Tools

Nabu is operated from the command line (bin/nabu), and everything the command line can do for a reader, a connected AI assistant can do in conversation through the built-in read-only MCP server — see Ask your model below. The commands are grouped by scholarly task; every pasted example is a real command with output from live runs (trims marked with …).

Finding text

Full-text search (nabu search QUERY) runs over the whole corpus (106,902,485 passages, 19 September 2026) with per-language orthographic folding: unaccented μηνιν finds μῆνιν; iuvenis, juvenis, and iuuenis all resolve; Old English æ/þ/ð and cuneiform determinatives are handled analogously. Results can be filtered by language, license class, and — where documents are dated and placed — by historical year and provenance:

$ bin/nabu search 'στρατηγ*' --from 101 --to 300 --place oxyrhynch%

finds the strategoi of the Oxyrhynchite nome in the papyri of the second and third centuries. The chronological axis covers 688,562 dated and placed documents as of 4 August 2026 — the CDLI tablets, Italy’s EDR (115,590), the Heidelberg inscriptions and the papyri foremost among them. On the faceted shelves (currently the inscriptions), genre facets compose with the same filters:

$ bin/nabu search manibus --type epitaph --province Britannia

draws on 256,518 facet rows recording inscription genre, province, material, and object type, with uncertain upstream attributions preserved as such.

Chinese, Japanese and Korean text searches first-class (September 2026): the classical Sinitic shelves — the Chinese canons, the Korean state records, the Japanese library — write without word boundaries, so the library maintains a dedicated character-pair index over those shelves (fifteen sources). A query in Han characters or kana (nabu search 六龍) matches exact character sequences at full speed, composing with every filter below; no segmentation is guessed, ever.

Stage-aware search (search --lect LECT-ID, August 2026, with the optional nabu-lects registry synced) filters hits by historical stage rather than bare code: --lect lat:med keeps only collections resolving to Medieval Latin, --lect lat matches every Latin stage — prefix semantics over the registry’s anchor[:stage][/variety][~script][@ortho] grammar, with stored codes never touched. See the Languages page for the lect layer.

Place-identity search (search --place NAME|ns:id, August 2026): the provenance filter understands place identities, not just spellings. A namespaced ref (--place cigs:GIR, --place pleiades:912855) matches every document whose place claim cites that identity in any spelling — upstream URL or namespaced mint alike — and a plain name resolves through the held gazetteers and the nabu-places registry’s decisions, crosswalk equivalences included, while still matching verbatim name fields. The page names the lane that answered:

$ bin/nabu search lugal --place Girsu --scan --limit 5
…
note: place: "Girsu" → cigs:GIR = pleiades:912855 = cdli-provenience:88 = geonames:93976 (identity refs + name match)

--scan there is the filter-first mode (August 2026): where plain search asks “where does this term rank corpus-wide?” — and answers with a corpus-wide sample when the term is too common to rank — --scan asks “which of these documents carry the term?”, walking the matches in corpus order intersected with the active filters: deterministic, never sampled, built for a common word under a selective filter.

Script search (search --script TAG, August 2026) cuts by writing system: the held text’s script axis or the artifact’s original system where the two differ. --script ogam walks the Ogham stones, --script grek finds Greek-alphabet Gaulish:

$ bin/nabu search --script grek --limit 3
urn:nabu:riig:ahp-01-01:HRD-a:1 [xtg-Grek]
  καρε[…]μ

Sign search (search --sign GLYPH|NAME, August 2026) is the inverse of the sign desk: one cuneiform sign expands into an OR over its OSL reading values, each hit bracketing the value that matched:

$ bin/nabu search --sign 𒊬 --lang sux --limit 3
sign: 𒊬  SAR — 26 reading values: e₁₄ kiri₆ ma₄ mu₂ mud₆ … and 14 more
urn:nabu:etcsl:0.2.06:31 [sux]
  cubur-e cu [mu2]

Geo-radius search (search --within LAT,LON,KM, August 2026) keeps documents whose recorded find-location falls inside a radius — --within 59.86,17.64,30 walks the runestones raised within 30 km of Uppsala. All of these compose with each other and with the date, genre, license, and lect filters.

