are reviewed the same way; one the review cannot settle is shown, marked Not yet checked, and not counted. A match found to be another work is removed from the report. That includes a registration for a different volume of the same series, and a registration that covers only what one edition added to an older text (another person’s introduction, notes, editing, illustrations, translation or adaptation, or a new edition of a public-domain classic) when the catalog copy is not that edition.
In a review of 800 matches against the Copyright Office record, every match that passed the checks was the registered work: 272 the same edition and 5 another edition of it. The 800 were 500 matches drawn 100 from each matching method plus 300 drawn from random parts of the dataset; each was compared with its Office record in a software review that was not blinded to the matching method. The checks are strict, so many correct matches start as Not yet checked until reviewed. A record match says the registration is for this book; it is not a finding of ownership, copying or eligibility for damages. The report summary gives the count for each report, and the CSV has a registration_check column.
Possible other edition (analysis revision 32)
A catalog copy tied to a registration by title and author only (not matched on an ISBN, and with no ISBN of its own on the Copyright Office record or given for the registration by the Works List) is held as a possible other edition when it names another edition than the record, or is dated two or more years after the registered edition was published. An edition statement alone does not hold a copy dated the year the registered edition was published (or the next) when the record names no edition, and a LibGen fiction series number is not read as an edition. The copy is listed with its reason, not counted as the registered edition, and not given a match carried from another copy of the same title. Any person’s review decision still governs. Later printings that the Office record shows to be the original text are still shown as another printing of the registered work.
Possible additional claims (analysis revision 32)
Reports first issued under revision 32 list, in their own section, the author’s registered works that are not on the Bartz settlement Works List as we hold it (the settlement site lookup, 476,444 of its 482,460 entries), or whose entry our review found names a different book, where a catalog copy in one of Anthropic’s download windows is a later printing, a later or revised edition, a possible other edition or a translation of the registered work. A work never on the Works List was never in the class, so the settlement did not release claims for it (Bartz Dkt. 680 at 9–10). Each needs more proof: the claim rests on showing that the copy reproduces the registered text, and which copy Anthropic took is a question for discovery. Outside the class, the five-year point bears only on the presumption of validity (17 U.S.C. § 410(c)), not on whether a claim exists. These works are listed for counsel to assess. The section adds nothing to the report’s counts and confirms no claim; a work listed there may also appear in the report as a match, which shows a copy, not a claim. From revision 33, such a line says when the report also counts the registration, on the same copy or another.
Works List status of each work (analysis revision 33)
Reports first issued under revision 33 say, for each registration they count, whether it is on the Bartz settlement Works List as we hold it (the settlement site lookup, 482,416 of its 482,460 entries, read through Oct. 31, 2026) and, if it is, whether it is on the opt-out list. A registration we do not find is marked “not found”; the 44 entries we have not seen could include it. From analysis revision 36, such a registration also says whether we can link another edition of the same work to the list: “not listed (this registration or any linked edition)” when no edition we can link by ISBN or by title and author is listed, or “another edition appears to be listed”, naming that entry and how the two are tied, for counsel to confirm. A registered work not on the Works List was never in the class, so the settlement did not release claims for it (Bartz Dkt. 680 at 9–10); whether any claim exists is for counsel to assess. A report shows only the searched author’s own works, which each author can also look up on the settlement site; we do not publish the list.
The listed copy, its torrent pieces and what shared them (analysis revision 33)
For a counted registration on the Works List, and for each work on the Bartz opt-out list, a report first issued under revision 33 gives the listed copy we tie to the entry: the library’s ID and MD5, the catalog it is in (the June 5, 2021 LibGen catalog or PiLiMi release 1, the two Anthropic downloads, or another), the torrent and its BitTorrent info hash, the pieces that carry the file, and the SHA-1 hash of each of those pieces, read from the torrent file’s own metadata. A BitTorrent client checks every piece it receives against that SHA-1, so a piece hash names the exact bytes a download received. The report also says how many other files share those pieces, how many are not on the Works List, and how many lie wholly inside a shared piece. PiLiMi release 2 is distributed as whole .tar files, so positions there are estimates and carry no piece hash. We tie Works List entries to files from catalog records (ISBN, title and author); the settlement did not publish which file it relied on, and Anthropic’s own copy and its hash come only from discovery. In the possible-additional-claims section, a copy whose file is itself a Works List copy listed under another registration, or shares a torrent piece with a Works List copy, says so. When the opt-out list lists the same title under another registration whose Copyright Office record names the same people, the report notes it without deciding whether the two registrations cover the same work.
Which registration, which download: the rules we apply
Not legal advice. This section explains how our reports apply the registration rules of the Copyright Act. It is general information for research, not legal advice. Whether any work qualifies, and what a claim is worth, is for a lawyer and a court.
A registration covers the work, not a printing. A registration reaches the text it registers in any printing or format, so a later paperback or ebook of the registered book with the same text is the same work. An edition matters only where its text differs. An earlier registration does not cover material a later edition added (17 U.S.C. § 103(b)), though it still covers the registered text the later edition carries (Montgomery v. Noga, 168 F.3d 1282 (11th Cir. 1999)). A registration that covers only a later edition’s new material, another person’s introduction or notes, or a translation does not reach a copy that lacks that material. Reports show edition details as a sign of how closely a copy matches the registration, not as a condition.
Earliest known download per company. Statutory damages and attorney’s fees depend on registration before the infringement commenced, or within three months of first publication (§ 412). Courts treat a series of infringements of the same kind as one continuing infringement that commenced with the first act (Derek Andrew, Inc. v. Poof Apparel Corp., 528 F.3d 696 (9th Cir. 2008)). So reports time each company from its earliest known download of a work, across every copy, edition and format of it in the report: a later download counts only if the registration also came before the company’s first one. Counts against documented downloads look at the company’s documented earlier downloads; counts against alleged downloads look at all of them. A registration that falls between two downloads by the same company is shown, marked “registered after the same company’s earlier download,” and not counted. This applies to all reports from analysis revision 10 (Oct. 4, 2026); it can lower a report’s counts, never raise them.
