Two words: "The author".
They open 5,642 of the answers on my flashcard site. Add "The chapter" and you are at 22.5% of 34,500.
I found that by parsing my own database, which sounds more impressive than it was. The hard part was admitting I had shipped 156 decks without ever reading them end to end.
The parse
Each deck page ships its whole card array inline as window.BC_CARDS — the viewer needs to flip through cards without hitting the server. So the data was already public. I pulled 157 sitemap URLs with curl, matched one regex per file, and loaded each array with json.loads.
156 decks, 34,500 cards, no malformed payloads.
| Measure | Value |
|---|---|
| Decks parsed | 156 |
| Cards | 34,500 |
| Unique normalized questions | 34,484 |
| Questions in 2+ decks | 9 |
| Exact duplicate answers | 0 |
| Mean answer length | 60.58 words (sd 8.29) |
| Answers 50–75 words | 88.0% |
| Answers over 90 words | 50 |
| Chapters | 9,286 |
| Chapters with exactly 4 cards | 7,297 (78.6%) |
The duplicate hunt was a bust
I went in expecting mass duplication. Same generator, 156 books, one prompt — surely the same cards came out under different titles.
Nine questions out of 34,484 appear in more than one deck. Zero answers repeat verbatim.
An earlier note of mine called duplication "the main structural problem". That note was a guess dressed up as a finding. Consider it withdrawn.
What the uniformity actually is
Three things, none of them duplication.
Openers. 5,642 answers begin "The author". 2,109 begin "The chapter". A quarter of the database opens with words that carry no information — the question already told the reader whose argument it is. Worse, 132 of 156 decks mix styles: "Mill argues..." on one card, "The author argues..." three cards down, same book, no rule.
Length. sd 8.29 around a mean of 60.58. 88% land in a 26-word window between 50 and 75. Fifty cards out of 34,500 get past 90 words; 381 come in under 40. A one-line fact and a four-step argument both arrive at sixty words, which means one is padded and the other is cut.
Chapter caps. 7,297 of 9,286 chapters hold exactly four cards. Only 154 hold one. Four is the ceiling and also the target, so most chapters carry a fourth card that exists to fill a slot.
And the ref field — the pointer back to the book — is just "Chapter 7" in 99.9% of cards. No page, no paragraph. Useless for finding the passage.
Limits
I have not read the 156 source books, so I cannot tell you the answers are correct. I can tell you they are unique and consistently formatted. That is not the same claim.
Whether any of this costs a student recall is a user question. I did not open user data for this post.
Disclosure
WakeCards is mine — wakecards.com, public-domain titles, five cards free, no account. The numbers above come from my own database and I am publishing them because nobody else was going to.
The regex took ten minutes. Not looking for three months is the part I would rather not put a number on.