Proximity search (search A --near B --window N) keeps only passages where the second term occurs within N words of the first — collocation probing in the tradition of the TLG interface, and composable with lemma search:

$ bin/nabu search λόγος --near θεός --window 5 --lang grc
urn:nabu:ddbdp:p.oxy:8:1151:18   [θεοσ] ην ο [λογοσ].          ← a papyrus amulet…
urn:cts:…:tlg0031.tlg004…:1.1    …και [θεοσ] ην ο [λογοσ].     ← …quoting John 1:1

Fragment search (search --fuzzy FRAGMENT) matches a damaged-text fragment anywhere inside a passage, mid-word included, typed straight off the edition with its editorial brackets:

$ bin/nabu search --fuzzy ']ανδρα μοι εν['
urn:nabu:ddbdp:bgu:6:1470:ctr:6 [grc]
  μαρτυροι. [ανδρα μοι εν]νεπε μουσα πολυτρο
1 hit (fuzzy substring; highlights are diacritic-folded)
fuzzy index covers: oracc, papyri-ddbdp

— BGU 6.1470, a Hellenistic writing exercise breaking off mid-word through the opening line of the Odyssey. The character-trigram index behind this is scoped to the documentary shelves, where fragment search earns its cost; the scope has grown in waves — the Heidelberg inscriptions joined 13 July 2026, Italy’s EDR and the Elephantine archives 28 July 2026 — and the production index covers 3,801,508 passages across the papyri, cuneiform, and inscription shelves as of 28 July 2026.

Concordance (nabu concord QUERY) prints classic keyword-in-context lines, column-aligned in the pristine (accented) text and in corpus order — for scanning usage rather than ranking relevance:

$ bin/nabu concord --lemma virtus --width 30
…vetii quoque reliquos Gallos virtute praecedunt, quod fere cotidi…  urn:nabu:proiel:caes-gal:52552 [lat]
…i populi Romani et pristinae virtutis Helvetiorum.                   urn:nabu:proiel:caes-gal:52635 [lat]
…b eam rem aut suae magnopere virtuti tribueret aut ipsos despicer…  urn:nabu:proiel:caes-gal:52636 [lat]
…

The attic (search --withdrawn TERM, September 2026) searches what the library no longer serves: passages upstream revised away and documents upstream withdrew, each hit labelled with why it was withdrawn (an upstream deletion, a revision that pruned it). The catalog never hard-deletes — text that once entered the library remains findable, honestly marked, in its own view.

Words and meanings

Above the letter-for-letter layer sit two ways of searching by what a word is rather than how it is spelled.

Lemma search (search --lemma FORM) queries by dictionary form rather than surface string — inflection and suppletion included, so one query finds a Latin noun in all its cases or a Greek verb across its stems:

$ bin/nabu search --lemma λέγω --limit 3
urn:nabu:proiel:chron:108755 [grc]  λέγω → ῥηθέντος  (lay)
  ὅ περ ἦν καὶ αἴτιον τοῦ μὴ ἐλθεῖν τὸν γενήσαντά με εἰς τὸν Μορέαν μετὰ τοῦ αὐθεντοπούλου κὺρ Θωμᾶ εἰ…
urn:nabu:proiel:chron:121080 [grc]  λέγω → εἶπον, εἰπὲ  (lay)
  Πολλῶν οὖν λόγων δαπανηθέντων, τέλος ἐστάλησαν πρὸς τὸν ἄνθρωπον δύο τῶν κελλιωτῶν καὶ συντρόφων μου…
urn:nabu:proiel:chron:121083 [grc]  λέγω → εἴπω  (lay)
  Ἐγὼ δὲ νὰ ἀκούω παρὰ μὲν τῶν, ὅτι καλή ἐστι, παρὰ δὲ τῶν, ὅτι οὐ καλή, διὰ τὶ νὰ μηδὲν εἴπω·
3 hits (exact lemma match; text is pristine)

The coverage comes in two honestly separated tiers: 21.9 million gold annotations in 45 languages — human-verified, from the treebanks — and 35.9 million silver annotations in 15 languages, proposed by machines and labelled [silver] wherever they surface; --gold-only restricts any lemma query to the human-verified tier, and a machine guess is never blended into a gold count. The silver tier is grown by the library’s own local lemmatization campaigns — the how and the honesty rules are written up in lemma enrichment. A morphology filter (--morph case=gen,number=pl, in Universal Dependencies vocabulary) restricts hits to attestations with the stated gold morphology.

Semantic search (search --similar URN, September 2026) finds passages by meaning: anchor on any passage of the literary core and get its nearest same-language neighbors — the same line in another held edition, a quotation drifted by memory, a paraphrase — ranked and banded close / near / loose, with the model named so a statistical match is never mistaken for a curated alignment. Behind it is a local vector store the owner builds once with nabu embed (fully on-box; no text leaves the machine) and tops up incrementally after every sync. The plain-language write-up — what embedding is, what it can and cannot answer — is embed.md.

