The Library

Live counts as of 19 September 2026: 2,274,296 documents / 106,902,485 passages across 159 registered, synced sources (the full registry counts 187 rows: 159 corpus sources, 5 local shelves, and 23 feature modules), plus 1,998,589 dictionary entries on the reference shelf and 21.9 million gold lemma annotations in 45 languages (a further 35,945,917 lemma rows in 15 languages ride an honestly labelled silver tier — machine-suggested or upstream-undisambiguated, excluded from gold-only search). Classical Arabic is the largest language in the catalog at 33.3 million passages, ahead of early-modern English (24.3 million — the EEBO-TCP corpus) and Literary Chinese (18.7 million), with Latin, Sumerian and Ancient Greek next (about 3 million each); 167 language codes appear in all, and since August 2026 stored codes also resolve through the lect layer — historical-stage ladders on the language cards and stage-scoped search, described on the Languages page. The authoritative shelf map, refreshed at every development gate, is docs/library.md in the repository; the numbers below are taken from it, and were read from the live catalog, not estimated.

The same shelves, grouped by the owner’s research desks rather than by collection — each with its instruments, CLI recipes and terminal setup — are the research axes.

Classical Greek literature

The backbone of the Greek canon, from the Perseus Digital Library’s canonical-greekLit collection: Homer, Hesiod, the tragedians, Aristophanes, Herodotus, Thucydides, Plato, Aristotle, the orators, Plutarch, Galen, and more — 768 Greek editions spanning the Archaic period to roughly the third century CE, of which 650 carry an aligned English translation displayable side by side. Citations follow the canonical schemes (book and line, Stephanus pages) under stable CTS URNs, so a range such as Iliad 1.1–1.32 resolves natively.

Post-classical Greek

The First Thousand Years of Greek project (Open Greek & Latin) supplies the long tail Perseus does not cover: Athenaeus, Philo, the church fathers, grammarians, scholia, minor historians, and medical and mathematical writers — 1,088 Greek editions, mostly Hellenistic through Late Antique. English coverage is thin (41 editions) but present. Swete’s edition of the Septuagint lives on this shelf and serves as the Greek witness of the Old Testament alignment axis.

The Greek silver layer

Two machine-annotated editions of the Greek canon sit deliberately unmerged beside the Perseus and First1K copies of the same works, and every count they contribute is labelled silver — machine-produced analysis, never presented as gold attestation, and excluded from gold-only search. The Diorisis Ancient Greek Corpus (Vatri & McGillivray, CC BY-SA) supplies 764 documents / 502,865 passages — roughly 10.2 million words tokenized, lemmatized, and morphologically analyzed, Homer to early Byzantium. GLAUx (“the Greek Language Automated”, Keersmaekers, Leuven; joined 25 July 2026) adds 1,421 documents / 968,578 passages — about 20 million tokens automatically annotated for morphology, syntax, lemmas, animacy, and word senses. Together they make lemma search possible across the whole Greek canon where the gold treebanks cover only samples.

Classical Latin literature

Vergil, Ovid, Horace, Cicero, Caesar, Livy, Tacitus, and Seneca, from Perseus canonical-latinLit — 353 Latin editions with 181 aligned English translations. Orthographic variation (u/v, i/j) is folded at the search layer, so iuvenis, juvenis, and iuuenis all resolve to the same word.

