Bibliographic Madrigal

[{"type": "function", "function": {"name": "read_record", "description": "Read this object's current Revise form by node: every field's id, label and value, plus the attached evidence's markup. No thread is returned.", "parameters": {"type": "object", "properties": {}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "read_markup", "description": "Read the markup of this call's one selected canvas, retaining its identifiers for citations.", "parameters": {"type": "object", "properties": {}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "read_quotation", "description": "Read this call's one selected quotation and its markup. No other quotations are available.", "parameters": {"type": "object", "properties": {}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "zoom_image", "description": "Read a region of the selected canvas image at higher resolution: top-left and size in a 0–1000 space.", "parameters": {"type": "object", "properties": {"x": {"type": "integer"}, "y": {"type": "integer"}, "width": {"type": "integer"}, "height": {"type": "integer"}}, "required": ["x", "y", "width", "height"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "search_records", "description": "Look up Gnorium record titles by typeahead query. Results are record metadata, never thread content or additional evidence.", "parameters": {"type": "object", "properties": {"title": {"type": "string"}, "kind": {"type": "string", "enum": ["bibliographic", "lexicographic"]}}, "required": ["title", "kind"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "suggest_revision", "description": "File an ordinary pending suggested revision, by gnorium, on this call's object. Use only concrete changes supported by the selected evidence and cite it in reason. fields maps exact human Revise field ids (read_record supplies them) to string values. For the selected canvas's page, use_confirmed_draft: true files the draft as confirmed by validate_tei (layout_ok with its review_token) as markup, or on a call in translation as translated-markup, with nothing retyped; an edit since the confirmation voids it. Or give it in fields: markup (translated-markup in translation) as a JSON string mapping the attached evidence id to that same confirmed page, byte for byte. Only the selected canvas can be rewritten. Do not change prompts. Filing does not accept the revision.", "parameters": {"type": "object", "properties": {"fields": {"type": "object", "additionalProperties": {"type": "string"}}, "reason": {"type": "string"}, "use_confirmed_draft": {"type": "boolean", "description": "File the draft confirmed with layout_ok as the selected canvas's page."}}, "required": ["fields", "reason"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "edit_tei", "description": "Correct the draft with a list of edits, applied together in one call: each names its place by locator, an exact span of the draft as it stands before the call that occurs once in it, and an operation on it: replace it with fragment, insert fragment before or after it, or delete it. All the locators are found in the same draft, so order does not matter; edits may not overlap. If any edit fails, none is applied and each edit's result says which failed and why. Returns each edit's result and what the validators find in the new draft.", "parameters": {"type": "object", "properties": {"edits": {"type": "array", "minItems": 1, "items": {"type": "object", "properties": {"locator": {"type": "string", "description": "An exact span of the draft, occurring once in it."}, "operation": {"type": "string", "enum": ["replace", "insert_before", "insert_after", "delete"]}, "fragment": {"type": "string", "description": "The TEI to put in its place or beside it; empty or left out for delete."}}, "required": ["locator", "operation"], "additionalProperties": false}}}, "required": ["edits"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "read_tei", "description": "Read the draft by line: offset is the first line (1-based), limit how many.", "parameters": {"type": "object", "properties": {"offset": {"type": "integer", "minimum": 1}, "limit": {"type": "integer", "minimum": 1}}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "validate_tei", "description": "Check TEI against the encoding rules: the draft, or the fragment or document given as tei. With features, the features the page shows, it also checks that each is encoded and that everything encoded is declared. Checks render spacing and returns the canonical rendered TEI beside the scan, registered as the reader's overlay lays it (its main text block—the printed text, never its marginalia—on the render's, the whole scan shown), and the structure as the reader's overlay shows it (the rendered