Reading and citing

Retrieval by citation (nabu show URN) returns a passage, a whole document, or a citation range, with license, revision, and the full provenance trail. --random pulls something off a shelf; --parallel pairs the aligned English translation, span-grouped and honest when the translation’s citation grain is coarser than the original:

$ bin/nabu show urn:cts:greekLit:tlg0012.tlg001.perseus-grc2:1.1-1.5 --parallel
urn:cts:greekLit:tlg0012.tlg001.perseus-grc2 — Iliad [grc]
  parallel: urn:cts:greekLit:tlg0012.tlg001.perseus-eng4 — Iliad [eng]
  aligned by citation: 0 paired, 1 block covering 5 lines, 0 grc only, 0 eng only
  :1.1
    grc  μῆνιν ἄειδε θεὰ Πηληϊάδεω Ἀχιλῆος
  :1.2
    grc  οὐλομένην, ἣ μυρίʼ Ἀχαιοῖς ἄλγεʼ ἔθηκε,
  …
  eng [:1.1 — covers :1.1–:1.39; range shows :1.1–:1.5]
    Sing, O goddess, the anger [mênis] of Achilles son of Peleus, that brought
    countless ills upon the Achaeans. …

Alignment (nabu align REF) renders one citation across every witness of a registered work — the parallel New Testament (up to fifteen registered witnesses, the Sahidic and Bohairic Coptic among them since 13 July 2026) and the Septuagint ↔ Vulgate ↔ English Old Testament ship as flagships; a full example opens the Home page.

Collation (align REF --collate) turns the aligned witnesses into a compact apparatus: a base reading with per-witness divergences, computed per language-and-script group. Witnesses in a different script family are rendered undiffed and labelled as such — the tool does not pretend that a Cyrillic and a Latin-transliteration witness can be mechanically collated.

$ bin/nabu align "MARK 2.3" --collate --base zographensis

Export (nabu export --format plain|jsonl) streams the corpus out with language and license filters — the exit format, so the collection is never captive to its own tooling.

The reference desk

Dictionary lookup (nabu define LEMMA) queries the reference shelf (114 dictionary shelves, 1,998,589 entries as of 19 September 2026: LSJ, Lewis & Short, Bosworth-Toller, Monier-Williams, the reconstruction dictionaries, the StarLing bases with Vasmer, the Slovenian historical dictionaries, the Hebrew and Egyptian lexica, the Sino-Japanese desk, and on), with the entry’s citations resolved to live passages in the local catalog:

$ bin/nabu define μῆνις
μῆνις — A Greek-English Lexicon (Liddell-Scott-Jones) [attribution]  urn:nabu:dict:lsj:n67485
  gloss: wrath

μῆνις, Dor. and Aeol. μᾶν-, ἡ, gen.
A. μήνιος Pl. R. 390e , later μήνιδος Ael. Fr. 80 , … —wrath; from Hom.
downwds. freq. of the wrath of the gods, Il. 5.34 , al., A. Ag. 701 (lyr.),
… but also, generally, of the wrath of Achilles, Il. 1.1 , al. …

resolved citations (in this corpus — nabu show <urn>):
  Il. 1.1 → urn:cts:greekLit:tlg0012.tlg001.perseus-grc2:1.1
  Il. 5.34 → urn:cts:greekLit:tlg0012.tlg001.perseus-grc2:5.34
  A. Ag. 701 → urn:cts:greekLit:tlg0085.tlg005.1st1K-grc1:701
  …

Etymology (nabu etym LEMMA) walks an attested lemma to the reconstructions that name it among their descendants, then up the proto-to-proto chain — through the intermediate shelves where they exist (Latin through Proto-Italic to Proto-Indo-European, for instance) — each stage listing cognates with corpus attestation counts, and curated loan events labelled as such. Plain-ASCII input is accepted for reconstructed forms:

$ bin/nabu etym богъ --lang chu
богъ [chu] → *bogъ [sla-pro] — gloss: god
← *bʰeh₂g- [ine-pro] — gloss: to divide, distribute, allot
  reflexes: [grc] ἔφᾰγον, [sa] भक्ष (bhakṣá), …

Since 14 July 2026 three expert-curated witnesses answer beside the Wiktionary-derived chains: the IE-CoR cognacy database (4,981 cognate sets), the LIV verbal roots, and de Vaan’s Latin etymological dictionary — so the same walk can be checked against independently curated scholarship. The StarLing bases joined on 17 July 2026 (Pokorny, Nikolayev’s PIE database, Vasmer, Common Germanic, Baltic), and where the crosswalk has no reconstruction path for a form, etym now falls back to the plain dictionary lookup rather than missing what define would find.

