The Missing Ancient Works

Almost everything written in the ancient world is gone, and a great deal of what survives has never been read.

1. The numbers

• Menander wrote about 108 plays. One survives complete, published from papyrus in 1952.
• Livy wrote 142 books of history. Thirty-five survive; the rest exist as one-paragraph summaries made by someone else.
• Aristotle’s published works: none. What we have is lecture notes.
• Nicolaus of Damascus wrote 144 books. Fragments.
• About 1,800 carbonised scrolls survive from one villa at Herculaneum. The first words were read in 2023.
• The British Museum holds around eight million objects.
• The Endangered Archives Programme has digitised over sixteen million images since 2004. The high-resolution files are not released.
• Sixty-two per cent: how often Ithaca, the best-published tool for restoring damaged Greek inscriptions, is right unaided.

2. What was lost, and how

The counts below are of works we know existed, because a surviving author names one, or a library catalogued it, or a later writer quotes it. They are not estimates of what was written.

Greece

Greek loss can be counted because Alexandria catalogued and later writers cross-referenced. The surviving tragedies are a school selection: seven each of Aeschylus and Sophocles because a late-antique syllabus settled on seven. Nine of Euripides survive for a different reason — one volume of an alphabetical set came through, which is why we have his plays with titles from E to K.

Rome

Latin loss is countable for the same reason and the proportions are worse. Varro was called the most learned of the Romans by Quintilian. Nicolaus was Herod's court historian and Josephus's source; everything in The Herod Question passes through those lost 144 books.

Egypt

Egypt cannot be counted. Manetho names sources no one else records, and the temple libraries are gone entirely — in some cases the catalogue is carved on the wall and the books are not there. The figures above are not a census. They are the few named items.

Mesopotamia

Mesopotamia inverts the problem. Clay survives where papyrus does not, so the tablets exist in enormous numbers and are unread rather than lost. The constraint is publication, not preservation.

China

China lost by decree and then rebuilt by canonisation. The Qin burning of 213 BC is reported by Sima Qian, writing under the dynasty that replaced the burners; the scale is disputed. What is not disputed is that the pre-Qin record survives thinly.

India

India lost by medium. Birch bark and palm leaf do not survive a monsoon climate, and no ancient manuscript of the Vedas exists — the text came down by memorised recitation for a millennium before it was written. The oral channel carried what the physical one could not.

Set the six side by side and the mechanisms differ completely. Greece and Rome lost by attrition and can be counted. Egypt lost the books and kept the catalogue. Mesopotamia lost nothing and reads almost none of it. China lost by decree. India never had physical copies to lose.

3. What was never named

Every figure above is a floor. A work enters the count only because somebody who survived happened to mention it — quoted it, catalogued it, or argued with it.

A book nobody engaged with leaves no trace at all. Papyrus finds regularly produce authors no surviving source names, which shows the category is real and gives no way to size it.

The true loss is not measurable. Nothing in this memo should be read as a total.

4. Why it went

The Library of Alexandria is the story everyone knows and it explains almost nothing. Caesar’s fire in 48 BC, Aurelian, the Serapeum in 391 — each did damage, and none of them accounts for the pattern of what we have.

Papyrus lasts a few centuries in Mediterranean conditions. A text stayed in existence only while somebody kept making new copies of it by hand, at real cost. Three filters decided which:

• The syllabus. Schools taught a selection; the selection got copied.
• The codex. Between the second and fourth centuries AD the scroll gave way to the bound book, and the transfer was made work by work.
• Language. When Greek stopped being read in the Latin west, Greek texts stopped being copied there.

The third filter is the one with a recovery. Greek survived at Constantinople and in Arabic translation at Baghdad, and returned to Europe through Toledo in the twelfth century. The second timeline carries that relay.

5. The conflicts

Where sources survive they frequently disagree, and the disagreements are more informative than the agreements.

Shalmaneser III’s Kurkh Monolith records Ahab at Qarqar in 853 BC with 2,000 chariots. 1 Kings gives Ahab three chapters and never mentions the battle. The Black Obelisk shows Jehu prostrate before the Assyrian king; 2 Kings covers his reign and omits it.

At Kadesh in 1274 BC both Ramesses II and the Hittites recorded a victory. Hatshepsut and Akhenaten were struck from the Egyptian king-lists by their successors. Sima Qian reports the Qin book burning while writing under the dynasty that replaced the burners.

Two chronologies for the same events remain in use: Ussher’s dates the division of the Israelite kingdom to 975 BC, Thiele’s to 931. This site uses Thiele, because the Assyrian eponym lists are fixed by a solar eclipse in 763 BC and Ussher’s reckoning cannot be reconciled with it.

Where two sources disagree there are two sources. Where they agree there may be one, copied twice.

6. What exists and has not been read

• Oxyrhynchus. About 500,000 fragments from an Egyptian rubbish dump, excavated from 1896. Roughly 5,000 published.
• Herculaneum. Some 1,800 carbonised scrolls, first passages read in 2023. The villa is not fully excavated.
• Cuneiform. About 500,000 tablets excavated, roughly nine in ten unpublished.
• The Cairo Geniza. Around 400,000 fragments from one synagogue storeroom.
• Timbuktu. Hundreds of thousands of manuscripts in private family libraries, largely uncatalogued.
• Sakya, Tibet. A sealed wall found in 2003 holding a reported 84,000 texts. Indexed since; over twenty per cent digitised by 2022.
• Palimpsests. Erased parchment reused for other text. The number is unknown because nobody has imaged them.

None of this is hidden. It is uncatalogued, unfunded, and in most cases nobody’s job.

7. Where the record is thin

Every entry on this site carries a grade for what attests it. Aggregating those grades across the fourteen registers produces a shape nobody designed.

