Quick Reference

A conforming output is a Markdown rendering of a document that satisfies every MUST of this protocol, optionally accompanied by a sidecar that satisfies The Sidecar. A producer conforms when every output it emits conforms.

The output MUST represent the document as rendered: what a careful reader extracts from the pages, semantic structure over visual appearance, with text transcribed verbatim — typos included.

Every piece of printed content MUST survive into the output (never drop), and the output MUST NOT contain a label, value, span, header, structure or relationship that the page does not print (never invent).

Every element of the output MUST be attributable to the place on the page it was read from. When boxes are requested, every Markdown element MUST have a box, and every box MUST cover exactly the printed content of its element.

A document is represented as a sequence of pages, each a sequence of elements in reading order. Every element MUST have exactly one class from the table below and MUST be rendered in Markdown as that class prescribes.

The output MUST be one Markdown document for the whole document, with pages in order separated by a line holding only ---, and each page's elements in reading order. Where Markdown has no syntax for a construct, the output MUST use the HTML element this protocol names, and MUST NOT invent a notation of its own.

A box is the region of the page that an element's printed content occupies. Boxes MUST be given at exactly the granularity that was requested, and at that granularity every element MUST have one box. A producer MUST NOT return several granularities in place of one.

When boxes are requested, the output MUST include a sidecar: a structured file that lists, page by page, every element with its class, its Markdown and its box. The sidecar is the source the Markdown is rendered from; the Markdown MUST be exactly the concatenation of the sidecar's element Markdown in reading order, each element preceded by its inline box anchor, with pages joined as Markdown Rendering requires.

Four situations have more than one conforming output, depending on what the producer was asked to produce. Working default: the narrower reading — fragments separate, graphics described, everything printed recorded — which conforms under either setting. Proposed resolution: make each an explicit, named option of the request, and judge the output under the options it was produced with.

Where the protocol accepts two renderings of a situation, the rendering chosen at the first page that forces the choice MUST be used on every later page of the same document. A document MUST NOT mix the accepted renderings of one situation.

A table that runs across a page break MUST be rendered as one fragment per page, each on its own page, complete and exactly as that page prints it. Fragments MUST NOT be joined. In the sidecar, each fragment MUST carry two continuation flags — continues from the previous page and continues onto the next page — decided at document level, from the pages on both sides of the break.

Whether a document is a slide deck or a paged document, and which sheet is its cover, MUST be decided once from the whole document and applied to every page. Only the cover MAY carry a Title element.

A new page MUST NOT restart the heading hierarchy. A heading MUST be given the depth its place in the document implies, and a top-level heading MUST appear only where the page prints one.

Whether a margin band is page furniture MUST be decided once for the document. A band that is page furniture on one page MUST be page furniture — a Page-header or Page-footer, never body text and never a heading — on every page it appears on.

Every meaningful visible region of a page MUST be one element with one class, one position in the page's reading order and — when boxes are requested — one box. Printed content covered by no element is dropped content, and makes the output non-conforming.

A Picture or Table element MUST contain the complete object and nothing else. Its caption, its title, its source note and nearby prose MUST NOT be absorbed into it; each is an element of its own.

Every element MUST have exactly one of the seventeen classes below. Its layout class MUST match the class of the corresponding Markdown element defined in Elements. A class MUST NOT be chosen merely to complete a page: where the class is genuinely unclear, the output MUST NOT assert a class the page does not support.

Every printed string of a court caption MUST be transcribed, the left column as one block of lines and then the right column, each printed line on its own line. Blank party and case-number lines MUST be [blank]. The § divider and the rules MUST NOT be transcribed, and the caption MUST NOT be rendered as a table.

A page reproduced inside another page MUST be rendered as the wrapper's furniture followed by the embedded page in its own reading order. Both page numbers MUST be kept, each with its own region, and titles inside an embedded form screenshot MUST NOT be promoted into the wrapper's outline.

A visible region MAY be left out of the output only on the narrow grounds listed below, and a page MAY be declared unrepresentable only when the problem is the page itself. Content that is merely difficult to read MUST NOT be left out on either ground.

A region's class MUST be settled before its content is rendered, because the class decides which rule governs the rendering. A region rendered under the wrong class is non-conforming however faithfully its text is transcribed.