The Latin–Romance continuum

Since late July 2026 the Latin shelf continues past antiquity as one readable band. digilibLT (Biblioteca digitale di testi latini tardoantichi, Vercelli, CC BY-SA) covers late-antique secular Latin prose of the second through seventh centuries — Ammianus, the grammarians, the agrimensores, the jurists — in 372 documents / 459,451 passages. openMGH (the digital Monumenta Germaniae Historica, CC BY) brings the medieval-Latin backbone: all 57 volumes of the SS rer. Germ. series — Einhard, Widukind, Liudprand, Regino, Adam of Bremen, Otto of Freising — with more volumes published upstream successively. CroALa (Croatiae auctores Latini, CC BY) adds 570 documents / 309,180 passages of Croatian Latinity across a near-millennium, 454 of them dated (999–1984). On the vernacular side, the Base de français médiéval (ENS de Lyon) holds 219 complete medieval French texts / 333,819 passages — roughly 6.45 million words from the Serments de Strasbourg of 842 through the end of the fifteenth century, every text dated. And the Cantigas Medievais Galego-Portuguesas (Projeto Littera, held under a written grant from the project’s coordinator; joined 31 July 2026) supply the complete secular Galician-Portuguese lyric corpus — 1,676 cantigas de amigo, de amor, and de escárnio e maldizer / 34,058 verse passages, with author, genre, manuscript sigla, and rubrics carried as metadata. On the reference shelf, DÉRom — the Dictionnaire Étymologique Roman — reconstructs the Proto-Romance side of the same band (see below).

Documentary papyri

The everyday written record of a millennium of Egypt, from the Duke Databank of Documentary Papyri via papyri.info: contracts, tax receipts, petitions, private letters, census returns, leases, and court records — 61,414 documents from Ptolemaic to early Islamic Egypt, chiefly Greek with Coptic, Latin, Arabic, and Demotic minorities. Leiden-convention editorial markup (restorations, cancellations) is preserved, and the fragment-search index (see Tools) is built over this shelf.

Latin inscriptions

The Epigraphic Database Heidelberg, ingested on 13 July 2026 as a preservation snapshot (the upstream project was archived in 2021): 81,881 inscriptions — epitaphs, dedications, honorific and building inscriptions, milestones — in Latin (80,561), Greek (1,290, the bilinguals), from provinces across the whole empire. 81,416 documents carry dates on the chronological axis, and the shelf brings the library’s first genre facets: searches can be filtered by inscription genre, province, material, and object type, composing with the date and place filters. Leiden-convention rendering and the fragment-search index apply here as on the papyri.

Since 26 July 2026 the Epigraphic Database Roma (Sapienza) completes the record geographically: 115,590 inscriptions of Italy itself — the peninsula Heidelberg’s provinces surround — in 596,064 passages, every document dated and faceted (genre, material, object type, and a fifteen-value region axis), released under CC BY 4.0.

Coptic

The Coptic Scriptorium corpora, synchronized on 13 July 2026 with 482 of the 483 upstream corpora aboard: the complete Sahidic and Bohairic New Testaments, monastic and patristic prose, and hagiography — 74,169 passages with upstream gold lemmatization, making Coptic the library’s fifteenth lemma-searchable language (233,020 lemma rows). The two New Testament dialects join the alignment hub as witnesses fourteen and fifteen, and the corpus’s language-of-origin annotations make the Greek loan stratum of Coptic queryable.

Celtic

The Celtic language axis, synchronized on 17 July 2026, in three sources. CorPH — the Corpus PalaeoHibernicum of the ERC ChronHib project at Maynooth — supplies 76 Early Irish documents of the seventh through tenth centuries (the Annals of Ulster, Adomnán’s Vita Columbae, the poems of Blathmac, and the Milan, St Gall, and Würzburg gloss corpora, with law, poetry, and computus), carrying 136,559 gold-lemmatized tokens: the library’s first Old Irish lemma layer, with per-token language honesty for the Latin the glosses interleave. The RIIG corpus (Recueil informatisé des inscriptions gauloises, Ausonius / Bordeaux, CC BY 4.0) brings 428 Gaulish inscriptions in Gallo-Greek and Gallo-Latin scripts, each with its editorial readings, dated findspot on the chronological axis, and French translation as an aligned sibling. The Ogham in 3D corpus (DIAS / Maynooth) adds some 500 ogham stones in real Ogham codepoints, each with aligned transliteration and romanization layers; its two published license statements conflict, so the shelf is held at the restrictive non-commercial reading pending clarification. Old and Middle Irish and Middle Welsh dictionary extracts on the reference shelf, and the two Old Irish treebanks under Universal Dependencies, complete the axis.

