What the fire didn’t do

A burning library is the most legible disaster in the history of writing. It has a date, a villain, and a smell; chronicles record it, survivors mourn it, and every later history can hang a chapter on it. Slow loss has none of that. When a work dies because nobody assigns it to students any more, or because the one climate that suited its parchment was also the one province that lost its monasteries, there is no day to commemorate. So the fires get the blame, and the question — a real, decidable question — is whether they earned it. Roughly a dozen conjectures in this corpus, posed against manuscript cultures from Armenia to Yucatán, are secretly one wager: put the catastrophe and the attrition side by side, in the same dataset, and see which one moved the count.

What a fire looks like when you can measure it

Catastrophe does leave fingerprints, and the first family of conjectures is about reading them precisely. The sack of Constantinople in 1204 scattered the imperial city’s libraries; the conjecture The dedication marks the loot says the disaster was selective enough to detect eight centuries later: manuscripts whose dedicatory verses name court figures — books physically concentrated in the palace quarter the Crusaders stripped — should show a sharper survival break at 1204 than equally luxurious books without such dedications. The fire, in other words, had an address.

Catastrophe also echoes in what happens to a book’s body afterwards. Hebrew manuscripts torn up for bookbinding material carry two dates — the script’s age and the dated host volume they were bound into — and The binder keeps the date of the pogrom wagers that the lag between them pulses: in the decades after an expulsion, confiscated libraries reach the binders’ knives in a datable wave. Armenian colophons — the scribes’ closing notes, which in that tradition kept recording a book’s later fortunes — log raids in a different currency: Books ransomed like captives predicts that the prices villages paid to buy their Gospel books back from raiders track the going rate for redeeming prisoners, not the price of commissioning a new copy — a seized book was priced as a person. And where a raid crushed the quiet, capital-heavy work of the scriptorium, memory switched materials: Stone booms when parchment busts wagers that Armenia’s dated cross-stones and its dated manuscripts rise and fall in opposite phase, because carving needs one mason and a month while a book needs a supply chain and a decade of peace.

Sometimes the catastrophe is not an army but a price. Khmer temple inscriptions inventory gold and silver objects that carried writing; almost none survive, because metal text is self-liquidating the moment its bullion outvalues its words. Melted libraries turns that into an experiment with a built-in control: compare attested-to-surviving ratios for inscribed metal against inscribed stone from the same temples, and you have measured how much of a literature the melting-pot ate.

The filter that doesn’t smoke

Then comes the other side of the ledger. The bluntest statement of it is The fire is a rounding error: the history of Hebrew books is told as a history of burnings — Paris 1242 above all — yet the conjecture wagers that when surviving manuscripts are counted by century of production, the curve is dominated by ordinary exponential attrition, the constant-rate decay that eats every manuscript culture, and the famous bonfires barely dent it. If that is right for the most persecution-marked book culture in Europe, it is a caution for every other.

What does the quiet filter look like in detail? It looks like a curriculum. Yijing, a Chinese pilgrim who studied at the North Indian monastery-university of Nālandā in the 680s, wrote down the Sanskrit grammar syllabus as a ladder from beginners’ primer to doctoral treatise; The syllabus was a survival ranking wagers that manuscript survival a millennium later follows enrolment — every student needed the primer, few needed the summit — so the ladder’s bottom outlives its top by an order of magnitude. It looks like an institution’s pulse: Survival follows the abbot, not the scribe wagers that what remains of medieval Ethiopia’s books depends less on where they were written than on which monasteries stayed continuously inhabited into the twentieth century — continuity of custody beating quality of production. And it looks like a climate: Nepal is everyone’s offsite backup wagers that the cool, politically continuous Kathmandu Valley holds the oldest surviving witnesses of works composed all over the Sanskrit world, because everywhere else the recopying treadmill ran hotter than the scribes could tread.

The filter has one more trick, stranger than either fire or neglect: sometimes the only door out is through the enemy. Kumārila Bhaṭṭa’s vaster verse-treatise, the Bṛhaṭṭīkā, was let die by his own school — and survives, in fragments, inside the Buddhist treatise written to refute it. The enemy kept the better book generalizes the pattern: preservation-by-refutation, a channel that selects precisely the passages an opponent found most worth attacking.

Counting what burned without a body count

If the two mechanisms compete, someone has to count the dead — and a cluster of conjectures imports the counting method from ecology. Mark-recapture — the technique by which ecologists estimate a fish population from the overlap between two independent nettings — works on books wherever two independent samples of one written culture survive. Mark and Recapture in the Ashes treats the four surviving Maya codices as one sample of the script and the thousands of stone monuments as the other, and estimates from their overlap how many signs the writing system once carried that we have never seen. The Width of the Filter That Ate the Library measures the survival bottleneck itself: all four codices are divinatory almanacs, and if that reflects the filter rather than the library, the codices’ sign-vocabulary should overlap the monuments’ less than any two monument sites overlap each other. Two Doors Out of the Fire adds the check: the three codices that survived through European collectors passed a different filter than the one that survived in a dry cave, and the two channels’ disagreements measure the collectors’ taste. The same arithmetic runs on a living library: Counting the uncounted overlaps two partly independent digitization campaigns in the Timbuktu manuscript world to estimate how many works exist that neither has touched.

The ledger, not the verdict

This essay has not told you which side wins, because nobody knows yet — and that is the point of the shelf. Every conjecture above names, in advance, its kill dataset — the public catalogue, census, or corpus whose numbers would prove it wrong: dated Armenian colophons, the binding-fragment surveys, the Hebrew manuscript censuses, the Khmer inscription corpus, Nālandā-curriculum survival counts, Ethiopian microfilm catalogues, the Maya sign inventories. None of these studies has been run. Each is an afternoon or a season of work for the right specialist, and each settles something the humanities have mostly argued by anecdote: how much of the missing written world did catastrophe actually take, and how much quietly walked out through the filter while we were watching the flames? If you can run one — run it. The verdict is credited, by name, on the conjecture’s page as it comes in.

Written by Claude (Fable 5). The conjectures linked here are pre-registered wagers, not findings; each page carries its own evidence ledger and the dataset that would kill it.