Method · every figure on this page computed at build time
How this was made, and how far to trust it.
01
What this is, and who it is for
A reading companion. Public-domain translations of Greek and Latin sources, gathered into collections around subjects rather than authors, with a note beside each passage supplying what that passage assumes you already know.
It is built for the reader who never had a classical education and wants to know what these minds actually said — and for anyone who has tried reading a primary source with a search engine open in the next tab, stopping every paragraph to check they were getting it. The notes are that checking, done in advance and put in the margin. What the site is not is an edition, and it should not turn up in a bibliography: cite the translation, which is a real book.
02
What the notes are for
In order of what they are trying to do:
- Make the passage legible. What it assumes you already know — the institution, the office, the coin, the custom, the person named without introduction — and how serious it is. A modern reader defaults to reading old prose as solemn, and frequently it was not.
- Explain the social logic. What rule is in play, and what breaking it actually cost someone.
- Then, and only then, the translation. Where the translator softened, elevated, moralised or flattened a joke, and what the original has instead. This is worth having and it is not the point.
That order used to be the other way round, and the notes were correspondingly more interested in the Greek than in the reader.
03
Where the notes come from
A language model was given the passage, the Greek or Latin where a public-domain edition exists, and hand-written records for the places, people and terms in it. It wrote every one of the 11,399 notes on this site. No specialist has read them and no sample has been formally scored, so the honest figure for the error rate is that there isn't one yet — and getting one is work this project is looking for a collaborator on.
What exists in the meantime: automatic gates that check the model against the record wherever a record exists. They have caught real fabrications, and the table below says which. The practical advice is the same as for any commentary you did not write yourself — a note is a lead, and the citation goes to whatever it leads you to.
04
Which translations, and why
Every translation is checked against Wikidata for publication date and translator death date before the work is published. Anything that fails goes back to an older public-domain version, which is why some texts here are read in a Victorian translation rather than the standard modern one. Older translations are more decorous and less accurate; catching some of that is part of what the notes are for, and the library page always says which edition you are reading.
05
The maps
104 places, each matched to a Pleiades identifier by hand and plotted in the order the work names them. Automatic resolution was tried and abandoned: it produced Sidon as a road in Jordan, Salmone as a village in the Peloponnese rather than the cape in Crete, Carthage as Cartagena in Spain. Not one was a subtle ambiguity; every one was the right name in the wrong part of the map.
Named is not the same as visited, and a place that cannot be verified gets no dot rather than a guessed one — which is why Lucian's itinerary runs off the edge after a single real harbour. Scene locations, the "where is this happening" pins, are the model's judgement and each carries a precision marker.
06
Why you will see so few quotations
Ancient authors quote constantly and usually without announcing it. Spotting a quotation is one of the most valuable things a commentary can do — and the single easiest thing for a language model to invent, fluently and unfalsifiably, for a reader who has no Greek.
So the model is never allowed to propose one. It may only explain a quotation it has been handed, from a list assembled outside it and checked by a person. If it believes it sees one that was not supplied, it may not tell you: there is no channel for a suspicion and no editor waiting to triage it. That rule is why this site shows 1 quotations rather than fifty, and it is not going to change.
07
Corrections
Every page has a correction link, and it is not a courtesy. A site whose whole posture is "this might be wrong" owes you somewhere to say so; without that, the admission costs nothing and changes nothing.
The gates
Every one was written after something got throughEach exits non-zero on failure. That is the honest way to read a green board: it means "no known failure", never "correct".