Sanskrit

The GRETIL corpus (Göttingen Register of Electronic Texts in Indian Languages): 780 editions and 703,068 passages spanning the Vedic saṃhitās, the Rāmāyaṇa, the purāṇas, kāvya, dharmaśāstra, philosophy, and the technical traditions — the longest chronological span of any shelf. Vedic accents are preserved; commentary layers are separately citable (kārikā against vṛtti). This shelf is licensed for non-commercial research use (CC BY-NC-SA) and is handled accordingly.

The Indic expansion

Three shelves, synchronized 18 July 2026, widen the Indic axis beyond GRETIL. The Digital Corpus of Sanskrit (Hellwig, CC BY) contributes 15,741 documents / 753,093 passages — 270 texts, about 5.46 million words with human-verified sandhi splitting, lemmatization, and morphology: the first gold-annotated Sanskrit in the lemma index. SARIT adds 78 scholarly TEI editions / 345,601 passages of works GRETIL lacks, including a complete Mahābhārata in the Southern Recension. SuttaCentral supplies the whole Tipiṭaka — 12,022 documents / 654,974 passages of roman-script Pali (the Mahāsaṅgīti edition, overwhelmingly CC0) with translator-picked English siblings riding the same segment identifiers, so a sutta reads with facing translation. Together with GRETIL these put the Sanskrit shelf past 1.8 million passages in three independent witnesses of the tradition — a breadth corpus, a gold-annotation corpus, and critical editions — with the canon of a second Indic language beside them.

Morphosyntactic treebanks

Gold-standard linguistically annotated corpora — lemma, morphology, and dependency syntax per token — from four families: PROIEL (the parallel New Testament in Greek, Latin, Gothic, Classical Armenian, and Old Church Slavonic, with classical prose), TOROT (Old East Slavic and OCS, from birchbark letters to Avvakum), ISWOC (Old English prose and the West-Saxon Gospels), and the ancient-language treebanks of Universal Dependencies — Thomas Aquinas, Vedic Sanskrit, Ruthenian, the two Old Irish glosses treebanks, Hittite (HitTB), Classical Chinese (Kyoto), Old Icelandic (IcePaHC), and the two Perseus conversions for Ancient Greek and Latin. Together with the gold layers of ORACC, DCS, AES, Coptic Scriptorium, CorPH, ReM, ReN, IcePaHC, goo300k, damaskini, TraCES, and ETCSL these feed a gold lemma index of 21.9 million rows in 45 languages (as of 19 September 2026).

Cuneiform and the Ancient Near East

The Open Richly Annotated Cuneiform Corpus (ORACC), 38 projects: 104,722 documents and 1,588,133 passages, including the complete State Archives of Assyria, the Achaemenid royal trilinguals (Old Persian and Elamite enter the library here), and the four ePSD2 corpora with the Ur III administrative mass — with upstream gold lemmatization for Akkadian and Sumerian, so a tablet displays transliteration beside running English, line by line. Released by ORACC under CC0.

Around it, since 19 July 2026, the rest of the cuneiform world: the CDLI universal catalog (353,156 artifacts under CDLI’s open grant, 135,201 of them transliterated — proto-cuneiform and proto-Elamite’s only machine-readable home, with periods, proveniences, and collections as browsable axes from Uruk IV to the Achaemenids); the TLHdig Hittite corpus (23,486 tablet manuscripts, >98% of published Hittite fragments, CC BY, with candidate morphology carried at an honest silver tier); the eBL Fragmentarium (23,288 fragments with inline English, cross-linked to their CDLI records); the Electronic Text Corpus of Sumerian Literature (394 hand-lemmatized composites with English prose siblings, deliberately unmerged from ePSD2’s editions of the same poems); and the Copenhagen Ugaritic Corpus (279 KTU tablets with per-sign cuneiform and damage flags).

