How We Choose
From thousands of listings to five
What we don't do
We don't buy anything automatically, and no listing is ever purchased or reserved on a reader's behalf. We don't work for or with any bookseller. We don't assert rarity, value, or authenticity — any concerns about a book's condition or completeness are flagged directly, in plain language, rather than implied by a score. We don't publish an issue without a human reviewing every find in it first. Nothing here is investment advice — verify condition and completeness yourself before buying.
Discovery
We watch listings across marketplaces including eBay, AbeBooks, and Catawiki — through saved searches, dealer alerts, and plain old browsing — looking for the kind of thing that's easy for a general audience to scroll past: a miscatalogued edition, an under-described binding, a book whose significance isn't obvious from the listing photo. Every candidate that makes it into our review process was found and put there by a person, on purpose. We look for diamonds in the rough and sort them into five categories:
[ THE BARGAIN ]
A genuinely good book at a price that doesn't reflect it. Nothing wrong with it — the market just hasn't caught up.
[ THE SLEEPER ]
Historically or culturally significant, but overlooked. The kind of book that shaped its field and nobody's bidding on.
[ THE BEAUTIFUL ONE ]
Chosen for the object itself: binding, illustration, plates, condition. A book you'd want on a shelf before reading a word of it.
[ THE WEIRD ONE ]
The delightfully odd find. Strange subject matter, an unusual format, or a book that simply shouldn't exist and does.
[ SOMEBODY SHOULD BUY THIS ]
A wildcard. Not always the best value or the most beautiful — just something we genuinely think deserves a home.
Automated first pass
Before anything else happens, a set of simple, deterministic rules clears out listings that couldn't plausibly be a real candidate — obvious reprints and facsimiles, modern print-on-demand editions, listings missing a price or photos entirely. This is housekeeping, not judgment: it never weighs in on subject matter or language, only on whether a listing is viable at all.
AI-assisted triage
Once a candidate is submitted, it goes through AI-assisted triage and write-up. Each one is read for six things: how interesting it is historically or culturally, how scarce the edition seems, whether the asking price looks anomalous next to comparable copies, how appealing the object itself is, how confident we are that the listing is correctly identified, and how much risk it carries — missing plates, an unclear edition, an ambiguous seller description. This is a sorting aid, not an appraisal, and it narrows the field to a shortlist a curator can reasonably review by hand — nothing more.
Where a write-up adds a specific fact beyond what the seller's own listing says — an edition detail, a collation point, a comparison to another copy — we check it against a real source before including it. If we can't verify something, we say so plainly instead of guessing, or we leave it out.
Human final call
A human curator reviews the shortlist and chooses exactly five books for the issue — one for each recurring category. The triage narrows the field; it does not choose it. Every write-up, every price comparison, and every choice of what makes an issue is made and reviewed by a person before it publishes.