Language cards (nabu language CODE) explain any language code the library surfaces — the corpus languages and the 803 Wiktionary etymology codes that appear in etym cognate lists — on one card: name, family, curated historical context, and the code’s live holdings in the catalog. Where the code is a registered anchor in the nabu-lects registry, the card ends with the stage ladder — live holdings per historical stage (Classical, Late, Medieval Latin…), reconstructed stages asterisked. An unknown code is reported honestly, with a family hint; nabu language --list prints the held languages.

$ bin/nabu language gkm

Sign cards (nabu char CHAR) put one glyph on the desk, dispatched by script. A Han character gets the dictionary-shelf card (structure, readings, and a diachronic column from Old Chinese through the Heian dictionaries) — and the door works in reverse: a pinyin reading (nabu char wen, tones optional) or a Japanese on/kun reading — kana (nabu char ひと) or plain romaji on the dictionary-caps convention, CAPS for on’yomi (nabu char TAI), lowercase for kun’yomi (nabu char kuni) — lists every character that carries it. A cuneiform sign — glyph, sign name, spelled value, or sign-list number (nabu char 𒊬, nabu char SAR, nabu char szesz, nabu char MZL535, bare nabu char 852 matched across every held list) — gets the Oracc Sign List card: sign name, the MZL/LAK/ABZL/HZL/ŠL print-list concordances, every reading by language, the CDLI meaning glosses, Wiktionary’s sense glosses (𒊬: “orchard”, “to write (on), inscribe” — the kaikki Sumerian shelf), variant forms, and how often each spelled value actually occurs across the held tablet corpora. An Egyptian hieroglyph (nabu char 𓅃, or by Gardiner-style code nabu char G5) gets the Unicode 17 Unikemet card: catalog code, description, function, phonetic value, the JSesh/Hieroglyphica/IFAO concordances, and the sign’s attestation across the held AES corpus’s sign annotations. An ambiguous value lists every candidate sign, never one silently; --json on the sign cards emits a frozen machine contract. The token-by-token sibling is nabu signs, which reads whole transliterated lines through the same sign list.

$ bin/nabu char 𒊬
𒊬  SAR  ·  U+122AC
sign lists: LAK 215 · ABZL 385 · HZL 353 · MZL 541 · …
values (OSL): kiri₆ · mu₂ · nisi · sar · šar · …
meanings (CDLI): sar → v. to write; n. vegetable · kiri₆ → n. garden
in the corpus: sar 100333 · nisi 27997 · kiri₆ 8737 · …

Place cards (nabu place NAME|ID) are the library’s third dimension, after language and time: one ancient place resolved through the local Pleiades gazetteer dump — title, id, place types, attested period span, representative point — followed by each source’s holdings at that place, counted over the Pleiades ids the epigraphic sources themselves assert in their headers (never fuzzy-matched). Input is a Pleiades numeric id or an exact title; id-less documents whose findspot text mentions the name are listed as a separate, honestly labelled tail. When a shown inscription carries such an id and the dump is on disk, nabu show adds a one-line findspot resolution too. Since the places program (August 2026) the holdings side reads a multi-gazetteer index (Pleiades, Trismegistos Geo, CIGS) and nabu place apply projects the nabu-places registry’s matching decisions into the catalog — see Places for the system and the coverage census.

$ bin/nabu place Segesta
$ bin/nabu place apply

Cognates in parallel (nabu cognates TARGET) crosses the etymological crosswalk with the alignment hub: verses of an aligned work where witnesses in two or more languages use reflexes of the same reconstructed root, found without any surface-form resemblance:

$ bin/nabu cognates "LUKE 14.34" --langs got,chu
LUKE 14.34  *sḗh₂l [ine-pro · attribution]
    chu  соль — attested as солъ
    got  salt

The whole Gothic × Old Church Slavonic New Testament yields roughly 300 such verses across 30 roots in under a second; each hit names the dictionary shelf on which the languages meet, so a Proto-Germanic meet for a Slavic word is flagged as a likely borrowing rather than common descent.