page with each line's and block's ink box outlined, the render's solid and the registered scan's dashed: gray where a line pair is within a quarter of the body's leading (an owner's mark's, of its own line pitch) and its aligned edges within an em of its type, orange where it is not, green a rendered line with no scan line, red a scan line with no rendered line), with layout_measure: the scan's and the render's text lines measured in em (1em the work's body type, the size at which the edition's face has the print's x-height (or cap height)) in that one frame and paired block by block, marginalia measured from the frame, and a lineation finding wherever a block's printed and rendered line counts differ. Inspect the whole page first, optionally up to 3 targeted regions per page only when needed (reason required), then confirm layout_ok with review_token before suggest_revision. Set the page as the edition, in standard TEI: physical lb/cb/pb breaks; rend for alignment (align(left|center|right|justify)), ink (color(NAME)) and typeface class. 1em is the work's body type, the size at which the edition's face (Noto Serif) has the print's x-height (or cap height), and the witness is set in one type scale, measured across its scans and declared in its header by the tools: its steps size-N (font-size), leading-N (line-height), indent-N (text-indent) and gap-N (margin-block-start), each in rem, the body type itself, so a step is the same wherever it stands. Set the type's size, leading, indents and gaps by referencing the steps, rendition=\"#size-2 #leading-1\" (several joined by a space), the step whose value layout_measure's measured value is; never write their values, and never write a <rendition> or <tagsDecl>. A font-size, line-height, indent or block margin the page writes off the scale (as computed in rem) is a finding. A block's style cascades to its children as CSS's does, and a font-size is held to 0.5em–4em of the body type as computed. Letter- and word-spacing are measured style in em. Numerals set as old-style figures in the print take font-variant-numeric: oldstyle-nums (lining-nums where a block of them sets lining figures). rend indent(1..6), hanging, larger and smaller are accepted but coarse: reference the step instead, never both. Gaps: a block's gap above it is its gap step (rendition=\"#gap-1\"); a vertical <space dim=\"vertical\"/> in whole lines (unit=\"lines\"), or in em a gap step's value; a gap within a line a horizontal <space/>, decimal characters (quantity=\"2.5\") or a length; a first-line indent its indent step. A printed rule is <milestone unit=\"section\" rend=\"rule\"/> (its measure inline-size), a printed frame border-style, padding and inline-size on the block it encloses. The frame: every page's main text is set in its work's frame, the body pages' main text block; write no page margins and no centering by margins, and give a drop as a vertical space from the frame's top. Marginalia—side and shoulder notes, call numbers, stamps, a bookplate—are set outside the frame where the scan has them, never in it: <note place=\"margin-left|margin-right|top|bottom\"> (type=\"provenance\" for an owner's or keeper's mark) with measured style, its white to the frame (margin-inline-end on a left note, margin-inline-start on a right one, margin-block-end on a top note, margin-block-start on a bottom one) and its offset along it (margin-block-start from the frame's top on a side note, margin-inline-start from its left on a top or bottom one), read from layout_measure's marginalia rows. The page is the frame and its marginalia with equal, empty margins around them. A page whose only ink is such a mark (a library's barcode label on a blank leaf) sets no main text: the mark alone, as its note at its measured offset from the frame, which the tools place on the scan as the edition sets it (centered on the page), never a vertical space to reach it. Main text taller than the frame shrinks its large vertical gaps, never its type (a frame finding blocks layout_ok). Compare the registered scan, the render and the structure (the render's lines and blocks outlined solid, the registered scan's dashed: gray within a quarter of the body's leading (an owner's mark's, of its own line pitch) and its aligned edges within an em of its type, orange a line pair out of it, green a rendered line the scan lacks, red a scan line the render lacks) with layout_measure's lines; some drift is acceptable. Contents pages: <table type=\"contents\">, the heading before it, one row per entry with four cells—label, numeral, title, page reference—an entry with neither label nor numeral spanning them with cols=\"3\"; PAGE a <row role=\"label\">; each cell's text a <p> with its measured