How RedLectern is built

Who runs RedLectern

RedLectern is a small, independent project, run under the name RedLectern.

Email: privacy@redlectern.com

Courses

Courses come from MIT OpenCourseWare, The Open University (OpenLearn) and TU Delft OpenCourseWare. The catalog lists 2,596 courses; 1,965 of them have a course page on RedLectern, and the other 631 link to the course on MIT OpenCourseWare.

Course pages are built from what the provider published: for MIT OpenCourseWare, the course's downloadable package of pages, files and video listings; for the other providers, the course's own pages. Each course page credits its source and license.

The licenses on 1,964 of the 1,965 course pages, as each provider states them:

• MIT OpenCourseWare: CC BY-NC-SA 4.0 (1,952 course pages)
• MIT OpenCourseWare: CC0 1.0 (4 course pages)
• MIT OpenCourseWare: CC BY 4.0 (2 course pages)
• MIT OpenCourseWare: CC BY-NC 4.0 (1 course page)
• MIT OpenCourseWare: CC BY-SA 4.0 (1 course page)
• The Open University (OpenLearn): CC BY-NC-SA 4.0 (2 course pages)
• TU Delft OpenCourseWare: CC BY-NC-SA 4.0 (2 course pages)

Course order

3 courses carry the Course order label: one order for the whole course, lectures with the problem sets and exams placed between them.

For a course built from an MIT course package, a script sets that order only when every problem set and exam comes from an ordered list in the course package itself (never a table read off a page), and each one is tied to a lecture by a quoted line of the course's own calendar or syllabus. The lecture sequence comes from the package's ordered lists (its lecture videos or its list of lecture notes) or, failing those, from a numbered lecture table on one of its pages. A course with no problem sets or exams is never labeled. A model never sets the script's order. A course that meets that bar is still shown without the label until its order has been checked against the course's live pages and approved.

2 of the labeled courses are pages from an earlier version of the site, made before the catalog existed.

The other 1,962 course pages show Browse materials: the provider's own sections, in the provider's own order, with no course-wide order claimed.

Books

The Books page has 10,833 listings of free books. 234 are for books cited in the reading lists and syllabi of MIT OpenCourseWare courses, quoted as the course gives them. The other 10,599 come from RedLectern's own list of free books, gathered from web searches and from Project Gutenberg's catalog. A book can be listed more than once, for example under two titles or inside a collection that is also listed: the 5,295 listings that open in the reader lead to 5,248 different reader pages, and 45 of those pages are reached from more than one listing.

Free copies were found and checked with AI language-model agents, which opened each link and recorded a quote from the page as evidence that the book is offered free there by its author, its publisher, an open-license host or a public-domain library; shadow libraries and unofficial uploads are rejected. A later check fetched each free-copy link outside Project Gutenberg; a script flagged the ones that might be a sample, a single volume, a sign-in page or a dead link, and AI agents judged each flagged copy that was not a dead link. Copies found to be a sample, the wrong book, a dead link or not free to read were replaced or removed, and where a flagged copy was found to hold only part of the book, its listing says which part (such as “Volume 1 of 2 only”). Project Gutenberg e-texts were not part of that check; some hold only one volume of a longer book, and not every such listing says so. Each listing shows its license or public-domain basis as found.

5,295 listings open in RedLectern's reader; the rest link to the free copy at its source.

The reader

Every book in the reader comes from Project Gutenberg (gutenberg.org), whose volunteers digitize and proofread editions that are public domain in the United States. Each book page names its Project Gutenberg e-text number and links to it. The reader has 5,277 book pages, 29 of them an alternate edition of another book (for example a play's original spelling). It also has 1 Project Gutenberg index of works; it is not counted as a book page or a free copy.

A script prepares each text from Project Gutenberg’s file:

• It keeps the part between Project Gutenberg’s start and end lines, leaving out Project Gutenberg’s header, license and footer text.
• Mostly near the start and the end of the text, it removes many of Project Gutenberg’s own lines that are not part of the book: the volunteers’ production credits (“Produced by …”), notes about Project Gutenberg’s other files of the same text (such as its HTML version), and stock lines about the file (how italics are marked, that obvious printing errors were corrected). A note that says something about the book itself stays.
• Where Project Gutenberg’s file carries its note on how the book’s U.S. copyright was cleared (“Extensive research did not uncover any evidence that the U.S. copyright on this publication was renewed”), it usually takes that note out of the text, and the book’s page quotes it beside the book’s source, as Project Gutenberg’s note.
• It drops illustration placeholders; the book’s page says how many it left out.
• It builds the contents from several limited patterns. A short, separate line beginning with a numbered Chapter, Part, Book, Letter, Canto, Act or Scene, or with Stave, can start a section directly. Exact front- or back-matter lines—Preface, Introduction, Foreword, Prologue, Epilogue, Afterword, Dedication, Appendix, Postscript, Introductory Note, Author’s Preface or Translator’s Preface—also start sections. In plays, a separate Induction, Prologue or Epilogue can start one, as can a cast heading when a short cast list follows. If the ordinary numbered patterns find little, a second pass tries ordered forms such as “CHAPTER ONE,” “Part First” or bare numerals; it uses that pass only when it finds a sequence of at least three sections, preserves every paragraph and does not put most of the text under one section. Another pass can match a later line to an entry in the book’s own contents list, or recover a malformed numbered heading when the recognized numbers before and after it show the gap. These patterns can still miss a heading or mistake a line for one.
• It adds “Opening” for text before the first heading and “(part n of m)” when a section is too long for one page. If an oversized section’s heading may run past the text it names—an opening, a front- or back-matter section, a heading recovered from the contents, or the book’s own title—the first piece keeps the book’s heading in the body. The contents list and chapter previous/next links keep that source heading where it begins. Later pieces use the book title and reader part number. Neutral wording is used for the part’s heading, part navigation and page metadata, and for paragraph labels in search, citations, progress, bookmarks and quote announcements, whenever the source heading is unsure. A reviewed list applies the same rule to exact headings found to run over other works or sections. Reader descriptions say that the page has the full text; they do not state a chapter count.
• It splits a very long book into parts.