First publication of the work. The three-month route in § 412 runs from the first publication of the work, anywhere. The publication date on an Office record is the one the applicant gave for the registered edition; where an earlier edition exists, such as a first edition abroad, the work’s first publication may be earlier. Most counted matches pass on the first route, registration before the download, where the publication date does not change the result. Registration Details reports show both dates for each registration so this can be checked.
The five-year mark. A registration made before or within five years after first publication is prima facie evidence that the copyright is valid (§ 410(c)); the weight of a later one is up to the court. Class definitions in these cases use the same window: the Bartz v. Anthropic Works List required registration within five years of publication, and the textbook-author classes proposed against Meta and OpenAI do too. Our date screen does not apply the five-year mark; reports show the dates so it can be checked.
Works first published abroad. A United States work generally cannot be sued on until it is registered or refused (§ 411(a)). Works first published outside the United States are exempt from that requirement but not from § 412, so statutory damages still depend on timely registration. Unregistered foreign works are not in our counts.
Releases and opt-outs. Works on the Bartz v. Anthropic Works List were released as against Anthropic. The 350 owners who opted out (1,802 works; Dkt. 619-3, Ex. J) kept their claims. Our “not on the Works List” lists are our screen against the settlement’s criteria; check each work with the settlement’s official lookup.
What else a claim needs. A match and a passing date screen are two pieces. A claim also needs evidence that the work was in the defendant’s download (its file lists, MD5s or logs), ownership of the rights today (transfers, work made for hire), a valid registration covering the copied text, a timely suit (§ 507(b); limitations for the 2021–2022 downloads are being argued), and an answer to defenses such as fair use.
Tell us about an earlier edition or another book with this text. If you know of an earlier edition, an edition abroad, a translation, or an anthology or other book that contains the same text, you can tell us from your report page. A person at PayAuthors checks each submission against the sources before anything in a report changes, and the report notes what changed and why.
Measured accuracy
We test the matcher against works lists filed in court, which pair a title with its registration number. Of the Bartz v. Anthropic opt-out works (Dkt. 619-3, Ex. J; 1,749 TX and TXu registrations) found in these catalogs, the matcher links 94.0% to the exact registration in the court record and 96.0% to some registration of the work. About four in ten of the remaining misses are entries where Ex. J itself pairs a title with another work’s registration. Of 7,001 textbook registrations in publishers’ works lists (mainly Cengage Learning v. Google, No. 1:24-cv-04274 (S.D.N.Y.), ECF 1236-1), it links 93.1% of those found in the catalogs to the exact registration and 96.9% to some registration of the work.
What a report counts. A report counts a match only when it passed every automatic check against the Copyright Office record or was confirmed in a review (see Record-match marks); every other match is shown as Not yet checked and left out of the counts. The figure that matters for a report is how often a counted match is right.
Person check (rescored October 23, 2026). A reviewer paid by the hour, who had no part in writing the matching rules, checked 450 matches by hand (400, then the 50 trial rows the second reviewer had checked): each row showed a catalog record next to a Copyright Office record, without the software’s verdict. The matches were drawn at random from the dataset’s matches. Under today’s rules (analysis revision 33), reports would count 310 of them, and the reviewer judged all 310 to be the same work (95% range 98.8% to 100%). Every match the reviewer called a different work, or could not decide, is one the rules hold back as Not yet checked. The reviewer has finished re-checking 48 rows on the edition question, so those rows are now included. Not included: 26 rows today’s name search no longer reaches. The first scoring, on October 13, sometimes compared a row with another catalog listing of the same title, not the listing the reviewer saw; it gave 228 of 228. Each row is now scored on the listing the reviewer saw. Some rows had been looked at by software earlier, while the rules were tuned.
Second reviewer. A second reviewer, who also had no part in writing the rules, is checking 350 of the same rows separately, in another random order and without the first reviewer’s answers or the software’s verdict, and has also checked a separate trial set of 50 matches drawn the same way. Of the 281 rows done so far, reports would count 194 under analysis revision 33: the second reviewer judged 191 the same work, could not decide on 3, and judged none a different work (95% range 95.6% to 99.5%, counting the 3 undecided as wrong). The first reviewer has now also checked all 281; the two gave the same answer to “same work?” on 266 (94.7%), and they never split yes against no on a match reports count. A separate blind check of 226 counted matches, drawn only for this purpose, will be added here when it is complete. Only a person’s check is used for these figures.
Same work, not same printing. A match is right when the registration covers the catalog book’s text. A registration covers the work in any printing or format, so a later paperback or ebook of the same text is the registered work; the edition matters only where the text differs (see the rules we apply).
Earlier software sample. A first sample of 100 matches, drawn from all candidate matches including those reports hold back, was checked in a software review: 91% were the same work (95% range about ±8 points). It is superseded by the person check above. The 100 reviewed matches are published as a CSV (20 for each matching method, so the totals above weight them by how often each method is used: title and author 61%, main title 20%, ISBN 16%, loose title 2%, publisher 1%), each with its verdict, a note on any mismatch and a link to the Office record, so anyone can recheck them. The review is ours, not an independent audit. A software review of 800 matches against the Copyright Office record is described under Record-match marks. These are measurements on these lists, not a guarantee for any one match: confirm each registration against the linked Office record.