Printed gutter line numbers MUST NOT appear in the Markdown: the body is rendered as continuous paragraphs, as if the margin were blank. When line numbers are requested, each page MUST carry a line_numbers list that links every printed number to the body line it stands beside.

The elements of a page MUST form one reading-order sequence that includes every element of the page exactly once — no element left out, none placed twice. The sequence MUST follow the order in which a reader reads the page, not the elements' geometric position.

Every printed navigation tab of a slide MUST be transcribed, in printed order, as furniture, and a highlight that identifies the active section MUST be kept with the highlight convention. Labels printed on a photograph MUST be kept with the photograph's description; purely decorative background art MUST NOT be described.

Each stat tile MUST be rendered as one printed label–value pair, in tile reading order, under the panel's heading where the panel prints one. Units and qualifiers MUST stay attached to their value.

Dot leaders MUST NOT appear in the output. The label and the value on either side of a leader MUST be kept exactly as printed; the dots themselves produce nothing.

One printed block set on several lines MUST stay one element, with a plain newline between its printed lines — no trailing spaces, no <br>. The same spelling MUST be used for every multi-line block in the document. Inside a table cell, where a newline cannot be written, the break is <br>.

A printed section separator MUST be rendered as <hr/>, at its place in reading order. --- MUST NOT be used for it: --- is reserved for page boundaries. A rule MUST NOT be inserted where the page prints only whitespace.

Each printed marginal note MUST be rendered as a separate block placed immediately after the element it annotates. Where the element it belongs to cannot be established, the note MUST still be kept as a separate block and MUST NOT be merged into body text.

A numbered line that stands alone as a title with body text beneath it MUST be rendered as a heading, one level deeper per level of numbering. A numbered line that is itself the content — including a bare number printed above its paragraph — MUST be rendered as a list item carrying that content. The printed number MUST be kept exactly as printed, on the same line as its text.

A printed page number (the folio) MUST be kept in the header or footer band it is printed in, verbatim, and MUST NOT appear as a loose element in the body. When the printed page number is requested, it MUST also be recorded on its own for the page, as printed, and MUST NOT be inferred for a page that prints none.

A standalone structural title that organises what follows MUST be rendered as a heading. Run-in labels, repeated item labels, captions and page furniture MUST keep their own roles and MUST NOT be rendered as headings. Font size alone MUST NOT decide whether a line is a heading.

A run-in label and the sentence, value or URL that follows it on the same line MUST be rendered as one paragraph, with the label's printed emphasis preserved. The label MUST NOT be rendered as a Markdown heading.

The wrapped lines of one title MUST be joined with spaces into one heading. An article number MUST stay with its title. A company name printed above a statement or document title, exhibit labels, units lines and other furniture MUST be kept separate from the title; a title and subtitle MAY be split only where the document establishes distinct roles for them.

A table of contents, table of authorities or index MUST be rendered as a list: one item per entry, nested as the page indents it, each item the label then its page number(s) exactly as printed, separated by a space. It MUST NOT be rendered as a table.

On a transcript page, each speaker turn MUST be rendered as one paragraph that begins with its printed marker (Q, A, THE COURT:), with its wrapped lines joined. Printed line numbers MUST NOT appear in the Markdown; when line numbers are requested they are carried in the page's line_numbers map, exactly as for Pleading Line Numbers.

Inline marks MUST be written with exactly the markup in the table below. Where Markdown has no syntax for a mark, the output MUST use the HTML element named here and MUST NOT use a notation of its own.

Bold MUST be written **bold** and italic *italic*, in Markdown, except where the Markdown delimiter cannot close, where <strong> / <em> MUST be used. They MUST be applied only where the printed glyphs are heavier or slanted than their neighbours.

Displayed mathematics MUST be written as LaTeX inside $$ … $$. Its printed equation number MUST follow the closing delimiter as plain text in the same formula block: $$ … $$ (1).

A drop cap MUST be joined directly to the rest of its word, with no space, line break or styling added. The output MUST NOT mark the letter as bold, enlarged or otherwise distinct.

A character the page prints is content, even when Markdown would read it as syntax. It MUST be escaped so that it cannot open a span — \*\*, \_, \| — and it MUST NOT be deleted to avoid the collision.

Highlighted text MUST be wrapped in <mark> carrying the highlighter colour as an attribute: <mark style="background-color: yellow">the highlighted phrase</mark>.