Apart from those labels, it doesn’t edit the book’s words, and no model writes or rewrites any of a book’s text. The 17 books with LibriVox audio keep exactly the text their audio was matched to. A long book whose cleaned-up text would come out in fewer parts gets less of this clean-up, so every part page it already had stays. Older Project Gutenberg files don’t always mark where the text starts, so some of Project Gutenberg’s own words or a wrong chapter break can slip through: please report one if you see it.

Audiobooks and text highlighting

17 books in the reader have human narration from LibriVox, whose volunteers record public-domain books; LibriVox recordings are public domain in the USA. A recording is used only when LibriVox read it from the same Project Gutenberg e-text the reader shows; its audio files are LibriVox's own MP3 files, unaltered, served from RedLectern's storage.

In 14 of these books the sentence being read is highlighted as it plays. A script matches each recorded section to the text sentence by sentence; a section highlights the text only when that match passed its checks for exactly the version of the text on the page, and only confidently matched sentences are highlighted. Every other section plays as audio only. See the audiobooks list.

Quotes

The Quotes pages show lines from books in the reader. Candidate lines were gathered by AI language-model agents searching quotation sites, study guides and the texts themselves. A candidate is kept only if a script finds the same words, in the same order, in our copy of the book; differences of punctuation or capital letters are allowed, a changed, added or missing word is not. A line whose only source was Wikiquote is not kept.

What a quote page shows is always the book's own text from the reader, never a candidate's wording, and each quote links to its passage. Every time the site is built, each quote is checked again against the reader page; one that no longer matches is dropped.

Authors

Author pages list the books by that author on RedLectern. Where an author's birth and death dates, occupation or notable works appear, they come from Wikidata (CC0), and the page links the Wikidata entry it used. They are shown only when that entry is tied to the books on the page: a birth or death year in it equals the one in Project Gutenberg's catalog entry for one of those books, or a notable work it lists is one of them; entries reviewed as a different person are not shown. A year is shown only when Wikidata states it to the year and its best-ranked statements agree on it, with “c.” where Wikidata marks it as approximate. A statement Wikidata marks as presumed, possible, disputed or otherwise uncertain (a place given only as “near”, for example) is not shown. Nor is a birth or death year that Project Gutenberg's catalog entry for one of those books gives differently for that person, or marks with “?” where Wikidata does not mark it as approximate.

Where AI is used

AI language models are used to:

• search for free copies of books and check their links (see Books);
• nominate candidate quotes, which are kept only if a script finds them in the book’s text (see Quotes);
• check a course’s order against the course’s live pages, among the reviews recorded before the Course order label is approved;
• choose which of MIT’s own lines the study routes for the courses on the Start page quote, and list their first sessions from MIT’s pages. A route shows MIT’s words in quotation marks, each checked word for word against a saved copy of its page, with plain labels; it doesn’t summarize them;
• pick the works in the curated reading lists and check each one against the archived MIT course page, and write a few short notes there on a book’s role in a course (such as “Recommended reference; not a required text”);
• write short labels: the title of a formula sheet whose MIT file name didn’t say what is on it (read from the sheet’s first page), and how much of a book a free copy offers (such as “Volume 1 of 2 only”);
• help draft RedLectern’s own interface text, such as this page: RedLectern describing itself, not a summary of a course or a book.

No AI-written summary of a course or a book appears on RedLectern. Models are not used to write or change the words of a book or of a quote, or to set the order the course-order script derives.

What RedLectern doesn't do

• Change the words of a book, or show a model's wording as a quotation.
• Show an AI-written summary of a course or a book.
• Sell placement: retailers cannot pay to add or rank books or courses here, and the site shows no display ads. Some book listings include store links that earn a commission; a print-edition link is added only where the edition's details have been checked.
• Use shadow libraries or unofficial uploads as free copies.

Reporting an error

Report a problem with a course, a book or a source (a broken link, an incomplete copy, a wrong source or a content error): email corrections@redlectern.com with the page’s address and what you found.

Privacy questions go to privacy@redlectern.com.