Copyright Office records, recorded documents and contact details
Each registration links to the Office’s own record in its Public Records System. The Registration Details report also lists every recorded document that cites the registration: transfers, assignments, licenses, security interests, notices of termination and others, each with its document number, date recorded, parties and, where the Office has them, addresses. These come from the Office’s bulk recordation data (published January 2026, documents recorded through June 26, 2025); later recordings are only in the Public Records System. The address on a registration is the one given at filing, so a recorded transfer can show that the claimant named there is no longer the owner.
For registrations from about 2010 on, the Registration Details PDF also reproduces the Office’s PDF of the registration record, unchanged, in an appendix. That record includes the claimant’s address and any rights-and-permissions contact the applicant supplied. Readers can use it to check whether those details look current. Supplementary registrations, which correct or add to an earlier registration, appear on the Office’s record but are not yet listed separately in our reports.
Books3 ISBN lists, and why we go by the file list
Some Books3 lists, and lookups built on them, identify books by the ISBNs printed inside the Books3 texts. The best known, Peter Schoppert’s, holds 129,635 ISBN-13s. It does not say which file each ISBN came from, and a book’s text prints other books’ ISBNs: on an “also by” page, in a bibliography, or in an excerpt from the next book in a series. So an ISBN on that list can belong to a book that was never in Books3.
We check every ISBN on that list against the Books3 file list itself, under any name. Of the 111,169 ISBNs our catalogs can tie to a title, 68,563 (62%) belong to a title that a Books3 file carries; 33,848 (30%) have a Books3 file with a similar title that a person would need to compare; and 8,758 (8%) have no Books3 file with that title or a similar one. A lookup that goes by ISBNs alone would show those 8,758 books as in Books3. Our reports do not.
In a report, a book counts as in Books3 only when a Books3 file carries it: under the name searched, or under the name of the book’s author or a co-author, which a search by name alone would miss. When a book’s only link to Books3 is its ISBN on that list, the report says it is not in Books3 by its file list and leaves it out of every count. When a file with the same or a similar title is listed under another name, the report names that file and asks you to check whether it is your book; it is not counted. Title matching is automated, so a book whose Books3 file carries a very different title can be missed; the report shows the evidence either way.
Snapshot dates and The Atlantic’s search tools
Each snapshot answers a timeline question: was an entry in the catalog by that date? With 25 LibGen snapshots from 2015 to 2025 a report can show when an edition was first listed and whether it was listed in the snapshot nearest each documented download; the gap is shown. The June 5, 2021 snapshot is the one closest to Anthropic's LibGen download; the PiLiMi release 1 index (July 2022) precedes its PiLiMi download by weeks; the June 2022 and January 2023 LibGen snapshots bracket Meta's first use; the January 18, 2025 snapshot is from the same month as The Atlantic's. According to the Bartz v. Anthropic order, Anthropic downloaded at least five million books from LibGen in June 2021. The Kadrey v. Meta court found that Meta “first used a shadow library in October 2022,” about sixteen months after the snapshot. A match in the snapshot can help document that the catalog listed an entry before that acquisition. It does not show that a later defendant obtained that exact entry or file.
The catalog at the download. Each download is compared with LibGen as it stood then. A LibGen line counts for a download only if its file, identified by MD5, is in LibGen’s catalog in force at that download, under this record or another: for a download period with a known start (or a company’s founding date as the earliest possible start), the library’s last catalog dated on or before the start, or a catalog dated inside the period; for a download known only by an outer bound, any catalog up to that bound. Fiction and non-fiction catalogs exist on different dates (the April 15, 2019 dump is non-fiction only). The test is on the file, not on the record’s details: LibGen often edits a record’s title, author or ISBN between dumps, which moves the file to a new record without taking it out of the library. We checked every LibGen file in our catalogs against this rule. Every file listed by the end of a download is in the catalog in force then, except where LibGen gave a catalog ID to another book: in December 2015 it reused 248 fiction and 61 non-fiction IDs, and only 81 of the original files were listed again under new IDs. A line for one of the other original files that is listed before a download but not in the catalog in force then carries that note and is not counted for that download; its earlier listing stays in the report. Our build had also kept the original file’s MD5 for the later book under each of these IDs. Reports now show the later book’s own MD5, which we checked against LibGen’s own dumps of December 2015 and later. The date agreed for the Bartz settlement is not a download, so it keeps the earlier test: any catalog up to that date.
More on period events and The Atlantic’s search tools
For an event dated as a period (for example, OpenAI’s alleged torrenting from September 2019 through January 2020), the timeline also reports the first LibGen dump after the period ends, if it falls within 45 days, and how many of the lines it lists had a catalog entry added on or before the period’s last day (LibGen’s own record). For OpenAI’s alleged period that is the February 2020 pair of dumps (fiction February 8, non-fiction February 13), 8 and 13 days after the period ends. For the event dated only to 2018, the timeline also counts the EPUB and MOBI files listed in the snapshot compared, because the 2018 download script is alleged to have taken only those two types.
The Atlantic’s LibGen search uses a snapshot “taken in January 2025, after Meta is known to have accessed the database, so some titles here would not have been available to download,” and notes that “it’s impossible to know exactly which parts of LibGen Meta used to train its AI, and which parts it might have decided to exclude.” The Atlantic (a subscription may be required); the same caveat is quoted on Bruce Schneier’s blog. For Books3, The Atlantic’s Alex Reisner explained that the books are stored “as large, unlabeled blocks of text,” so he “extracted ISBNs from these blocks of text and looked them up in a book database”; of the 191,000 titles he identified, 183,000 had associated author information. The Atlantic, September 25, 2023.
The two approaches answer different questions. Ours documents what each dated catalog snapshot and a published file list contained, so a title added after a download shows up with the date it first appeared; a single later snapshot cannot separate those. A defendant’s own acquisition records, file manifests or logs would be more direct evidence of what that defendant obtained. Counts also differ because sources count different things: files, catalog records, identified titles, editions, or books with author metadata.