style; dot leaders as printed in one <pc> whose letter-spacing gives their pitch; a continuation line in its title cell. Never join entries or simulate columns with spaces. Printed page numbers are fw type=pageNum; the canvas label is not page text. No text coordinates, text bounding boxes or absolute positioning (a figure or decorated initial still takes its bbox). Returns each finding; it changes nothing.", "parameters": {"type": "object", "properties": {"tei": {"type": "string"}, "features": {"type": "array", "items": {"type": "string", "enum": ["type_changes", "long_s", "letterforms", "spellings", "abbreviations", "running_head", "page_number", "signature", "catchword", "title_pages", "headings", "marginal_notes", "footnotes", "provenance", "columns", "dictionary_entries", "speeches", "stage_directions", "verse", "other_languages", "figures", "initials", "tables", "formulas", "illegible", "corrections", "blank"]}, "description": "Every feature the image of the page shows (empty when it shows none of them): type_changes — a change of type or letter style within the text: italic, roman within italic, small capitals, bold, blackletter, spaced letters, larger or smaller type, another class of typeface, superscript, subscript, underline, a stretch in another ink, encoded <hi rend=\"…\">, or a block set in other type than the page's text (style font-size, font-style, font-weight, font-variant-caps or letter-spacing; rend larger, smaller or a typeface); long_s — the long s (ſ), encoded <g ref=\"#slong\">ſ</g>; letterforms — ligatures and special letterforms other than the long s (ct, st, æ, œ, r rotunda), encoded <g ref=\"#…\">; spellings — historical or nonstandard spellings (vertue, publick, ſhewed), encoded <choice><orig>…</orig><reg>…</reg></choice>; abbreviations — abbreviations and contractions: a macron or tilde for m or n, a raised letter, a brevigraph, the ampersand in &c., a shortened word with its point (Dict., Ham.), encoded <choice><abbr>…</abbr><expan>…</expan></choice>; running_head — a running head, encoded <fw type=\"header\">; page_number — a page or leaf number printed or written by the makers, encoded <fw type=\"pageNum\">; signature — a signature or gathering mark, encoded <fw type=\"sig\">; catchword — a catchword, encoded <fw type=\"catch\">; title_pages — a printed or written title page, encoded <titlePage>; headings — a heading or title, encoded <head>; marginal_notes — the makers' notes or glosses in the margin, beside the text block or above or below it, encoded <note place=\"margin-left|margin-right|top|bottom\">, set outside the frame where the scan has it (<note place=\"margin\"> where its side is not known); footnotes — footnotes, encoded <note place=\"foot\">; provenance — marks this copy's owners or keepers added: a bookplate, an ownership or library stamp, a label, a call or accession number, a barcode, encoded <note type=\"provenance\"> (a stamp in it as <stamp>); columns — text set in two or more columns, encoded <cb/>; dictionary_entries — a sequence of entries, each headed by the word it enters, followed by what is said of it (a dictionary, glossary or word list), encoded <entry> or <entryFree>; speeches — speeches headed by the speaker's name, encoded <sp> with its <speaker>; stage_directions — stage directions, encoded <stage>; verse — verse lines, encoded <l>; other_languages — words or passages in another language than the text around them, encoded <foreign xml:lang=\"…\"> or xml:lang on the word, sentence or quotation; figures — an illustration, diagram, map, photograph, ornament or printer's device, encoded <figure>; initials — a decorated or drop-cap initial, encoded <hi rend=\"initial\"> inside the <w> of the word it begins; tables — a table, encoded <table>; formulas — a mathematical or chemical formula, encoded <formula>; illegible — damaged, faded, blotted or hard-to-read text, or text lost, encoded <unclear>, <damage>, or <gap> (other than reason=\"blank\"); corrections — the makers' own insertions, deletions or corrections (a later hand's are not transcribed at all), encoded <add>, <del> or <subst>; blank — a surface with nothing written or printed on it, encoded <gap reason=\"blank\"/>"}, "regions": {"type": "array", "maxItems": 3, "items": {"type": "object", "properties": {"x": {"type": "integer", "minimum": 0, "maximum": 1000}, "y": {"type": "integer", "minimum": 0, "maximum": 1000}, "width": {"type": "integer", "minimum": 0, "maximum": 1000}, "height": {"type": "integer", "minimum": 0, "maximum": 1000}}, "required": ["x", "y", "width", "height"], "additionalProperties": false}}, "reason": {"type": "string"}, "layout_ok": {"type": "boolean"}, "review_token": {"type": "string"}, "lineation_judgment": {"type": "string", "description": "With layout_ok, when a lineation finding stands: why the measured line counts differ though every printed line end has its <lb/> (a stray mark measured as a line, a block the measurement split differently)."