The corpus reads itself

Intertext (nabu parallels URN) is passage-anchored quotation and echo discovery: point at one passage and find where the corpus quotes or reworks it, ranked by shared rare phrases, with elision folded across editions (so Matthew 4:4 finds Septuagint Deuteronomy 8:3). Gold-lemmatized anchors also surface re-inflected allusions through rare lemmas. Its statistical sibling is search --similar above — parallels ranks shared phrases with evidence, semantic search ranks shared meaning; the two answer different questions and neither pretends to be the other.

$ bin/nabu parallels urn:nabu:sblgnt:matt:4.4

Formula mining (nabu formulas SCOPE) points the same machinery inward: the repeated formulas within a corpus slice, ranked by count and length — Homer’s ὣς ἔφαθ᾽ οἵ δ᾽ (72 occurrences), the Beowulf maþelode bearn Ecgþeowes speech formula, the Old English riddle refrain saga hwæt ic hatte.

$ bin/nabu formulas aspr --min-count 5

The links graph (nabu links URN) reads back the mined cross-reference graph: every batch-produced edge touching a URN — parallels, formulas, cognates — with its evidence and a provenance footer naming the producing run. Edges live in their own journal database and survive catalog rebuilds.

Vocabulary profiling (nabu vocab URN) computes a lemma-frequency profile of a document or range against the gold-lemma corpus: distinctive vocabulary by log-odds (Caesar surfaces legio and proelium; Cicero’s De officiis surfaces officium, honestas, decorum) and the in-document hapax legomena. With --by-century it plots the shape of the dated corpus, or one word’s fortunes, across historical centuries:

$ bin/nabu vocab --by-century 'στρατηγ*' --lang grc

peaks in the second century CE.

Ask your model

Everything above is also available in conversation: the library ships a read-only MCP server (bin/nabu mcp) exposing thirteen tools — nabu_search, nabu_show, nabu_define, nabu_etym, nabu_links, nabu_parallels, nabu_cognates, nabu_align, nabu_concord, nabu_place, nabu_char, nabu_signs, and nabu_status — so an AI assistant wired to your library can search, cite, align, and etymologize over it mid-discussion. Registration for Claude Code and Claude Desktop is in the server documentation; restricted material is excluded by default, license classes ride every payload, and nothing is ever written. The command-line search modes travel with it: nabu_search takes the lect historical-stage filter and, since September 2026, a similar_to anchor — the semantic-search lane in conversation, banded and model-named like its CLI twin.

The point is composition: a research question is usually two or three tools, not one. Every example below — and every “Ask your model” block on the axis pages — was run live against this library before it was written down; nothing is invented.

• “Who quotes the opening of the Iliad?” — nabu_parallels on Iliad 1.1 returns Galen, Aristotle’s Ars Rhetorica, and Sextus Empiricus (seven loci) sharing μῆνιν ἄειδε, each with the matched phrase and a urn nabu_show opens in pristine text.
• “Show me MARK 2.3 in every witness.” — nabu_align returns fourteen columns at once: Greek, Latin, Gothic, Armenian, four Old Church Slavonic codices, Old English, Sahidic Coptic, English. collate: true turns them into an apparatus.
• “Where does Gothic guþ come from?” — nabu_etym walks it to Proto-Germanic *gudą (64 cognates, nine attested here with counts) and the PIE ancestors above it.
• “What survives from Segesta?” — nabu_place resolves the gazetteer card (Pleiades 462487) and counts the library’s holdings at that place, per source.
• “Attestations of šarru ‘king’?” — nabu_search by lemma resolves the logogram LUGAL through ORACC’s lemmatization; the first hits are the Cyrus Cylinder.

Each desk has its own examples — and its own conventions (Hittite is typed syllabified, Old Japanese romanized) — on its axis page under Ask your model.

Stewardship

Lect rulings (nabu lect …, added 4 August 2026) manage the layer beneath language codes: which historical stage of a language a document actually attests, in nabu-lects identifiers (lat:med, akk:ob, grc:koi). Assignments live in their own journal (db/lects.sqlite3) that survives nabu rebuild and rides backups, at three grains: lect assign URN LECT records a single ruling; lect apply-rules compiles the ratified facet rules (a CDLI period label, an AES corpus slice, a DCS Vedic school tag names the stage of every document carrying it); lect infer-dates assigns a stage when a document’s date interval sits inside exactly one stage band — containment, never overlap, so an inscription dated across a stage boundary stays honestly unstaged. Both batch commands render a full census under --dry-run before anything is written, re-runs supersede only their own rows, and a hand ruling is never overwritten by a rule. lect check-dates audits the reverse direction: assignments whose document dates contradict the assigned stage’s band. The resolved lect is materialized as an ordinary document facet, so show prints it, search --lect filters by it, and the language card’s stage ladder counts by it.

