PayAuthorscom

Methodology

Changes and builds

The dataset build in use now is payauthors-2026-10-01-m64, built October 5, 2026. Every page, report and /api/health reads it from the same place.

Each paid report records the dataset build and analysis revision it was issued under, and each report version lists what changed from the one before. This page lists every build and revision, so a report can be traced to the data and rules behind it. Each dataset build is kept; the methodology describes the sources and rules.

Dataset builds

BuildBuiltWhat changed
payauthors-2026-10-01-m64 in useOctober 5, 2026Same catalogs. Records filed under a pen name that Copyright Office records tie to a legal name are matched again with the legal name added (held for a person to confirm); co-author index folded into the main index.
payauthors-2026-10-01-m63October 1, 2026Same catalogs. Matching rule v6.3: translations, study guides, adaptations and disagreeing volume, issue or edition numbers are set aside.
payauthors-2026-10-01-m62October 1, 2026Same catalogs. Registration matching rule v6.2 (precision fixes to title matching) and journal-article registrations.
payauthors-2026-10-01October 1, 2026Added later Z-Library catalog records and the Internet Archive’s lending-library catalog (22,370,394 records in all).
payauthors-2026-09-30September 30, 2026LibGen fiction catalogs parsed again (2,511,642 → 2,515,792 records).
payauthors-2026-09-29September 29, 2026Added Z-Library records from the Pirate Library Mirror (PiLiMi) index and further LibGen non-fiction snapshots.
payauthors-2026-09-26September 26, 2026First build: dated LibGen non-fiction and fiction catalog snapshots, the published Books3 file list, and U.S. Copyright Office TX registrations.

Analysis revisions

An analysis revision changes how the frozen catalog results of a report are checked and counted. The same list is printed in each report’s notes and JSON. A report first issued under an earlier revision gets a new version when it is next opened.

RevisionDateWhat changed
Revision 16October 6, 2026A registration that covers only something added to an existing text (an abridgment, a performance, an introduction, illustrations or notes, or work made for hire by someone other than the author) is no longer confirmed automatically; it is held for review. Recorded transfers and other recorded documents are listed under a registration only when the document belongs to it: about 4,200 links in the Copyright Office’s recordation file that cite a registration number for an unrelated title were removed, and each document now shows the title it lists.
Revision 15October 6, 2026Books of a series are no longer joined into one book because they share a series name, or a registration our matching found for several of their titles: “Skin Walkers: Mason” and “Skin Walkers: Taken”, or “Slave Ship” and “The Mandalorian Armor”, are now separate books, each tested against its own confirmed registration. A registration the Copyright Office record confirms for two titles still makes them one work, so a collection counts once. Catalog titles that begin with the author’s name are read without it. Book counts, the § 412 counts and Work IDs are recomputed.
Revision 14October 5, 2026The registration check now states confirmed matches as its figure; matches not yet confirmed are given as a separate number instead of in the denominator. The method notes describe the full-record check behind matches confirmed by main title.
Revision 13October 5, 2026Translations: a catalog copy in another language is no longer counted as the book’s presence in a dataset. It is listed as a translation with the registration of the original, and books found in a dataset only in translation are named under the § 412 table. 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. Two titles that share only a subtitle are no longer joined as one book.
Revision 12October 5, 2026Books3: a book now counts as in Books3 only when a Books3 file name carries it. An edition whose ISBN is printed somewhere in the Books3 texts, with no Books3 file under the name, is shown as supporting evidence and is no longer counted in the date screen, the § 412 table or the company counts. Reports now show books per company and no dollar amounts.
Revision 11October 4, 2026One registration-date test for every part of a report. The § 412 table, the dollar maximums and each line’s date-screen mark now test a book as a whole: its earliest confirmed registration against the earliest date any copy of it in that company’s dataset could have been taken, and against the company’s first known download of it (revision 10). Before this, the table tested each title line on its own and the maximums did not apply revision 10, so the two could disagree. Books found only in another language, with no registration, are now named separately as likely translations instead of being counted as more books.
Revision 10October 4, 2026The registration-date screen now times each company against its earliest known download of a work. A later download by the same company counts only if the registration also came before that company’s first download of the same work (any copy, edition or format in the report), because § 412 runs from the start of a continuing infringement (Derek Andrew, Inc. v. Poof Apparel Corp., 528 F.3d 696 (9th Cir. 2008)). Counts against documented downloads look only at the company’s documented earlier downloads. This can lower, never raise, a report’s counts.
Revision 9October 4, 2026Three matching rules tightened. A catalog author that names an organization (a college, bureau or board) no longer matches on a shared word. A short catalog title is no longer confirmed automatically when the registration adds a subtitle, which often marks another book of a series; it goes to review. A registration for an edition published two or more years after the catalog copy is listed but not counted, because a revised edition’s registration covers only its new material. These apply to reports first issued under this revision or later.
Revision 8October 4, 2026Matches found only by main title, by title words or by publisher are no longer confirmed by the automatic checks alone; they are counted once a review confirms them (a catalog ISBN printed on the Office record still confirms them). This applies to reports first issued under this revision or later; reports issued earlier keep the earlier rule for these matches until a person reviews them.
Revision 7October 3, 2026OpenAI: the 2018 LibGen download is dated October–November 2018 as alleged (the date screen now counts from October 1, 2018, not January 1, 2018), and an alleged February 2021 download of Books3 (via The Pile) is added. Labels: court orders are labeled as orders, and Meta’s Books3 use is dated by its own LLaMA paper.
Revision 6October 3, 2026Counts, the date screen and dollar figures now use confirmed registration matches only; a match not yet checked is shown but not counted, and is double-checked when the report is opened.
Revision 5October 3, 2026Software review applied to a first random slice of the data (about 8,400 matches, each reviewed twice): matches the review confirms are marked, others found to be another work are removed, and a registration that covers only one edition’s added material (another person’s introduction, editing, illustrations, translation or adaptation, or a new edition of a public-domain classic) now matches only that edition.
Revision 4October 3, 2026Record checks updated: a catalog ISBN printed on the Office record now confirms the match; a different volume number now drops the registration from the line; the date window now runs from the publication date; the word “translation” in a book’s subject no longer counts as a translated edition; and matches decided in a software review against the Office record are applied.
Revision 3October 3, 2026Registration-scope check: a registration that excludes the searched author’s contribution, or registers a collective work by others, is kept as evidence but no longer counted; work counts and IDs recomputed.
Revision 2October 3, 2026One work ID per book, shared by the page, PDF, CSV and JSON.
Revision 1October 2, 2026Registration matching rule v6.3, the timing rule (download start or the library’s own record date, whichever is later) and the person check (a registration naming a different author with the same surname is left out).
Earlierbefore October 2, 2026Reports generated before October 2, 2026 used earlier, unnumbered analysis.

Site releases

Changes to how reports are issued, licensed or presented. Changes to wording that do not affect results are not listed.

DateWhat changed
October 6, 2026Reports are issued as numbered versions. Each version’s PDF, CSV, notes and JSON are built once and never rebuilt; the report page lists every version with what changed and each file’s SHA-256, and payauthors.com/verify checks a file. Pages and reports name the dataset build from one source. Terms: the license covers the counsel, experts and agents a buyer retains; refund on request if the review lowers the counts.