How this site is built

BookStoic is a catalogue of the books recommended on a set of podcasts. For each book it records who recommended it, in which episode, and how many times it has come up across every show being tracked. That last number is the reason the site exists: from inside a single episode there is no way to see that the same book was raised in fourteen earlier ones.

The shows

Three to begin with, and the list will grow.

All three are hosted or co-hosted by Amit Varma. That is worth saying plainly rather than implying a breadth the catalogue does not yet have. It makes the person-level views meaningful from the first day, because one recommender spans all three shows, and it means this is at present a record of one intellectual circle. Shows outside that circle are the obvious next addition.

What counts as a book

Books, ebooks and audiobooks. Not music, film, theatre, essays, articles, threads or other podcasts.

The rule matters because the sources do not separate these. A single episode’s resource list will run a book, a YouTube clip, a newspaper column and a song together in one numbered sequence. The pilot episode of Watching the Wheels lists two songs among its resources, which is a fair test of the rule: they are not catalogued.

Two failure modes get special handling.

An anchor that reads “English” or “Hindi” is a link to a different edition of the book above it, not a book called English. Left alone, that one pattern would have put a book named “English” into the top three most recommended titles on the home page.

An Amazon link is not proof of a book. The same lists link author pages and the occasional non-book product, so the link is a signal and the bibliographic match is the decision.

How a recommendation is found

1. Every episode page is fetched once and cached, so that re-running the process costs nothing and does not go back to the source.
2. The resource list is parsed out of the page, with each link kept alongside the text around it.
3. Each link is classified as a book or not a book, with a confidence score.
4. Titles are grouped by normalising case, punctuation, leading articles and subtitles, so that one book with several surface forms becomes one record. Matching against a bibliographic database for authors, years and identifiers is the next step and has not been done yet.
5. Anything that fails to match cleanly goes to a review queue and is looked at by a person. It does not enter the catalogue on a guess.

Titles are extracted programmatically and never transcribed by hand. A title read off a screen and retyped is a title that can quietly lose a word, and the difference between two books is often one word.

What the numbers mean

One show has been fully parsed: 451 episodes of The Seen and the Unseen. From 20,589 links in those pages, the pipeline keeps 3,262 book recommendations covering 2,035 distinct books.

Most of the difference is deliberate. 886 links point at Amazon author pages or search results rather than at a book, and they are excluded by rule. 47 are links to another language or format of the book above them, which are recorded as editions of that book rather than as books of their own. Left unhandled that one pattern alone would have created a book called “English” with 21 recommendations. 11 links remain genuinely unclear and sit in a review queue waiting for a person, which is where anything doubtful goes rather than into the catalogue on a guess.

Two numbers are worth reading together. 469 books have been raised in more than one episode. The remaining 1,566, more than three quarters of the catalogue, were raised exactly once. A catalogue of this kind is mostly long tail, and a book mentioned once in passing is not the same claim as a book the show has returned to thirty one times. The site should not present them as though they were.

126 of the 451 episodes link no book at all. Almost all are early ones: the show’s notes became detailed somewhere around episode 150, and before that they rarely linked anything. The archive is heavily weighted towards its later years.

These counts are still provisional. Titles have been grouped by normalising punctuation, articles and subtitles, so “Early Indians” and its full subtitled form are one book. They have not yet been matched against a bibliographic database, which is the step that will catch the remaining duplicates and supply authors and publication years.

Identifiers, and how a book is matched

Each book carries whatever identifiers can be established for it: an ASIN, an ISBN-10, an ISBN-13, and a link to its Open Library record where one exists. They are established in a strict order of preference, because that order is what keeps the catalogue honest.

1. The ASIN comes from the link itself. Where the show notes used a short link, the redirect is followed once to recover the product identifier behind it. All 592 short links in the archive resolved this way.
2. An ISBN-10 shaped ASIN is the ISBN. Amazon uses the ISBN-10 as the ASIN for print books, so those need no lookup and cannot be wrong.
3. ISBN-13 is derived arithmetically from the ISBN-10. Also cannot be wrong, being a checksum rather than a guess.
4. Where an ISBN is known, the bibliographic lookup is by ISBN. An identifier match is exact. This matters because searching by title is the weak point for India-published books, and most of this catalogue is India-published.
5. Only where no ISBN exists is the title searched, against Open Library and then Google Books. Those results are scored against the title we hold, and a match is accepted only when it is both strong and clearly better than the runner-up.

Anything that fails those tests goes to a review queue for a person to settle. It does not enter the catalogue on a guess. A book page with no author is a book we have not confidently identified, and saying so is better than attributing someone else’s work to it.

Where a page shows an author and a year, it also says which source supplied them, so a wrong one can be traced rather than argued about.

Structured data

Every page carries schema.org markup as JSON-LD, so that search and answer engines can read the catalogue as data rather than inferring it from prose.

A book page describes the book itself, with its authors, identifiers, publication year and subjects, and then describes each episode that raised it as a podcast episode which mentions that book. That direction is deliberate. The episode refers to the book; it is not about the book, and the markup should not claim otherwise.

Two things are deliberately absent. There is no price markup and no rating markup, because this site has neither, and marking up either would be asserting something untrue in a format designed to be trusted.

Who recommended a book, and the limit of what we know

The show notes list the books an episode touched. They do not say who raised each one. That distinction matters and the site should not blur it.

What can be stated honestly: a book appeared in the notes for episode N, that episode was hosted by Amit Varma, and the guest was X. What cannot be stated from this source is that X recommended that particular book rather than the host raising it, or it coming up in passing. Attributing individual books to individual speakers would need the audio, and the transcripts are both machine-generated and currently gated.

So the catalogue records the episode and the people in it, and says “raised in an episode with X” rather than “recommended by X”. 236 guests have been identified this way, from the show’s own phrasing in its notes. Where a name could not be read confidently it is left empty rather than guessed.

What is missing

The two YouTube shows are identified and queued but not yet catalogued, which is why the totals above cover one show rather than three. Individual book pages, the written summaries, and the views that slice by person, guest, subject and year are being built. The site is published while incomplete rather than after, and it should be read that way.

Sources and what is reproduced here

The catalogue holds facts: that a book was recommended, by whom, in which episode, on what date, with a link to that episode. Facts of that kind are not anyone’s property and the shows are credited on every page that uses them.

What is not reproduced is the shows’ own writing. Episode descriptions and show-note prose belong to the people who made them. Any summary of a book on this site is written here, and every book page links back to the episodes it came from so that the source gets the visit.

Money

There are no affiliate links on this site today and nothing here earns anything.

That is expected to change. If bookshop links become affiliate links, the disclosure will appear on the page carrying them rather than buried in a policy, and the links will be marked as paid links in the markup. Any advertising, if it comes, will be labelled. Where an inbound link already carried someone else’s affiliate tag, that tag is stripped rather than passed along, because sending a reader through it would credit this site’s traffic to a third party.

Recommendation counts are computed from the sources and are never influenced by what a book earns. If that ever stops being true, this page will say so.

Corrections

A miscounted book, a wrong author, a recommendation attributed to the wrong person: these are worth reporting and they get fixed. Where a correction changes a number that was published, the change is noted rather than quietly applied.