$ bin/nabu lect assign urn:nabu:perseus-latin:phi0119.phi001 lat:arch --note "Plautus"
$ bin/nabu lect apply-rules --dry-run
$ bin/nabu lect infer-dates --source edh --dry-run
$ bin/nabu lect check-dates

Enrichment campaigns (nabu embed, nabu lemma-enrich) are the two owner-fired long runs that grow the layers above: nabu embed builds and incrementally maintains the semantic-vector store behind search --similar, and nabu lemma-enrich runs the local lemmatizer over a language’s uncovered passages to grow the silver tier. Both are census-first (an honest count and time estimate before anything runs), resumable at fine grain, and fully on-box; their tool stacks install with one idempotent command each (rake tools:embed, rake tools:stanza[la]). The plain-language write-ups: embed.md · lemma-enrichment.md.

Ingest (nabu ingest FILE...) files your own material — scanned grammars, offprints, reading notes — into the local library shelf: the file is copied in (never moved), metadata candidates are derived mechanically (PDF metadata, filename heuristics, sha256) and confirmed interactively, with AI assistance (--assist pipes a brief to any suggester command and prefills the prompts), or scripted (--yes plus flags); the shelf then syncs and the new urn is printed. The command also accepts http(s) URLs, downloading first (redirects followed) and recording the given address in the manifest. Everything on this shelf defaults to the research_private license class — catalogued and searchable locally, never served or redistributed. The same command scaffolds a language dossier (ingest --shelf language CODE) or a source dossier (ingest --shelf source SLUG).

The content census (nabu list [SOURCE]) is the what-is-held view beside nabu status’s sync-state view: bare, it prints one line per shelf with document, passage, and entry counts, languages, and the effective license-class mix; with a source it prints that shelf’s card — identity, credit line, the curated dossier description, per-language breakdown, and date-axis and facet coverage — and --documents / --entries / --collections enumerate the holdings with filters.

$ bin/nabu list
$ bin/nabu list corph --documents --limit 10

Working by research desk. The flat source list is also tagged into 25 research axes — the owner’s scholarly desks (the Classicist, the Assyriologist, the Sinologist…). A source wears every desk it serves, and four surfaces read those tags: nabu enable NAME puts a desk’s shelves in this box’s profile (the first-time step — sync acquires only enabled sources), nabu list --axis NAME groups the census under a desk, nabu search --axis NAME scopes a query to a desk’s shelves (the multi-source generalization of --source), nabu sync NAME syncs a desk’s enabled members, and nabu axis NAME prints the desk card — members, live holdings, and gold coverage. Each desk’s own page collects its shelves, instruments, CLI recipes and terminal setup.

$ bin/nabu axis celtic
$ bin/nabu search μηνιν --axis celtic

Owner notes (nabu note URN [TEXT]) record your own annotations — scholia of one’s own — against any citable URN the corpus knows: documents, passages, ranges, or dictionary entries. The URN is resolved against the catalog before anything is written; the notes live as plain YAML on a local shelf, render wherever the target is shown (show, define, links), and are served to AI clients under the same withholding rules as their targets. A bare nabu note URN reads back what you said; --list enumerates.

$ bin/nabu note urn:nabu:ccmh:mar:mt "Collate against Jagić 1883 before citing."

The remaining commands keep the collection alive: nabu quickstart syncs a curated starter shelf (four sources, about 690 MB) and prints the first commands to try; nabu sync fetches and loads a source (idempotent and non-destructive, with every run recorded); nabu status and nabu health report per-source counts, run history, and upstream drift — health also checks a set of mechanical invariants (failed or partial loads, configuration-versus-catalog mismatches, pending migrations, quarantine counts measured against an audited baseline) so that a sync that did not do what it claimed is surfaced rather than discovered; nabu verify re-parses every canonical file and compares content hashes against the catalog; nabu rebuild regenerates the entire database from canonical data, proven byte-identical by test; and nabu backup copies everything non-derivable to external storage, with a rehearsed restore drill. These are documented in the repository’s operations runbook.

These instruments read best in use. The Examples put them in the hands of a classicist, a papyrologist, an assyriologist and others — real sessions, end to end — and each research desk then gathers the commands, shelves and terminal setup for one field.