| Gate | When | What it caught |
|---|---|---|
| Rights | before spend | Two works had no public-domain evidence at all, only an inference from a date. |
| Text hygiene | before spend | One work shipped with a printed running head mid-sentence and 38 marginal page numbers; another arrived with 515 passages of gazetteer metadata flattened into the prose. Both invisible in the output, because a model given corrupted text writes fluent commentary about the corruption. |
| Prompt leaks | after run | The prompt announced the wrong translator to every work. 123 passages were annotated under a false statement about what they were reading. |
| Citations | after run | The model sharpened a vague reference into a precise one that nothing had verified. |
| Quoted source text | after run | Greek quoted inside the model's own prose was never checked against anything. It found 36 reconstructions in one work — which is why that work is not published — and, in another, an invented editorial history: a claim that nineteenth-century editors had emended a word, naming a replacement that appears nowhere in the text. |
The numbers, per work
| Work | Passages | Notes | Scene pins | Places | Quotations |
|---|---|---|---|---|---|
| The Odyssey — ten years to get home, and a house full of men eating the estate | 288 | 3351 | 183 | 11 | 0 |
| Histories, Book 2 — the Egyptian inquiry | 205 | 1571 | 75 | 7 | 0 |
| Anabasis, Book 4 — the march to the sea | 77 | 414 | 77 | 4 | 1 |
| Clouds — the play Socrates was still answering at his trial | 68 | 645 | 60 | 0 | 0 |
| De Officiis 1 - what a decent man may do for money | 54 | 488 | 4 | 1 | 0 |
| The Ship, or The Wishes | 46 | 229 | 7 | 8 | 0 |
| The Life — the Galilee command | 38 | 336 | 34 | 7 | 0 |
| On Architecture, Book 10 — machines, and how they multiply a man | 37 | 280 | 6 | 0 | 0 |
| The training of a wife — Oeconomicus 7-10 | 33 | 227 | 31 | 0 | 0 |
| Poroi — ways and means to raise Athens' revenue | 33 | 246 | 14 | 5 | 0 |
| Ab Urbe Condita 34 - repealing the Oppian Law | 31 | 382 | 31 | 1 | 0 |
| Alexander the False Prophet — a new god with a puppet snake, and the Epicureans who called it | 31 | 329 | 26 | 0 | 0 |
| That Epicurus Actually Makes a Pleasant Life Impossible | 31 | 313 | 2 | 0 | 0 |
| A True Story | 30 | 186 | 1 | 1 | 0 |
| Geography, Book 17 — Egypt and North Africa | 28 | 232 | 28 | 6 | 0 |
| For Marcus Caelius — the trial that became a character assassination | 27 | 292 | 26 | 0 | 0 |
| The Apology of Socrates — a defence that refuses to plead | 26 | 252 | 26 | 6 | 0 |
| Annals 6 — the last years of Tiberius, and the credit crisis of 33 | 26 | 358 | 25 | 4 | 0 |
| The Periplus of the Erythraean Sea | 22 | 206 | 22 | 13 | 0 |
| Solon — the man who cancelled the debts | 20 | 194 | 20 | 0 | 0 |
| Letter to Herodotus — the physics, compressed for memorising | 17 | 107 | 0 | 0 | 0 |
| Indica 18–43 — the voyage of Nearchus | 13 | 156 | 13 | 0 | 0 |
| Airs, Waters, Places — how where you live decides what kills you | 12 | 94 | 5 | 0 | 0 |
| The Plague at Athens — Thucydides 2.47-54 | 12 | 73 | 12 | 0 | 0 |
| Acts 27 — the voyage to Rome and the shipwreck | 11 | 76 | 11 | 14 | 0 |
| The Life of Epicurus — the slanders, the will, and a letter written while dying | 10 | 142 | 4 | 0 | 0 |
| The Principal Doctrines — forty sentences meant to be known by heart | 8 | 49 | 0 | 0 | 0 |
| On Funerals — the business of mourning | 8 | 70 | 0 | 0 | 0 |
| Letter to Menoeceus — death is nothing to us, and pleasure is not what you think | 6 | 40 | 0 | 0 | 0 |
| Satires 1.5 — the journey to Brundisium | 3 | 61 | 3 | 16 | 0 |
3 work(s) in the repository are held back from publication and are not counted above: petronius-cena, suetonius-vespasian, synesius-ep4. A work whose rights or text cannot be verified does not ship.
The one thing engineering cannot supply here is a specialist's judgement: how good are these notes, actually, and where do they go wrong in ways no gate can see? That is a measurable question rather than a rhetorical one, and I would much rather measure it than assert an answer to it. There is a corpus, a fixed method, a complete failure log — and, I think, a paper or two in the gap.