The Ras Shamra Tablet Inventory (University of Chicago, OCHRE; joined 31 July 2026 under a written grant) adds the inventory spine of Ugarit: 5,075 records — one per inscribed object of Ras Shamra and Ras Ibn Hani, with RS number, KTU/CTA/UT concordances, findspot, and museum number — and line-grain published editions (740 passages, Ugaritic and Akkadian) where OCHRE publishes one. Inventory-first by design: the concordance layer is what Ugaritologists join their citations against.

Egyptian

Three millennia of Egyptian in Unicode transliteration (synchronized 18 July 2026): 101,793 gold-lemmatized sentences from the Pyramid Texts to the sawlit literary canon (AES, CC BY-SA), the only bulk Demotic anywhere (13,383 sentences) plus Late Egyptian from the Thesaurus Linguae Aegyptiae’s official datasets, the 35,052-entry Ägyptische Wortliste as the dictionary shelf, and the Comprehensive Coptic Lexicon whose egy↔cop crosswalk turns Egyptian etymology into resolvable edges.

Hebrew and Aramaic

The Masoretic text of the Leningrad Codex byte-verbatim (the combining-mark order is never normalized), fully morphology-tagged with ketiv/qere preserved; the ETCBC BHSA beside it with full clause/phrase syntax — the library’s first constituency data; the Dead Sea Scrolls (1,001 scrolls under Abegg’s own CC BY-NC grant); the verse-aligned Targums; the Peshitta Old Testament as the alignment hub’s Syriac leg; the Digital Syriac Corpus (632 documents — a millennium of classical Syriac, CC BY); the augmented-Strong’s lexicon that resolves every OSHB lemma to its BDB entry; the UBS Semantic Dictionary; and the epigraphy of Israel/Palestine (5,499 inscriptions).

Elephantine

The multilingual documentary record of one island community (the Berlin ERC papyri and ostraca database, Ägyptisches Museum und Papyrussammlung, CC BY-SA; joined 28 July 2026): 15,539 documents / 69,350 passages at the artifact grain. Imperial Aramaic leads with 5,568 passages — the Judean garrison’s archives, the library’s first substantial Aramaic document shelf — beside Greek (18,164), Demotic (5,322), earlier Egyptian (3,194), Coptic (3,237), Arabic, Latin, and Phoenician slices of the same community’s record, with 32,877 English translation siblings displayable page by page and line by line. 7,973 documents carry dates on the chronological axis, with genre, material, and object-type facets on roughly ten thousand.

Arabic and Persian

The OpenITI corpus — the premodern Islamicate library, and the largest openly licensed historical corpus in existence — was ingested on 22 July 2026: 9,079 documents / 34,631,499 passages. Classical Arabic (33,294,039 passages) is thereby the library’s largest language: Quran and hadith, history and biography, fiqh, kalām and falsafa, the dīwāns and adab. The Persian shelf (1,337,472 passages) rides the same release — Ḥāfiẓ, Niẓāmī, ʿAṭṭār, Ibn Sīnā. Passages are cited by volume, page, and paragraph; the Arabic-script search fold strips tashkeel and bridges the Persian keyboard (ی/ي, ک/ك), so Arabic-typed queries find Ḥāfiẓ and vice versa. Licensed CC BY-NC-SA — non-commercial research use, never redistributed.

Ethiopic (Gǝʿǝz)

The Ethiopic axis, joined 26 July 2026. Beta maṣāḥǝft (Hiob-Ludolf-Zentrum, Hamburg, CC BY-SA) supplies the text-bearing works of its catalog — 3,796 documents / 66,516 passages, carrying the Gǝʿǝz Bible: a near-complete Old Testament, all four Gospels, and Acts, the Epistles, and Revelation. The TraCES corpus (ERC, Hamburg) adds 15 documents / 75,436 morphologically analyzed word tokens — the Gospel of Matthew, the Kebra nagast prologue and twenty-five chapters, the royal chronicle of ʿAmda Ṣǝyon, the Testamentum Domini, and seven Aksumite royal inscriptions — the gold lemma lane of the shelf, with each analyzed token linked to its entry in Dillmann’s 1865 Lexicon linguae aethiopicae (13,727 entries) on the reference shelf: text, gold analysis, and the standard lexicon closed into one loop.