Every equation MUST be written as LaTeX inside $$ … $$, inline ones included. Single-$ delimiters MUST NOT be used.

A word broken across a line end MUST be joined, and a hyphen that exists only for the break MUST be removed. A hyphen that belongs to the word MUST be kept. Where the two cannot be told apart, the visible hyphen MUST be kept.

Hyperlinks and email addresses MUST be transcribed as raw text. Link syntax — <email>, [url](url) — MAY be used only when link extraction was explicitly requested.

A quoted or inset region MUST be flattened: its content is rendered as ordinary text, headings and tables included. The output MUST NOT add > block-quote markers.

Text set in small capitals MUST be transcribed as uppercase letters. The output MUST NOT add small-caps styling, and MUST NOT substitute the text layer's mixed case for the capitals the page shows.

Raised and lowered text outside maths MUST be written with <sup> and <sub>. A Unicode superscript or subscript character MUST NOT be substituted for the tag.

Underlined text MUST be wrapped in <u>…</u>. The output MUST NOT use <ins>, and MUST NOT use _ text _.

Text whose line breaks carry meaning — verse, song text, deliberately broken lines — MUST be written one printed line per source line, separated by a plain newline, and MUST NOT be reflowed into running prose. No trailing spaces, no <br>, no fenced block: the same spelling as Line Breaks Inside a Block.

Text cut off at the page edge MUST be rendered as the characters actually visible. A clipped word MUST NOT be completed, and an ellipsis or marker that the page does not print MUST NOT be appended to it.

A field that carries only an e-signature tag is unsigned and MUST be rendered as a blank, [blank]. A tag that prints a word — SIGN, INITIAL — MUST be rendered as that word, separately beside the field, and MUST NOT be rendered as a signature or as the field's value.

A handwritten note on a printed page MUST be rendered as [handwritten: legible words], in a block of its own next to the content it annotates. Its legible words MUST be kept; words that cannot be read MUST NOT be reconstructed.

An icon that carries meaning MUST NOT be dropped. It MUST be rendered as a bracketed semantic tag in the icon's position: one lower-case word naming what the icon means — [yes], [no], [locked] — with : degree where the icons are graded ([water: 3], [sun: partial]), and [icon: shape] naming the shape where the meaning is unclear ([icon: sun]). The vocabulary is open: the word is whatever names the meaning on this page.

Rotated text printed along a side margin MUST be rendered once, as a furniture block of its own outside the body flow, with its words kept — branding included. It MUST NOT be tagged as a watermark unless it actually overlays the content.

Where two printed strings overlap, each legible string MUST be rendered whole, in a block of its own, with the stamp first. Legible fragments MUST be kept; characters hidden by the overlap MUST NOT be reconstructed.

A QR code or bar code MUST be rendered as an image reference whose alt text holds the decoded content and whose target holds the literal label: ![decoded content](QR Code), ![decoded content](Bar code). The code MUST also be an element of its own, carrying the decoded value and a box around the code.

Each contiguous redacted span MUST be rendered as one [REDACTED] marker, in the position the suppressed text occupied. Visible text and any printed exemption code MUST be kept; the hidden characters and their count MUST NOT be inferred.

A seal or round stamp MUST be rendered as [seal: printed text], holding all of its legible rim and centre text in reading order, in a block separate from body prose. The bare [seal] MUST be used only when none of its text is legible.

A handwritten signature MUST be rendered as a bracketed placeholder: [signature:{name}] when its glyphs are genuinely legible, the bare [signature] otherwise. A name MUST NOT be guessed.

Ink applied over or beside the printed page — overprinted, boxed or rotated, or in a different ink — MUST be rendered as [stamp: its words], exactly as printed, at its place in reading order. A stamp with no legible words MUST be rendered as [stamp]. The page's own printed text is never a stamp, however official it looks.

Text colour MUST be carried only when it is the only thing distinguishing some text from the rest, written as <font color="NAME">…</font> around exactly the coloured words, with the nearest HTML colour name. Colour that is decoration MUST NOT be rendered.

A watermark MUST NOT be deleted and MUST NOT be left inside the body text. It MUST be rendered once per page that prints it, as its own element at the top of that page's content: [watermark: DRAFT].