The court record
There is no single public ledger of every entity that copied every collection. The court records below are summarized as of the review date above; each is a statement about that case and its procedural stage.
The record, company by company
Huckabee v. Bloomberg (S.D.N.Y., November 24, 2025). The court denied Bloomberg’s motion to dismiss. The plaintiffs identified their works, alleged the works were in Books3, and relied on Bloomberg’s own paper saying it trained BloombergGPT on Books3. The court rejected the argument that the plaintiffs had to specify which files were theirs. This was a pleading-stage ruling, not a finding of liability. Decision.
Bartz v. Anthropic (N.D. Cal.). A June 2025 order described evidence that an Anthropic cofounder downloaded Books3 in early 2021, that Anthropic downloaded at least five million books from LibGen in June 2021, and at least two million from the Pirate Library Mirror in July 2022. Anthropic’s own interrogatory answer, filed on the public docket in January 2026, gives the days: Books3 on or around February 25, 2021; LibGen June 8–19, 2021; the Pirate Library Mirror July 8 to August 10, 2022 (Dkt. 554-6 at 4). We use those days for the timing check and mark them as the company’s statement, not a finding. The court treated building a central library from pirated books differently from using copies to train models. The case settled, with final approval in July 2026; the settlement does not resolve other companies’ conduct. 2025 order; final approval order.
Kadrey v. Meta (N.D. Cal.). The June 2025 opinion describes Meta downloading LibGen in October 2022 and using Books3 in training data, and granted Meta summary judgment on the training claims on that record; the judge said the ruling “stands only for the proposition that these plaintiffs made the wrong arguments and failed to develop a record.” Distribution through BitTorrent was left for separate proof. The opinion also describes Meta downloading Anna’s Archive, a compilation that includes LibGen and Z-Library, in early 2024, and says it is undisputed that Meta torrented LibGen and Anna’s Archive (Dkt. 598 at 11–13). In March 2026 the court granted leave to add a contributory infringement claim based on BitTorrent uploading. In its July 13, 2026 Answer in Elsevier v. Meta (S.D.N.Y. 1:26-cv-03689, Dkt. 78), Meta admits obtaining data from LibGen, Anna’s Archive and Sci-Hub to train some Llama models, and obtaining Anna’s Archive data in 2024 (¶¶82, 97); that is a company statement in a pleading, not a court finding. 2025 opinion; March 2026 order.
NVIDIA. A court allowed claims to proceed past dismissal on allegations connecting Books3, The Pile and a model-training dataset; that was not a finding that NVIDIA copied any particular book. The amended complaint alleges that NVIDIA sought high-speed access to Anna’s Archive in August 2023 and that management approved within a week, and that the offer included millions of Internet Archive lending books (ECF 235 ¶¶53–57). NVIDIA withdrew its motion to dismiss as to Anna’s Archive, Z-Library, LibGen and Sci-Hub (ECF 271, March 6, 2026), and its June 2026 Answer admits downloading data for AI research from annas-archive.org/datasets and libgen.rs, without dates, while denying that it improperly copied the plaintiffs’ works (ECF 321 ¶¶58, 77). Order.
DeepSeek. DeepSeek’s own DeepSeek-VL paper (arXiv 2403.05525, March 8, 2024, §2.1) says it cleaned 860,000 English and 180,000 Chinese e-books from Anna’s Archive for its document-OCR training data. That is a company statement; we know of no U.S. court ruling on it. Paper.
OpenAI. A November 2025 discovery order described as undisputed that an OpenAI employee downloaded books from LibGen in 2018; other details about derived datasets came from testimony and party assertions. Order.
Where a company’s own statements or a court record show it copied an entire collection, what an author then needs is the link between the work and that collection: the work, its registration, and the specific entry or file name in a dated source. That is the link a PayAuthors report is built to document. What the company copied, and when, has to come from the company’s own records, statements or court findings.
Time stamps
From October 13, 2026, when a report version is issued, the SHA-256 of its PDF is sent to an independent time-stamp authority (DigiCert, or FreeTSA if DigiCert does not answer), which signs that hash with its own clock under the RFC 3161 standard. Only the hash leaves PayAuthors, never the report. The signed time stamp is kept with the version and linked on the report page as “Time stamp (RFC 3161)”. Versions issued before that date were time-stamped afterwards, and the report page says so. To check one with OpenSSL: openssl ts -verify -in <report>.tsr -data <report>.pdf -CAfile <the authority’s certificate chain>.
Provenance: dataset build payauthors-2026-10-01-m64
Checksums of every source file
Additional sources
Z-Library records after PiLiMi, the Internet Archive lending catalog and the journal-article DOI lists. The two scimag files were read as a stream and not stored; their checksums are the Internet Archive’s.