Pre-Roman Italy and Sicily

The Etruscan corpus (6,248 OpenEtruscan inscriptions with English siblings and the ETP glossary), CEIPoM’s lemmatized epigraphy of the whole peninsula (3,871 texts — Oscan, Umbrian with the complete Iguvine Tables, Venetic, Messapic, South Picene, Faliscan, archaic Latin including the Fibula Praenestina), Corpus ItAnt’s critical editions, the Lexicon Leponticum and Raetic corpora of record from Vienna, and I.Sicily’s 5,074 inscriptions across all the languages of ancient Sicily. Synchronized 18 July 2026.

Biblical editions

Editions serving the alignment hub: the complete Clementine Vulgate (73 books, public domain), the SBL Greek New Testament (CC BY), and the World English Bible (public domain) as the readable English witness. With these aboard, one Gospel citation renders across up to fifteen registered witnesses (the two Coptic dialects joined on 13 July 2026), and the Old Testament axis runs seven-legged — Masoretic text (twice, at two annotation grains) ↔ Septuagint ↔ Vulgate ↔ English ↔ Targum Onkelos ↔ Peshitta — with the Greek/Hebrew Psalm numbering divergence mapped explicitly.

Old English

The complete six-volume Anglo-Saxon Poetic Records — Beowulf, the Junius Manuscript, the Vercelli Book, the Exeter Book, the Paris Psalter, and the Minor Poems — cited by the canonical printed line numbers, alongside the ISWOC treebank (Ælfric, Apollonius, the Chronicles, Orosius, and the West-Saxon Gospels, gold-annotated) and the Bosworth-Toller dictionary on the reference shelf, with æ/þ/ð-tolerant lookup.

Slavic and Slovenian

The Old Church Slavonic canon: four gospel codices in the Helsinki CCMH transliteration (Zographensis, Marianus, Assemanianus, Savvina kniga), folio-line cited, beside the PROIEL/TOROT editions — a substrate for collation. The ~1000 CE Freising Manuscripts, the oldest Slavic text in Latin script, appear in three aligned transcription layers with five modern translations (CC BY-ND; held under the library’s strictest access class). Early Modern Slovenian print is covered by goo300k (gold-annotated, 1584–1899) and the IMP corpus (658 documents), all dated per document and feeding the chronological axis. Since 17 July 2026 the damaskini corpus (CLARIN.SI, CC BY-SA) extends the axis southward: 23 gold-annotated Balkan Slavic witnesses of the fifteenth through nineteenth centuries on the Church Slavonic–Bulgarian continuum — roughly ten of them independent witnesses of the Life of St. Petka — each paired with a full English sibling translation, with the corpus’s own norm and origin classifications carried as document facets.

The Chinese libraries

The step change of 20 July 2026: the Kanseki Repository (Kyoto’s premodern-Chinese library, KR1 classics through the KR5 Daoist canon) and the CBETA Buddhist canon (Taishō and Xuzangjing in TEI) together hold 13.2 million passages of Literary Chinese — the library’s second-largest language, behind only Classical Arabic. The Thesaurus Linguae Sericae adds sense-level attributions into the classics, and the character instruments (Unihan, the BabelStone IDS decompositions, Baxter-Sagart Old Chinese, the Guangyun rime dictionary, the Heian-period HDIC dictionaries) feed the nabu char card — a Han character answered across four millennia of structure, sound, and attestation. Traditional, simplified, and z-variant spellings fold to one search skeleton, so a modern-script query reaches the traditional-script corpus.