A top-level bulleted item MUST be rendered with the * marker and a sub-item with the - marker. The two markers carry the nesting and MUST NOT be interchanged.

A whitespace-aligned block with columns MUST be rendered as a table. Its column tracks MUST be read off the widest row, every value MUST be placed in the track it sits in, and the block MUST NOT be rendered as preformatted text.

When a chart prints its data grid beside or beneath the canvas, the output MUST render that grid exactly as printed — same orientation, same headers, same row order — and MUST NOT render any other table for the chart.

A printed grid with no values entered MUST be rendered as a table, with its headers and every printed empty row as a row of empty cells. It MUST NOT be skipped, and MUST NOT be reduced to its headers alone.

The output MUST NOT contain a table that is a header with no body. A header row MUST either be rendered together with its body or, where no body can be joined to it, rendered as ordinary text — never as an orphan header row standing as a table.

A region MUST be rendered as a table when, and only when, its rows and columns encode a relation — when the position of a value in the grid is what gives the value its meaning. A region that encodes no relation MUST NOT be rendered as a table.

A full-width band inside a grid — a date line, a section line, a change of shading — MUST NOT end the table. The rows above and below it MUST stay in one table, and the band MUST be rendered as one row whose single cell spans every column.

One long series printed as side-by-side blocks SHOULD be rendered as one table per printed block, each carrying its own copy of the headers, including repeated blocks inside one frame. A checker MUST accept source-supported joined equivalents without losing or reassigning records.

A continuation marker — (cont'd), (continued), Table 5 — … (cont.) — is printed text and MUST be kept verbatim in the title. The fragment on this page MUST be rendered as a table in its own right.

Every printed occurrence of a table MUST be rendered, in reading order, each as its own table — including a table the page prints twice. Transcription is not summarisation: page content MUST NOT be deduplicated.

Bands over stacked grids resolve on one question: is there a shared row spine? Grids sharing one spine MUST be rendered as one table. Grids with no shared spine and no column headers MUST be rendered as one table per band, with the band as that table's header row.

A table too wide for the page, printed as stacked column blocks, MUST be rendered as one table per printed block, each keeping its repeated row-label column. The blocks MUST NOT be re-widened into a single wide table.

Within one page, a column-header row printed again after data rows MUST close the table and open a new one at the repeat. Header repetition is the table-boundary signal.

Boxes sitting side by side with their own titles and their own schemas MUST be rendered as separate tables — even when their rows line up perfectly. Neighbouring tables MUST NOT be woven together, and a small side table MUST NOT be folded into the middle of another table's body.

Fragments that share a header row or a label column printed only once MUST be rendered as one table. The missing header is the proof that the second fragment was never meant to stand alone.

The output MUST transcribe what is printed and add nothing. A column or a header MUST NOT be dropped because it is empty, plainly styled or easy to miss, and a label MUST NOT be invented for a cell that has none.

A header line with a rule beneath it, standing over two or more self-standing leaf labels, is a real tier and MUST be rendered as its own header row. Fragments that simply stack with no border and read as one phrase are a wrapped label and MUST be joined with a space into a single header cell.

Every printed header tier MUST be rendered as its own header row, with exact colspans. A group header MUST span only its data columns; the row-label stub takes a rowspan across the header rows instead.

A financial period stack MUST be rendered as its printed tiers: period groups over their own date leaves, an apex qualifier above them, a stacked month and year joined into one leaf label, and the units note kept verbatim as the corner header cell.

A units line under the header MAY be rendered as a second header row, or folded into the header label above it ("Premium ($000)"). Both forms conform, and a document MUST use one of them throughout (see Choices Made Once Per Document). A units line MUST NOT become the first body row.

A currency glyph or a separator typeset in a column track of its own SHOULD be rendered in the same cell as the value it belongs to — $ 277.09, $1,200 — $1,800. A marker numeral such as (i) or (ii) SHOULD be rendered in a column of its own.

Every row of a table MUST carry the full column count. Where the header and the body disagree, the short side MUST be padded with empty cells, one per missing position, and MUST NOT be padded with a duplicated or invented value.

A genuinely printed blank row inside a table MAY be rendered as an empty row or left out; both forms conform, and a document MUST use one of them throughout. A blank row that is the only thing separating two groups of rows SHOULD be rendered as an empty row.