| Source | File | SHA-256 |
|---|---|---|
Hugging Face P1ayer-1/annas-zlib3-index | train-00000-of-00005-421d14d609b06e30.parquet | c06f449014008b8a340e2867d410cca603600bed7b0789faf086e4d5f91ec062 |
Hugging Face P1ayer-1/annas-zlib3-index | train-00001-of-00005-33b56dc710868fcc.parquet | 0a10340de0523a670668108181b37c1bbd6fdfe3d6b72e0ca778b30f0cdb885a |
Hugging Face P1ayer-1/annas-zlib3-index | train-00002-of-00005-8801c913706c9de7.parquet | 03f8fb4e2130d081025c1b0678a43e16fd496f590f336481694ccfe3ccd16e1b |
Hugging Face P1ayer-1/annas-zlib3-index | train-00003-of-00005-11a7abbf65bdb652.parquet | 6eddb91d0a0133654434d69824e74344dabdba8de43cdc459744e1612a8e8735 |
Hugging Face P1ayer-1/annas-zlib3-index | train-00004-of-00005-9f57636ca249d4d4.parquet | d4ee2a61f1d75b2ebd758638f6162089b4cabf8d258be21b92df22fd3345ef7e |
| Zenodo 10.5281/zenodo.17609567 (MD5 5876dc0cb3ef4be79a35fbaf59d5bcbe) | majinbook_zlibrary.jsonl.gz | da1d50b04c97f79c6c8ac1030f74cd0f083104e62da19a3bffe1e1ed626ad071 |
Internet Archive scrape API, collection:inlibrary, fetched Sept. 30 – Oct. 1, 2026 (4,423,866 items) | inlibrary.ndjson.gz (our download) | 30d7c22621e6992cd323a5801617cd99762a309e1066a90c9228f8b99c2b0934 |
Internet Archive item scihub_doi_list_2017 (MD5 62b253ebdb9456a7e632e81d7e6ed800) | doi.7z | (MD5 as published by the Archive) |
Internet Archive item libgen-meta-20200213 (MD5 bdcdb69511fe0aa328b1afc659ab5714) | scimag_id_doi1_doi2_isbn_md5_pmid_pmcid.tsv.gz (read as a stream) | (MD5 as published by the Archive) |
Internet Archive item libgen-meta-20210806 (MD5 084992920a17ce3dc9fe65b69803339c) | libgen_scimag_dbbackup-2021-08-06.rar (read as a stream) | (MD5 as published by the Archive) |
U.S. Copyright Office, data.copyright.gov Registrations/Tabular/ (journal-article registrations) | reg_serial_2026_01.csv (401,155,206 bytes) | 02e034580e8d638d7a3c8fc471861452d22ceaec725d7b9cb56b686effffc71a |
U.S. Copyright Office, data.copyright.gov Raw unparsed uncategorized XML/ (serial records only, read as a stream) | USCO public data.zip (3,581,316,904 bytes, dated Apr. 2, 2024) | (read as a stream; not stored) |
Every source file behind the current dataset build. LibGen items link to their Internet Archive pages; “SHA-1 (archive)” is the checksum the Internet Archive publishes for the file (for the Wayback Machine capture, the capture’s recorded payload digest), which we matched before use, and SHA-256 is our own. Only catalog metadata was read from these files. The file IDs and MD5 checksums shown on report lines were collected in a second pass over the same files, each downloaded again and matched to the same SHA-1 and SHA-256 (the February 8, 2020 fiction dump, added later, in a single pass).
| Snapshot | Catalog | Archive item and file | Bytes | SHA-1 (archive) | SHA-256 |
|---|---|---|---|---|---|
| November 14, 2015 | fiction | libgen-meta-2015111420151114-fiction.rar | 169,417,419 | b66dc4473c41935a5061174c6adca77815b79c39 | fb28917f8f8102782092ba37c53ed89823202d0b900530706a86cff7b2a3b6f7 |
| November 14, 2015 | non-fiction | libgen-meta-2015111420151114-libgen_compact.rar | 335,440,447 | 722d792dcef25dec518bba879be42990329d9141 | 0927f0659320624e0d00a8288cc54548bee676d244df7ae1055f507d33b9f506 |
| December 26, 2015 | fiction | libgen-meta-2015122620151226-fiction.rar | 229,964,419 | b1575d11a43eddc8660a77383e53376a95b690ed | 2d8f895eca25dcc74e05221998b2cf06aa998dcb26c7ae4b7d37ca559ed850ea |
| December 26, 2015 | non-fiction | libgen-meta-2015122620151226-libgen_compact.rar | 343,348,007 | 0e1a623afee6070b2c5566e25d3fb386f1f5d3b9 | 43a80461adb48fbf380bc3c385b0558c122d082cef02c1586fbf1d6c17452177 |
| January 9, 2016 | fiction | libgen-meta-20160109fiction.rar | 230,176,429 | 44a2dd92c6eedb1378613be529aca3bd23bea904 | c3aa13dfa7ddb4e842a3ba55dddbaf9fe30908831b5f4faba2e66631a3fb88f5 |
| January 9, 2016 | non-fiction | libgen-meta-20160109libgen_compact.rar | 343,922,223 | 6a627c137f3b54aa9b182e9d6b532905fc85a7a3 | 598af05e5b821e7eaec45f8bdb8575639bd38eca5ed5620f0feb6d27a346edab |