The Tibetan canon

The fourth canon leg — after the Pali, Sanskrit, and Chinese Buddhist canons — joined on 28 July 2026, complete. The Digital Derge Kangyur (1,200 documents / 461,304 passages: 103 volumes, Tohoku 1–1108) and Derge Tengyur (3,362 documents / 897,142 passages: 213 volumes, Tohoku 1109–4569, the commentarial half of the canon) are cited by page and line as an exact representation of the Derge woodblocks — carving variants preserved on purpose — and are public domain. Beside them, 84000 (“Translating the Words of the Buddha”) supplies the published English Kangyur translations — 388 documents / 124,223 passages, CC BY-NC-ND — for orientation. Old Tibetan Documents Online (ILCAA, Tokyo, CC BY) brings the earlier stage of the language: 413 critically edited documents in Wylie transliteration — Dunhuang manuscripts, the imperial-period pillar inscriptions including the Zhol pillar and the 821/822 Sino-Tibetan treaty pillar, and five Old Zhangzhung texts — with the Old Tibetan Annals and Chronicle (the two foundational histories, the Annals with an aligned English translation) in a companion corpus. The SOAS Classical Tibetan corpus contributes four texts with hand-corrected segmentation and gold part-of-speech tags (318,230 tokens). The desk’s dictionary rack sits on the reference shelf: the Mahāvyutpatti — the ninth-century imperial Sanskrit–Tibetan translation glossary, the crosswalk between the Sanskrit shelves and this canon — plus the Tibetan Verbs Database and a Wiktionary-derived Tibetan lexicon.

The Japanese reading desk

Aozora Bunko — the volunteer-transcribed Japanese public-domain library — joined on 21 July 2026: 2,991,807 passages of Meiji-and-later literature with a thin classical tail, ruby readings preserved as annotations rather than flattened into the text, and the ~2,285 works in kyūjitai orthography reachable from modern-form queries through the kanji reform fold (国 finds 國). Only copyright-expired works are ingested — the in-copyright remainder is excluded before a single file is opened. Beside it sit ONCOJ’s gold-morphology Old Japanese (the Man’yōshū-era corpus) and the EDRDG dictionaries.

The Germanic shelves

22 July 2026 completed the desk across all three branches. North: Menotec’s seven Old Norwegian treebanks with the Poetic Edda of Codex Regius (20,308 gold-annotated sentences), and Rundata — the Scandinavian runic corpus, ~6,800 inscriptions in up to five text lanes each (scholarly transliteration primary, Old-West-Norse and runic-Swedish normalisations, English and Swedish translations), the damage notation stored verbatim as content. West: the Old Saxon Heliand parsed to the token with gold form-lemma pairs (HeliPaD), and the Referenzkorpus Mittelhochdeutsch — 406 manuscripts / 355,449 diplomatic manuscript lines whose gold annotation makes Middle High German one of the corpus’s largest lemma pools (2.10 million rows) on day one. East was already home: Wulfila’s Gothic on the PROIEL treebank. Old Icelandic went live with the owner’s same-day Universal Dependencies sync — 44,029 IcePaHC passages whose 812,484 gold rows entered among the largest lemma pools. The Hanseatic rung followed on 26 July 2026: ReN, the Referenzkorpus Mittelniederdeutsch/ Niederrheinisch 1200–1650 (CC BY), with 235 documents / 297,504 passages — 161 of them gold-annotated, about 1.49 million tokens with lemma, part of speech, and morphology — between ReM’s Middle High German and the Norse and Old English shelves.

Reference shelf

Dictionary shelves as structured data rather than page images — 1,998,589 entries as of 19 September 2026. Liddell-Scott-Jones (116,497 entries), Lewis & Short (51,636), Bosworth-Toller (62,815), the Monier-Williams Sanskrit-English dictionary (193,890, synchronized 13 July 2026, with transliteration-tolerant lookup and citations resolving into the Sanskrit shelf), a Wiktionary-derived Old Church Slavonic lexicon (4,615), seven reconstruction dictionaries (Proto-Indo-European, Proto-Slavic, Proto-Germanic, Proto-West Germanic, Proto-Balto-Slavic, Proto-Italic, Proto-Indo-Iranian) whose entries carry machine-readable descendant trees, and — since 17 July 2026 — Old Irish, Middle Irish, and Middle Welsh extracts on the same Wiktionary shelf. Three etymological witnesses joined on 14 July 2026: the IE-CoR Indo-European cognacy database (4,981 expert-curated cognate sets with loan events flagged), the LIV-LOD dictionary of Proto-Indo-European verbs (305 verbal etymons), and de Vaan’s Etymological Dictionary of Latin (2,860 etymons, LiLa linked-data edition) — independent, expert-curated chains beside the Wiktionary-derived ones.