A small grid printed inside one cell of a table MUST be rendered by subdividing that cell into the outer table's own grid: the inner grid's cells become rows and columns of the outer table, and the neighbouring cells span them. The output MUST NOT contain a nested <table> element, and MUST NOT flatten the inner grid into text.

Independently printed records MUST each be rendered as their own row, with rowspan only on cells the page genuinely merged. One prose cell MUST stay one cell, with <br/> between its title and its paragraph. In a boxed row holding lists in several columns, lines that pair up one-to-one MUST be separate rows, and a list whose neighbours print a single value for the whole list MUST stay in one cell, <br/>-joined, beside that value.

A rowspan MUST be used only where the page genuinely merged the cell. A value printed k times at k distinct vertical positions MUST be rendered as k cells; repeated content MUST NOT be merged.

Where a table's row labels are indented to show nesting, every label cell MUST carry its absolute level as a data-level attribute: <td data-level="2">iPhone</td>. Level 0 is the leftmost label position in the stub column, 1 the next indent step, and so on — counted from the left edge of the column, never from the row above.

In a total or subtotal row, the label cell MUST be a <th> and the value cells MUST stay <td>. The label cell MUST carry the row's absolute level as data-level, read as for the stub hierarchy — a top-level total is data-level="0". Label and values MUST be kept exactly as printed, bold only where the page prints them bold.

Every number in a cell MUST be transcribed exactly as printed — parentheses, currency dashes, separators and change markers included. The output MUST NOT carry a normalised twin of the value beside it, and MUST NOT replace the printed form with a normalised one.

Inside a table, bold, italic and underline MUST be marked per cell, and only where the printed glyphs of that cell are heavier, slanted or underlined. A whole column or a whole header row MUST NOT be styled on principle. Inside an HTML table cell, bold MUST be written <strong>, italic <em> and underline <u>.

The text of a cell MUST be what the rendered page shows — not the text layer, not OCR, and not what the value ought to be. A character that cannot be read at any zoom MUST be written as [illegible], never as a plausible guess.

A cell whose only content is a coloured fill MUST be rendered as an empty <mark> carrying the colour: <td><mark style="background-color: red"></mark></td>. The colour MUST be the nearest HTML colour name, never the meaning a legend assigns it, and the cell MUST NOT be left blank.

Handwriting over a printed value MUST keep both, the struck-through printed value first: ~~60~~ 55 — the same markup as a printed redline replacement. A stamp over a table MUST keep its text and MUST be rendered outside the table, as page-level content; it MUST NOT be assigned to a cell.

A footnote marker MUST stay in the cell of the value it is attached to. An em-dash alone in a cell is a value and MUST be kept; a cell printed blank MUST stay blank. A cell whose content is a symbol or icon MUST be rendered as a bracketed transcription of what the mark asserts, and MUST NOT be replaced by an emoji.

A cell holding a struck value and its replacement MUST render both, struck first — ~~1.316~~ 1.335 — and MUST NOT merge them into one figure. The ~~ MUST cover exactly the glyphs struck on the page: 1.~~316~~ 335 when only the digits are crossed out.

Content printed below or beside a table MUST NOT be dropped, even when it sits outside the ruled frame. Where it continues the table's ruled grid it MUST be rendered as rows of that table; where it stands apart as its own unbordered block it MUST be rendered as its own small key-value table.

A table footnote MUST be rendered outside the table, as footnote text placed after it — even when the page prints it inside the table's ruled border. It MUST NOT be rendered as a row of the table. It retains the Footnote class and MUST NOT be wrapped in <caption>. This is a deliberate exception to the frame test.

For a table-only title, the frame and the fill decide. A band that shares the table's border or fill MUST be rendered as a header row inside the table; a table-only title outside the ruled frame MUST be rendered as <caption> inside the table HTML. Table and chart captions MUST be the first or last content inside <table> according to reading order and MUST appear once.

A rotated table MUST be read in its own orientation, never in page orientation. A diagonally split corner cell MUST stay one cell holding both labels separated by a slash — Region \ Year — the label across the top first.

Every row MUST carry the full column count. The jagged half MUST be padded with individual empty cells — one <td></td> per missing position. A colspan or rowspan over the empty region MUST NOT be used, and a 0 or any other value MUST NOT be invented.