}}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "measure_page", "description": "Measure the canvas's scan for this page: its main text block, its blocks, lines and words (the textblock, block, line and word zones on the scan) and each word of the draft linked to its word zones, all computed by the tools and placed when the page is filed (you never write a zone, a facsimile or a word's facs). Where the witness's document declares no renditions, they are measured on this scan and filed with the page. Changes no word of the draft; returns what was measured: the zones by kind, the words linked, any word left without a zone, and the renditions. Then validate_tei, confirm layout_ok and suggest_revision.", "parameters": {"type": "object", "properties": {}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "check_annotation", "description": "Cross-check the words' annotation against an independent tagger (Stanza, trained on Universal Dependencies treebanks of some 70 languages): the draft, or the fragment or document given as tei. Each <s> is read by the tagger of its language, on the TEI's own tokens, and only the disagreements come back: type (as UPOS), lemma, and UD features where the word class agrees, each with the tagger's confidence. A tagger is often wrong where you are right: judge each, correct the TEI only where the evidence of the page says you were wrong. A language without a model is listed as not checked, and where the tagger is not available at all the answer says so: your annotation stands. languages (field language) checks the xml:lang: identification flags a sentence fastText's lid.176 reads, without its parts in other languages, as another language with more than half the probability; inline gives every <w>, <foreign> and <mentioned> in a language other than the one around it, word by word, with whether UniMorph attests the form in its xml:lang and in the language around it (or that UniMorph has no table for one). Short sentences, names and borrowings are weak evidence; UniMorph's tables hold inflected words, some languages only some word classes (French only verbs), so a form missing from one is weak evidence too.", "parameters": {"type": "object", "properties": {"tei": {"type": "string"}, "lang": {"type": "string", "description": "The language of text that carries no xml:lang. A language code as the TEI writes xml:lang (ISO 639-3, e.g. eng, lat; a BCP 47 tag is read too)."}, "fields": {"type": "array", "items": {"type": "string", "enum": ["type", "lemma", "msd", "language"]}, "description": "Which comparisons to return (all four when left out)."}}, "required": ["lang"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "check_consistency", "description": "Compare the draft's word annotation with every page of this work already transcribed, before or after this one, in any language or script. Records: a form (case-folded, per language) read with different type, lemma or msd (lemma and msd within one type); an inflectional feature written on some words of a type and omitted on others with the same context (feature); words of one type and features whose lemma is their own form beside others whose lemma is not (lemma_form). Each gives every reading with its count and pages, the majority reading by its page ranges, the others page by page with their places [sentence, word] as check_annotation numbers them. Judge each: correct this page where it breaks the work's convention, and record a convention the pages should keep with write_conventions. By default only records this draft takes part in; whole_work returns all. Then, for fields type, lemma and msd, each form this draft reads in a way the permitted biblio-records do not, in two tiers that never count a word twice: other_witnesses, this work's other editions, issues and copies, each apart by its label, how their markup tags the form (lemma, type, msds when compared; occurrences, years); then site, every other work's markup, with how many works, and with each reading the lexico-records of its language, lemma and type, with their forms. Each is given only where this draft disagrees with it. Both are weaker evidence than this copy's own pages: another edition may spell otherwise, another work, of another time or place, may spell another word so.", "parameters": {"type": "object", "properties": {"tei": {"type": "string"}, "lang": {"type": "string", "description": "The language of text that carries no xml:lang. A language code as the TEI writes xml:lang (ISO 639-3, e.g. eng, lat; a BCP 47 tag is read too)."