← SkillSafe / Book Summary

Name a book. Get your bearings.

An orientation to any book, with the facts a model invents most readily checked against a retrieved source, confirmed or unverified.

The book

As printed, if you know it. A translated title is fine.

Optional, and it does more than it looks like: it is what makes the lookup land on one article instead of a disambiguation page.

Poems, songs and short stories are handled more strictly — see the note by the button.

What you want from it

Changes what the brief leads with more than anything else here.

Grounding

Retrieves a source before the brief is written. Needs an account; it is the cheapest part of a run.

Name a book to price the run.

Each example opens a whole saved brief — the marks, the retrieved abstracts, every check this page ran — free, signed out, nothing spent.

Why the facts carry marks, and how the check works

Books are the classic place a language model confabulates: plausible titles, wrong authors, editions that never existed, all delivered with total confidence. So the five facts most likely to be wrong and most likely to be repeated — author, year, original language, form, series — are checked rather than trusted. The page retrieves a source, hands it to the model, and then re-reads the source to see whether it actually contains what the brief claims. A fact confirmed that way is marked confirmed. A fact the model recalled on its own is marked unverified, which is honest and often fine. And a fact cited to a source whose text does not support it is flagged loudly, as falsely cited — because a citation pointing at a real page is not evidence for a claim the page does not make, and that failure is worse than no citation at all, not better.

  • Grounding runs a Wikipedia search for the title and author you typed and hands the matching abstract to the brief. It is what turns “the model thinks the author is X” into “the retrieved text says the author is X”. What each retrieved snippet literally states is printed alongside the brief, so you can see for yourself what the grounding does and does not carry.
  • Every claim is reported with what backs it. A claim cited to a source that does not support it is escalated; one the model volunteers as its own recollection is downgraded to unverified, and that admission is on the page rather than in a footnote.
  • A book that cannot be identified is reported as not found rather than invented, with any similar-titled works listed so you can tell a typo from a book that does not exist.
  • The fact ledger exports as Markdown, JSON or CSV, recording every claimed fact and what was found to support it. Briefs are saved to your account and searchable by semantic similarity, not just by title.
What this app will and will not do

What it is for

  • Summary and synopsis. What the book is, what happens in it, what it argues. Asking for a summary so you do not have to read it is a normal request and this app answers it properly.
  • Analysis. It analyses themes, structure, the shape of the argument, how the prose works, what the book is trying to do and whether it manages it, and reports the thematic reading behind each.
  • Context. It contextualises the book: the history around it and the historical moment it came out of, the biographical life behind it, how it reached publication, and its reception then and since, including the criticism around it.
  • Guidance. A read-it-if and skip-it-if control on whether to read it, a difficulty report, where to start, what it asks of you, and what to read next and why.

What it will not do, whatever you type

  • Reproduce the text. No passages, no chapters, no “write out the opening”, no lyrics. Those are the book, and handing them over is not summarising. It summarises a work; it never reproduces one.
  • Retell it densely enough to replace it. A structural walkthrough says what each stretch of the book is for. This page measures the density of that breakdown and says so when the thickening starts to read as a retelling.
  • Quote past a budget. At most three extracts, 25 words each, 50 words altogether, each attributed and each carrying a stated point. Quoted runs hidden inside the commentary are measured and counted too, and quotation is capped by the same fields either way.
  • Quote poetry or lyrics at all. A poem is short enough that any excerpt is a substantial part of the whole work, so quotation there is forbidden entirely. Those get analysis and no reproduction.

A request for the book's own text is refused before the run starts and before a credit is spent, so you are never charged for a refusal. The in-scope version of what you asked is offered in one click, and such requests are turned into answerable ones rather than simply denied.

What a brief covers

What it is, what it argues, how it is built, the themes running through it, the world it came out of, who it is for and what to read next. Useful before you start, after you finish, or while you are deciding whether to bother.