| June 17, 2016 | fiction | libgen-meta-2016-06-17fiction_dbbackup-2016-06-17.rar | 240,314,699 | a9b07c67a247c69d90a56218ce01e3a5a9d941c0 | 3590b1ead76d2e0bd8247f917ccfe90d237744b05e62d085152f9544a062ba4c |
| June 17, 2016 | non-fiction | libgen-meta-2016-06-17libgen_dbbackup-2016-06-17.rar | 190,512,122 | 15856da1e67fd0c630f6d39a8760dd6a9d50c30d | 3d54db847419b450df4e40830ffc95288230bf66c8c9ddc661e7809647dfb241 |
| November 4, 2016 | fiction | libgen-meta-20161104fiction_dbbackup-last.rar | 241,894,340 | 5bfb65ba9defb1c572c612d094e19b1fc126c7ec | 9db15b0c17e6f8e6cce4109512a187d50057c19ba525f27502b92408543a3d61 |
| November 4, 2016 | non-fiction | libgen-meta-20161104libgen_dbbackup-last.rar | 199,327,817 | 4609fce0f64bab73352b4e775e58de2747f920ef | 615b16cdcc31e849c2eab83b11bb54caa7343b710595c25dc9fefad5f7c62029 |
| December 30, 2016 | fiction | libgen-meta-20161230fiction_dbbackup-2016-12-30.rar | 243,419,703 | f4a26cd213ffda944bea06e1e33c615646cdae97 | 972a14c00e53de461f7ac9c3cbfc9a0739a5dd02ad3a4cf3588c2376fa40bb83 |
| December 30, 2016 | non-fiction | libgen-meta-20161230libgen_dbbackup-2016-12-30.rar | 205,108,150 | acae4bc3f6952239fccc10f51a183e597f3044c5 | 18572f6615ffce8d5cb8503d52a4751ca3abb0b641572ee7e368599da43a8447 |
| May 12, 2017 | fiction | libgen-meta-20170512fiction_dbbackup-2017-05-12.rar | 456,237,043 | 8d9be107c93d3e4bf454f4c6cd8720210de01a1f | fdfe5d94a98ffd1c39e091fb99815a6998a287e572e8bc74c581a08345f19ff2 |
| May 12, 2017 | non-fiction | libgen-meta-20170512libgen_dbbackup-2017-05-12.rar | 213,571,994 | db4d349b5e22a672dac1828369353c3a683374c0 | e37193c21088da07c8e6a4f1290dc66678caf13975ce658c26bcf587148fa3aa |
| April 15, 2019 | non-fiction | libgen-meta-20190415libgen_compact_2019-04-15.rar | 282,762,388 | 00f3772fc4942f7a48bc8c1b91a9de8880e6d877 | b5fae3c84a6a40512371dd6a8a26ab422feb67c6184283f6e64568f36e6f4660 |
| February 8, 2020 (folded with the February 13, 2020 non-fiction dump) | fiction | Wayback Machine capture of 2021-02-17 (web.archive.org/web/20210217010544/http://libgen.rs/dbdumps/fiction_2020-02-08.rar)http://libgen.rs/dbdumps/fiction_2020-02-08.rar | 801,097,717 | c0761e9e7d8fe98d32a98195f89e564a37cd41c6 | f28145748b8cb9f730eb879c406638ebe78f43423b7b99217b6e90800e6a6bd4 |
| February 13, 2020 | non-fiction | libgen-meta-20200213libgen_compact_2020-02-13.rar | 269,446,563 | dbf10c896c4d30c5284f3ec6229990a0d7cdee30 | 9fb734297d87765fbf2332f139dda4cf56d2f2f2cf0c2ec91425f75380259363 |
| February 12, 2021 | fiction | libgen-meta-2021-02-12fiction_2021-02-12.rar | 926,040,033 | fed262df72583676e838ae3a18ec9fc205065d73 | b2ff2775dff911f3e9f6599d18d34a4ee4a714ac4f9e76513e0e2c1013a15dec |
| February 12, 2021 | non-fiction | libgen-meta-2021-02-12libgen_compact_2021-02-12.rar | 328,162,453 | 5b93fb99ff9b821387ff8f12c63dfebd651f0589 | bef9a8430de38848561849d7b93aeae06ee0c2b62b9658e3949da281871f60fd |
| June 5, 2021 | fiction | libgen.rs-dbdumps-fiction_2021-06-05.rar-2021-06-05libgen.rs-dbdumps-fiction_2021-06-05.rar-2021-06-05-b4787ae8-00000.warc.gz (Internet Archive web capture) | 934,951,854 | f1ed9bc84cc931cd72c9abe1397617de5670c56f | 8c9b782a51b4fb358ad3118c6eaf10a622576b55c21536c80ef3b0c13f59fd5b |
| June 5, 2021 | non-fiction | libgen.rs-dbdumps-libgen_2021-06-05.rar-2021-06-05libgen.rs-dbdumps-libgen_2021-06-05.rar-2021-06-05-a77164ac-00000.warc.gz (Internet Archive web capture) | 4,075,381,025 | 1b459a44a83ee0ab0505c1c43a8da857d1c5c8b4 | 6f40db38486bad5250cbef257b74a23e5cd25837729be9713910b24a7d4020b2 |
| March 7, 2022 | fiction | libgen-meta-20220307fiction_2022-03-07.rar | 985,753,783 | 5e9b00e36162480f70fc8971bf5bd7ab79521de0 | 30ac56c795a57ace4bd9d549bccb61ce45bf31fc0860cb99cd6d79bf3d596d00 |
| March 7, 2022 | non-fiction | libgen-meta-20220307libgen_compact_2022-03-07.rar | 367,472,632 | 8e906f5f22ccdac807c23c015e6fbbebd6d0d186 | e2645975a21a0507365b886f1d7eb61f3d3f9690a4912150b0958319099de3ed |
| March 8, 2022 | fiction | libgen-meta-20220308fiction_2022-03-08.rar | 985,829,360 | f24216a4dca7c7d0fecdb5d424644f4d885d3dd0 | a68eee73aa4337f5a5126da32f9868b6a2700bf078c9c0fd00300d4c50f48318 |