Two further shelves arrived with the 17 July 2026 synchronizations. The StarLing / Tower of Babel etymological databases (27,397 entries, held under a written grant from their maintainer with per-compiler credit rendered on every surface) carry the Moscow-school apparatus: Pokorny’s complete Indogermanisches Etymologisches Wörterbuch (2,222 roots), Nikolayev’s Walde-Pokorny-based PIE database (3,291 etymologies with per-branch reflex columns), Vasmer’s etymological dictionary of Russian in the Trubachev edition (18,239 entries), and the Common Germanic and Baltic databases. The Slovenian historical dictionary shelf (ZRC SAZU via CLARIN.SI, CC BY, 139,405 entries) unites Pleteršnik’s Slovene-German dictionary of 1894–95, the lexicon of Janez Svetokriški’s Baroque sermons, and the complete word inventory of sixteenth-century Slovenian print — closing the gold-lemma-to-dictionary loop for Slovenian as LSJ closes it for Greek. The expansions of 18–21 July 2026 added the Hebrew desk (the augmented-Strong lexicon, the BDB outline, and the UBS Semantic Dictionary of Biblical Hebrew), the Egyptian dictionary chain (the TLA Wortliste and the Comprehensive Coptic Lexicon), and the Sino-Japanese desk — Unihan’s 65,092 codepoint records, KANJIDIC2 and JMdict, the Heian-period HDIC dictionaries, the Guangyun rime dictionary, and the ONCOJ Old Japanese lexicon.

Late July and early August 2026 added four more groups. The Tibetan rack: the Mahāvyutpatti (9,379 entries, public domain), the Tibetan Verbs Database (2,491 entries with the grammarians’ disagreements deliberately uncollapsed, CC0), and a Wiktionary-derived Tibetan lexicon (3,651). Dillmann’s Lexicon linguae aethiopicae for Gǝʿǝz (13,727 entries, the very entries the TraCES gold tokens link to). Two cross-linguistic typological instruments: CLICS³, the database of colexifications (1,647 concept entries whose cards show which languages express two meanings with one word), and WOLD, the World Loanword Database (64,289 entries across 41 recipient languages with a curated borrowed status per word). And — 2 August 2026 — DÉRom, the Dictionnaire Étymologique Roman (ATILF): 233 Proto-Romance etymon articles with per-language reflex sections across the whole Romance family, the comparative-method bridge between Latin and its daughters (CC BY-NC-SA). Dictionary citations are resolved to live passages where the cited work is held: the LSJ entry for μῆνις points at Iliad 1.1 as a resolvable URN, not a printed abbreviation.

The local shelves

Four shelves hold material that is authored or acquired rather than downloaded. A language-dossier shelf carries the library’s per-language curation — one plain Markdown file per language code, feeding the nabu language reference cards. A local-library shelf files the owner’s own PDFs, scans, and offprints as catalogued, page-cited collections through the nabu ingest command (see Tools), which since July 2026 also accepts http(s) URLs, downloading first and recording the given address in the manifest; everything on this shelf is held under the library’s strictest access class by default and is never served or redistributed. A source-dossier shelf carries a curated description of every registered source, served on the nabu list census and checked against the shelf map at every development gate. And a notes shelf records the owner’s own annotations on any citable URN through nabu note — scholia of one’s own, rendered wherever the target is shown. What these shelves hold is the owner’s private business: their contents and counts are deliberately absent from this site.