Fill-in fields and checkbox regions are key-value content, not tables: they MUST be rendered as the key-value list of The Key-Value List and MUST NOT be rendered as a table. The output MUST NOT contain a two-column key/value table that the page does not print as a ruled grid.

Field groups that repeat without a printed heading MUST be kept as anonymous groups in printed order, separated by a blank line in the Markdown. The output MUST NOT invent a heading, label or positional id — no Vehicle 1, no Child 2 — for a group or field the page does not name.

A field's id is its printed designator, written exactly as printed in escaped brackets outside the bold label. Its label is the printed caption, verbatim. A label MUST NOT repeat the id, and a caption MUST NOT be written as an id.

Fillable form content MUST be rendered as a Markdown key-value list, not as a table. Every field MUST be one list item using the - marker.

Entered content MUST be transcribed exactly as it appears on the page. A value MUST NOT be rounded, reformatted, computed or summarised.

A sentence with fill-in gaps MUST be kept intact, with each gap rendered as [blank] at its printed position. A hint printed with a blank MUST be kept next to that blank, once.

A checked box MUST be rendered as [x] and an unchecked box as [ ]. An unchecked box MUST be transcribed explicitly and MUST NOT be dropped.

Each checklist item MUST be rendered box-first — the box before its printed label — and items MUST be nested under their group label. A group-level box MUST stay on its group's line. A form's group labels MUST NOT be promoted to document headings.

Each printed entry field MUST be rendered as one line. An empty field MUST be rendered as [blank] and MUST NOT be omitted.

A question answered by ticking one or more of several printed boxes MUST be rendered as one group: the caption line, then one nested box-first line per option. Every printed option MUST be listed, ticked or not.

The text of a printed button, and the caption of a box reserved for a later stamp or seal, MUST be transcribed as text where it sits, in reading order. A reserved box is not a fill-in and MUST NOT be rendered with [blank].

A field printed in a header, a footer or anywhere outside the main form body MUST still be rendered. A field printed above the form MUST be placed before the form content; any other such field MUST be placed after it.

The fields of a form MUST be ordered by the form's printed indexing scheme — its ids — rather than by column-first or row-first geometry.

A printed heading that names a group of fields opens a section. Only a printed grouping MAY get a parent: fields MUST NOT be wrapped in a section or group the form does not print.

Inside a form, a table MUST be used only for a genuine grid: one shared set of printed column headers over repeating rows of the same shape, or a printed row-by-column matrix. Where it is ambiguous, the fields SHOULD be rendered as individual field lines.

A chart MUST be rendered as a chart table that holds every data point the chart draws: every category position × every series, plus every printed callout that belongs to a position. A data point MUST NOT be skipped because it is hard to read; it MAY be left out only when no printed text can locate it or it is genuinely unreadable.

The category labels and series names of a chart table MUST be text printed on the chart — an axis tick, a legend entry, a direct label — copied verbatim. A label MUST NOT carry unit text imported from elsewhere on the page, colour words, chart-type words or an expanded abbreviation, and two dimensions MUST NOT be merged into one label.

Each chart or panel on a page MUST get its own chart table, and every panel MUST be covered (see Cover Every Data Point). Reference ink — labeled averages, trend lines, shaded bands, event callouts — draws no data points and MUST NOT be rendered as data: it gets no row, no column and no value in any chart table. Its printed text MUST still be kept.

Each cell of a chart table holds one of three values. A value printed on the chart MUST be copied character for character and in full. A value not printed MUST be estimated at the mark's visual centre against the mark's own axis. A mark drawn at the baseline, or a labeled position with no mark, MUST be 0.

A bar chart MUST be rendered as a chart table with one value per bar: one row per category, one column per series. A histogram is a bar chart whose category is the bin: a bin the axis does not label MUST still get its row, labeled by its axis value, and a labeled position with no bar MUST hold 0.

Bars and lines on one canvas, or a second value axis, change only how values are read: each series' values MUST be copied or estimated against its own axis. The chart table is laid out as usual — one row per x position, one column per series — and a series name MUST NOT gain its mark type or axis side (no (line), no right axis).

A gantt chart MUST be rendered with one row per mark, giving the mark's label, its start and its end on the time axis — for a milestone, start and end are the same period. A sankey diagram MUST be rendered with one row per flow, giving its source node, its target node and the value it carries. Each family uses its one shape throughout, whatever the chart prints.