| March 8, 2022 | non-fiction | libgen-meta-20220308libgen_compact_2022-03-08.rar | 367,637,914 | bafbe9add97c7fdcffe9d64f27e1d8f4ce8c3b6c | 0cc175e41f042f31f485b2b669a52c258ddc2ed40e43b82ad479a87f437fde35 |
| June 19, 2022 | fiction | libgen-meta-20220619fiction_2022-06-19.rar | 997,491,750 | a0b5ab9a0074f381f8fd729e59cb1fe55db1e583 | cdb8a821e702785da84ddb448fdf92bb291e06f30aeb3a20792e773d9b7bbc12 |
| June 19, 2022 | non-fiction | libgen-meta-20220619libgen_compact_2022-06-19.rar | 376,799,809 | aa9a072af84d2315d0e2c3e0ca1eb119dee66617 | c989f6e8300a444c01b2bb99ac21d51a3aca2b3a635366cc6a42751310b9ef56 |
| January 13, 2023 | fiction | libgen-meta-20230113fiction_2023-01-13.rar | 1,103,284,283 | 9d1724ad11bf03053c7e5d7ee6fa105a7caa9b4b | 197b72839f0d9665726c8cbd4168482e1c42654926a55d6ed109f7d7541c8a3e |
| January 13, 2023 | non-fiction | libgen-meta-20230113libgen_compact_2023-01-13.rar | 415,009,403 | e597c13ff1455eb7efeeb5f29275c019aeb96750 | 5f87a00eefba1e93f858bf74c96a5a9a1d0f9aa5188415202e7c247b6921c367 |
| April 29, 2023 | fiction | libgen-meta-20230429fiction_2023-04-25.rar | 1,133,284,237 | b6c71aa3940bd795c741124b957f57b89b3dc4bc | c5aa780f89412ba8fbbfff8b75bc17a0276e249b27b75ff97d864860e912b508 |
| April 29, 2023 | non-fiction | libgen-meta-20230429libgen_compact_2023-04-29.rar | 429,176,031 | f178bc51cfb942d56891275c2b0a549230b32a72 | b4dbb18e038b0f4229b4ab2b957993ab42bb116b48c56e8f5ba715af5083eee1 |
| May 13, 2023 | fiction | libgen-meta-20230513fiction_2023-05-13.rar | 1,138,348,960 | 3dc1076c8f7b5c231423576b39c134dcc6aa5cd1 | f2c488fd745fbd3ca3d2232ecb5b9c64c8f709dbe077f1e758d0d548cdc4e028 |
| May 13, 2023 | non-fiction | libgen-meta-20230513libgen_compact_2023-05-13.rar | 430,711,935 | 0442b5658304a837677a5f7b5f916d3fbe8d06a1 | 54705986f983d0dfcec53846113d1f2faa7fe46f0438331357c985dc83d15506 |
| January 6, 2024 | fiction | libgen-meta-20240106fiction_2024-01-06.rar | 1,188,758,769 | 3364b11ac9c02353d9c14162e6606252e9199dc7 | be3ee7bf05f824af431fbe821cd3783894438bace8f09ca5936e0d7d06caa0e7 |
| January 6, 2024 | non-fiction | libgen-meta-20240106libgen_compact_2024-01-06.rar | 488,627,470 | a121566db551406e62acdcfc79ca339a8943da53 | fcbd99d7fa18e67ef7732ad03316e6a723ff8556149ac9789c87c160817e4bc5 |
| June 28, 2024 | fiction | libgen-meta-20240628fiction_2024-06-28.rar | 1,261,925,704 | 262fcf82d4e93d15d74c85860d4bfe3502678644 | 3456d0a3199df8735eb42f50210baa275ef0344fcff5d7a8bcb12cbc0f417b32 |
| June 28, 2024 | non-fiction | libgen-meta-20240628libgen_compact_2024-06-28.rar | 507,660,382 | 2daaa7353a3283ef859d0c0e8f7f8cc90fe1a365 | 4e3746afd983e61521956e4b1c6783fba58072a2b3f01cc88bfeffb94ed6dece |
| October 2, 2024 | fiction | libgen-meta-20241002fiction_2024-10-02.rar | 1,278,582,785 | 296dc599358becc9b0dac7b9f09c7998b967845e | 7cdd3d2b92cf80b86013f1d8cd3e24f3f689a3ab5bcfd0a627a20327784bc1c9 |
| October 2, 2024 | non-fiction | libgen-meta-20241002libgen_compact_2024-10-02.rar | 514,088,078 | d1336ba78c25353bf69551c627684a7951625e17 | a8bd80490de629733e61419fbf5ad063ad3565da92cae621d2c6b187f8b982b8 |
| November 16, 2024 | fiction | libgen-meta-20241116fiction_2024-11-16.rar | 1,288,550,033 | da2faa57b3f603b7125fde1dd5a4ff793b582625 | 92ead19d68e4b992c3e9c935bb4d6a5cc653bed6eed44ed237d3e88143fce202 |
| November 16, 2024 | non-fiction | libgen-meta-20241116libgen_compact_2024-11-16.rar | 518,055,736 | 44bbbf2cde48ee61bd127e05a904f9b67b4232a6 | 8fd4ef4aa5c97326ddadb3dc70f11723d33a76f26c9b4b7aea93be3f357516a0 |
| December 17, 2024 | fiction | libgen-meta-20241217fiction_2024-12-17.rar | 1,296,846,448 | 557c037cb4b75d26b7e361ffba5faa7d92eb0fda | 8bba3c8f90a2800a32252253501a98d0c5a311479a7527ffd9ac41846e58695b |
| December 17, 2024 | non-fiction | libgen-meta-20241217libgen_compact_2024-12-17.rar | 520,829,318 | 4a95bd454e455a3416b29a05c2ab26ca7020e74c | 87a74ca74c07f9766c9e5392eacf6a6d354b4edce9983225e4ba19fbd8dcb4a0 |
| January 18, 2025 | fiction | libgen-meta-20250118fiction_2025-01-18.rar | 1,304,646,878 | 5e51ccfba19e2f13bfdda653dae4429ef677cbd5 | f43a5f4e0e7782964d50c3e8b9f47d86abc3cd6039cfffb135e32ef14596c7c9 |
| January 18, 2025 | non-fiction | libgen-meta-20250118libgen_compact_2025-01-18.rar | 522,699,977 | 208a4fa0dfa4aa06308c253d968a7aec5c7066f9 | 1dd45de89caafacbe9f1b412d027fbf2807e955924e2799348906e9c4460a117 |