}, "fields": {"type": "array", "items": {"type": "string", "enum": ["type", "lemma", "msd", "feature", "lemma_form"]}, "description": "Which kinds of record to return (all when left out)."}, "whole_work": {"type": "boolean", "description": "Every record of the work, not only those this draft takes part in."}}, "required": ["lang"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "lookup_morphology", "description": "A form's inflectional analyses in UniMorph (lemma and features in the UniMorph schema; some 170 languages). Exact form only: look up the form as printed, then its lowercase.", "parameters": {"type": "object", "properties": {"form": {"type": "string"}, "lang": {"type": "string", "description": "A language code as the TEI writes xml:lang (ISO 639-3, e.g. eng, lat; a BCP 47 tag is read too)."}}, "required": ["form", "lang"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "identify_language", "description": "The three likeliest languages of a text by fastText's lid.176, with probabilities. Short texts, names and transliterations are weak evidence.", "parameters": {"type": "object", "properties": {"text": {"type": "string"}}, "required": ["text"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "segment", "description": "ICU's word boundaries in a text (UAX #29, with dictionaries for scripts written without spaces: Thai, Lao, Khmer, Burmese, Chinese, Japanese), each segment with its kind (letter, number, kana, ideographic, none). Advisory: the printed text decides the <w>s.", "parameters": {"type": "object", "properties": {"text": {"type": "string"}, "lang": {"type": "string", "description": "A language code as the TEI writes xml:lang (ISO 639-3, e.g. eng, lat; a BCP 47 tag is read too)."}}, "required": ["text", "lang"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "expand_abbreviation", "description": "Gnorium's lexicon of abbreviations already expanded in hand-checked TEI: every expansion of exactly this abbreviation (as printed, with its periods) in this language, each with the sentences it was read in, beside context, the sentence it is in here. Empty when the lexicon has none.", "parameters": {"type": "object", "properties": {"abbr": {"type": "string"}, "lang": {"type": "string", "description": "A language code as the TEI writes xml:lang (ISO 639-3, e.g. eng, lat; a BCP 47 tag is read too)."}, "context": {"type": "string"}}, "required": ["abbr", "lang"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "segment_lines", "description": "This canvas image's text lines by Kraken's line segmenter (BLLA), in reading order: each line's box (0–1000), the points its letters stand on, and a confidence. Evidence for you to judge, never written into the TEI.", "parameters": {"type": "object", "properties": {"region": {"type": "array", "items": {"type": "integer", "minimum": 0, "maximum": 1000}, "minItems": 4, "maxItems": 4, "description": "[x, y, width, height] in the 0–1000 space: only the lines whose middle lies in it (the whole page is read either way)."}, "direction": {"type": "string", "enum": ["horizontal-lr", "horizontal-rl", "vertical-lr", "vertical-rl"]}}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "detect_regions", "description": "This canvas image's regions that are not text (figure, initial, rule, frame) by SAM 2: each region's kind, box (0–1000) and score. Evidence for you to judge, never written into the TEI.", "parameters": {"type": "object", "properties": {}, "required": [], "additionalProperties": false}}}, {"type": "function", "function": {"name": "ocr_lines", "description": "Each of segment_lines' lines read by PaddleOCR and by Kraken (CATMuS-Print), with confidences and how far the two agree: a cross-check of your reading, not a reading. langs are the page's languages (ISO 639-3). Evidence for you to judge, never written into the TEI.", "parameters": {"type": "object", "properties": {"region": {"type": "array", "items": {"type": "integer", "minimum": 0, "maximum": 1000}, "minItems": 4, "maxItems": 4, "description": "[x, y, width, height] in the 0–1000 space: only the lines whose middle lies in it (the whole page is read either way)."}, "direction": {"type": "string", "enum": ["horizontal-lr", "horizontal-rl", "vertical-lr", "vertical-rl"]}, "langs": {"type": "array", "items": {"type": "string"}, "minItems": 1}}, "required": ["langs"], "additionalProperties": false}}}, {"type": "function", "function": {"name": "done", "description": "Finish with the reply to the query. Cite evidence markup identifiers; state uncertainty and any suggested revision IDs.", "parameters": {"type": "object", "properties": {"reply": {"type": "string"}}, "required": ["reply"], "additionalProperties": false}}}]