8. The tools

The reading problem is now partly a computing problem, and the change is recent.

The Vesuvius Challenge released the Herculaneum scan data publicly, stated the problem, and attached prize money. The first passages were read within months of launch. The structure is the finding: open data and a reward, not a committee.

Ithaca restores missing text in damaged Greek inscriptions. It is right about 62 per cent of the time alone and around 72 per cent when a historian works with it. Neural translation from Akkadian was demonstrated in 2023. Multispectral imaging reads erased under-text; the Archimedes Palimpsest returned two treatises that existed in no other copy.

A machine restoration is a plausible reading, not an attested one. Sixty-two per cent likely is a hypothesis with a number on it, and belongs in a register at the lowest grade, marked as restored.

9. How Ptolemy did it

There is a precedent for solving this at scale, and it is the dynasty this site already covers.

Ships putting into Alexandria were searched for books. What was found was copied, and by the account that survives the library kept the original and returned the copy, the seized texts marked in the catalogue as from the ships. Ptolemy III is said to have borrowed the official Athenian state texts of the tragedians against a deposit of fifteen talents, then kept them and forfeited the money.

Translation was paid for out of the treasury. The Septuagint was commissioned. Manetho wrote Egypt’s history in Greek under royal patronage and Berossus did the same for Babylon — which is why we have any of either.

The dynasty lasted three centuries. The institution lasted seven. Claudius Ptolemaeus was still working at Alexandria in the second century AD under Roman rule, and the astronomy he wrote there stood for fourteen hundred years, preserved in Arabic at Baghdad and returned to Europe through Toledo.

The methods do not transfer. The decision does: a state judged that assembling the written record was worth serious public money, and spent it. That judgement has not been made since. The Ptolemy Project sets out what it would look like now.

10. Sources

Reference codes read source.work.passage and resolve on the Master Source Register, which carries every source used across this series.

1. Survival counts, Greek and Latin — the standard figures given in the Oxford Classical Dictionary and in the introductions to the Loeb and Oxford Classical Texts editions of each author. The counts derive from the Suda, Quintilian, Diogenes Laertius, Photius and the authors’ own cross-references. Where ancient catalogues disagree the figure is given as approximate.
2. Lost work: Varro, Antiquitates rerum humanarum et divinarum — 41 books on Roman religion and institutions, lost. It is the principal source for Augustine’s account of Roman religion in De civitate Dei, which is how we know what it contained. Supports §2.
3. Lost work: Nicolaus of Damascus, Histories, 144 books — Herod’s court historian, and Josephus’s source for the reign. Survives in fragments and in Josephus’s use of it. Supports §2 and the channel argument in The Herod Question.
4. Lost work: Manetho, Aegyptiaca — Egypt’s dynastic history, written in Greek for the Ptolemies. Lost as a book; survives in quotation by Josephus and the Christian chronographers, who cited it to argue against it. The dynastic numbering still in universal use is his. Supports §2 and §9.
5. Lost work: Berossus, Babyloniaca — Babylon’s history in Greek, same period and same patronage, same fate. Supports §2.
6. Oxyrhynchus — excavation and publication figures from the Egypt Exploration Society’s Oxyrhynchus Papyri series. The half-million counts fragments, not texts; many fragments belong to one document. Supports §1 and §6.
7. Herculaneum — scroll counts from the Officina dei Papiri Ercolanesi, Naples. The 2023 readings are the Vesuvius Challenge results. Supports §1, §6 and §8.
8. Cuneiform — excavated totals and publication rates from the Cuneiform Digital Library Initiative. Supports §1, §2 and §6.
9. Ithaca — Assael, Sommerschield et al., Nature 603 (March 2022). The 62 and 72 per cent figures are the paper’s own, measured on its own test set. Supports §1 and §8.
10. Akkadian translation — Gutherz et al., 2023. Supports §8.
11. Endangered Archives Programme — British Library, funded by Arcadia. Project counts, image totals and the £60,000 project cap are the programme’s own published figures. Supports the callout and §6.
12. Sakya — the 84,000 figure traces to a single Chinese news report of 2003 and has not been independently confirmed. The indexing and digitisation progress is reported by the Tibetan Academy of Social Sciences. Claims circulating online that the collection records ten thousand years of history are false; the monastery was founded in 1073. Supports §6.
13. Ptolemaic acquisition — the searching of ships, the catalogue mark, and the forfeited fifteen-talent deposit all come from Galen’s commentary on Hippocrates’ Epidemics III, written some four centuries after the events. Single channel, and a late one. Supports §9.
14. Qarqar and the Black Obelisk — Kurkh Monolith, BM 118884; Black Obelisk, BM 118885. Both carry the entries on The Line from Noah to Jesus. Supports §5.
15. Chronology — Edwin R. Thiele, The Mysterious Numbers of the Hebrew Kings, against Ussher’s Annales Veteris Testamenti (1650). Supports §5.
16. Source, not reviewed for this memo: L. D. Reynolds and N. G. Wilson, Scribes and Scholars: A Guide to the Transmission of Greek and Latin Literature, 4th edn (Oxford, 2013), ISBN 978-0-19-968633-6 — the standard account of transmission, which bears directly on §4.
17. Verification register. Carried as reported and not settled: the Oxyrhynchus fragment total; the cuneiform unpublished proportion; the Timbuktu manuscript count; the number of Menander’s plays; the Sakya figure; the Ptolemaic acquisition stories; and whether a second library remains at Herculaneum, which is asserted as possible and is not established. The Egyptian, Mesopotamian, Chinese and Indian counts in §2 are named items, not censuses, and are softer than the Greek and Latin figures.