| March 8, 2025 | fiction | libgen-meta-20250308fiction_2025-03-08.rar | 1,313,545,516 | cedd7c973133431b5b69dbf02bdf1cbee53c80dd | 946a7efa87fa51fd436b0117db51b290d16ca954558b10729679fb2f3ed725a4 |
| March 8, 2025 | non-fiction | libgen-meta-20250308libgen_compact_2025-03-08.rar | 524,887,099 | f8ec8baaf8a4405c7543093535ef9dbe531cb981 | 3cc299f25c24ee7ba75c9adc890aae5c9aed8c3ceebf8a71eb9c72a554306299 |
| June 3, 2025 | fiction | libgen-20250603fiction_2025-06-03.rar | 1,324,697,887 | 0b67a4117d62dcdd0b2e63691d6c6ba8f9315bf6 | e22c00ea1d31b7d154e76813552b72aba480461d919c4bf843465fbef6180e26 |
| June 3, 2025 | non-fiction | libgen-20250603libgen_compact_2025-06-03.rar | 530,821,763 | 7b763fdc80b1e6cc3f4c6def3bcb2fe9b8067024 | e3e702e6496364073cb38f536efc43b325606690f76cba4c18cf10904075f5b2 |
| Source | File | SHA-256 |
|---|---|---|
| PiLiMi (Z-Library) index | Metadata-only copy re-hosted on Hugging Face, P1ayer-1/annas-archive-index, 18 parquet files; records tagged by PiLiMi release (July and September 2022) | b3005a74ca30e2ca6dc5a8ddc8a275db340f3877271672ae752555b9d8f08f15 train-00000-of-00018-0de08d15f83d5a95.parquet086083645d2dee61e9ca8e069de6f5dfa854a96b63417a60e72a0f5d8e339907 train-00001-of-00018-c5b87dfcfaaca263.parquet4d8dc8fdd98d0b065eed786c2756873b652a8383a62fbb76805f113276ecc7a0 train-00002-of-00018-b0dd6be7e63187a1.parquetf731700cb6681f16b5d7011d33b6cfe29352daf6ac15f38c54662db0f9586f37 train-00003-of-00018-0bbb9ba4be152d8b.parquetdc27ecd99d3245632291262d7957e579cff2a2929b8324891b3c5d2674b207d8 train-00004-of-00018-ea10c700bc0c6835.parquete44756886570f26764a317b66345cb85166a872022d87ba1aafe89750dc1d350 train-00005-of-00018-fa58b4f7bc3ec283.parquet47505e79575f144f45eefeae30e882fcd08e0d58841aadf397911072c6cf8971 train-00006-of-00018-ed59a37788bfedc5.parquet7f811cfd2ae5a2d03ada29221430af4875b1602c38bae298bbde1d0885e85370 train-00007-of-00018-e36786f5069ebd1d.parquet2ae0a906abe3ec0538f70f7948556800ecb264160b9bfe338c592a3cd36a96e2 train-00008-of-00018-91a3071dc8471e38.parquet1ee01c5c8864d43a1393eca3adc49fc16639b1ccadf88366c0cec1d035d2ac40 train-00009-of-00018-0b5b1757686f33bd.parquet500159efccc2fe74ec365f244843c92d2eb9d81b68dc7dacab7c390e4363f2cf train-00010-of-00018-23cd38004aa9fe5e.parquetb1046096da29e2aff76a45951bbc96bfeff2f1029aa425823e99e7f4a583927a train-00011-of-00018-f0423707a2f379dd.parquet60b3d2c0ee333db27437ab87d8c8b6cb1b9a5dcd378e0d30d2cf546fe6084667 train-00012-of-00018-3541a214f73c36cf.parquete851a1ef375b8a71fd39563cd733b3fc619a3ddaf5e9589abe7a9e4cc3e3e24f train-00013-of-00018-94ea5b888df94a3e.parquet5242e4917d79406d4a57dacb93874b8931051fe114901f55aee61cc67ca02c7f train-00014-of-00018-ff19e1d3c674dce3.parquet048cdf4b90fcad0142968b56defe7ed43a26b27d2f828882173109da14912938 train-00015-of-00018-dd633849fdd8070c.parquet90d4903cb35e4f9fd5ddb9542ed57d046c8c4e18df5e76af09c200862921fe83 train-00016-of-00018-cb1438e48baf670a.parquet858089c6992df6962bfb05cef701558ba6fe4360dea77c0b63953ea7a7a79c5e train-00017-of-00018-b2b51d65a0cfde19.parquet |
| Books3 file list | filenames.txt, github.com/psmedia/Books3Info (Peter Schoppert) | eb34489806bc36674a36515f9d5eb1d07a06edbd53ef2ce6aeb532e027bd7faa |
| Books3 ISBN list | Books2_ISBNs.txt, same repository (129,635 ISBN-13s extracted from the Books3 texts) | 6d67cb232fc60b10e34fe0afc7ffb409528f519b49d57cc79647179fd9ca99e3 |
| U.S. Copyright Office registrations | reg_non_dramatic_literary_work_2026_01.csv, data.copyright.gov Registrations/Tabular (non-dramatic literary works, as published January 2026; 4,789,987,093 bytes) | 4bfba9f70bf47954d964f9350c495c4bc953149521a1e25b10f3bc9d2b925336 |
| U.S. Copyright Office recorded documents | recordation_2026_01.zip, data.copyright.gov Recordations (published January 2026; documents recorded through June 26, 2025; 1,630,294,321 bytes) | 99ec235192fd03203e7d6f95466774f3f5f07cef37ac7d606fd196425cd707a9 |
This page is general information, not legal advice. Court records and source datasets may change or be interpreted differently as litigation proceeds.