When a chart's values are printed only in its legend or as data labels over its marks, those printed numbers are the data: every one of them MUST be in the chart table, copied exactly. A companion series without printed values is estimated as usual — unless no printed label can locate its points, in which case it has no column (see Labels).

A line chart MUST be rendered with one value per series per x position: one row per x position, one column per series, each series named by its legend entry or by its direct label at the line's end. A value printed at a point MUST be copied; every other value is read against the axis. Wiggles between x positions are not positions.

A logarithmic axis changes the reading, not the chart table. Unprinted values MUST be estimated against the printed decade bands, by the mark's position inside its band, and SHOULD be given to no more than two significant figures.

A treemap MUST be rendered with one row per tile: the tile's label and its printed value. Tile area MUST NOT be turned into a value: no axis calibrates it. A map with printed values per region MUST be rendered with one row per printed region–value pair. A shading-only map with a discrete legend MUST give each region whose fill matches one swatch unambiguously that legend class, copied verbatim.

A named scatter point MUST be rendered as one row holding its x value and its y value. Marker series over categories are ordinary series columns. An error bar's centre is an ordinary value, and its whisker span MUST be given with it: the low and high ends read at the whisker caps. A box plot MUST give its median, first and third quartiles, and its whisker ends when they are drawn. Points with no printed name have nothing to locate them and are not rendered as rows.

A stacked segment's value MUST be the segment's own span — the size it draws between its two boundaries — never the cumulative height at its top edge. A waterfall step follows the same rule on a floating segment: its value is the rise or fall the step itself draws. Each segment series is a column; a printed total is a value of its own.

The content of an image MUST be put into the text stream as whatever the image actually is — transcribed text, a table or a diagram. The output MUST NOT contain a file reference or a bracketed placeholder in place of an image.

The output MUST NOT contain an image reference ![caption](path) or a bracketed placeholder such as [The image shows…]. A figure's content MUST be rendered as text, a table or a diagram — whichever the figure actually is.

A logo MUST be rendered as a marker carrying its printed words, [logo: OSTREVANE], at the logo's place in reading order. A logo that prints no legible text MUST be rendered as [logo]. The artwork MUST NOT be described.

A floor plan or architectural drawing MUST use a faithful diagram/<language> encoding, preferring diagram/tikz then diagram/svg. If no suitable language can represent it faithfully, it MUST use a fenced description block. The output MUST preserve visible geometry, labels and printed measurements without inferred relationships or computed values. It MUST NOT be a table.

A diagram or blueprint MUST use a fenced diagram/<language> block when a suitable language can faithfully represent its content. If no suitable language can do so, the output MUST use a fenced description block. A diagram that is really tabular SHOULD be rendered as an HTML table. A diagram MUST NOT be left out of the output.

A component-level circuit schematic MUST use a faithful diagram/<language> encoding, preferring diagram/circuitikz, then diagram/tikz, then diagram/svg. If no suitable language can faithfully represent it, it MUST use a fenced description block. Every component symbol, designator, value, pin, terminal, net label and visibly drawn connection MUST be preserved.

How a map is rendered depends on what it is for. A data map whose values sit in drawn regions MUST be rendered as a Region | Value table — the region names MAY be recognised from the drawn outlines when the page does not print them, and the key column is then marked data-recognised. A map of values placed only by position MUST be described, with its legend transcribed, and no table. A map whose content is geography MUST be described. Every printed label MUST be kept.

A labelled figure MUST use a suitable diagram/<language> encoding when one can preserve its content faithfully, following the language priorities in Diagrams and Flowcharts. A labelled figure that no suitable diagram language can faithfully express MUST be rendered as a fenced description block in the fixed shape: one line saying what the figure is, then one bullet per printed label, verbatim, followed only by what is visibly drawn there, then printed values and notes. The block MUST NOT explain what a label means.

Linear formulae, ions and reactions MUST be written with \ce{...}, and drawn structures, rings and branches with \chemfig{...}. Standalone chemical markup MUST be placed in a fenced latex block. Legible structure MUST be preserved; an unreadable portion MUST be described rather than guessed.

Every string printed on a music page MUST be transcribed as ordinary text. The staves themselves MUST be described, not encoded: each staff region is a fenced description block in the shared fixed shape, and no notes are read into note names, ABC, LilyPond or MusicXML.