OPF

Search docs

Loading documentation…
    OpenPresentation/opf
    Browse docs
    On this page
    View source

    Pyramids, matrices and SmartArt

    Several layouts of the default catalog draw a list as a diagram: pyramid (levels that build on each other, the first entry at the top), matrix (four parts of a whole, or the four cases of two choices, as quadrants), cycle (steps that repeat, as a loop; see Cycles) and chevrons, flow and vertical-flow (stages or steps in order). The PowerPoint export writes them as live SmartArt where PowerPoint lays the SmartArt out exactly as the preview draws it (PowerPoint's designs on its built-in layouts, OPF's own designs on OPF's own layout definitions), and as the same drawing in native shapes everywhere else. A slide on either layout never turns into a stray bullet list: the arrangement has a second design of its own, drawn when PowerPoint's does not hold the entries.

    json
    { "layout": "pyramid", "title": "Our strategy", "items": ["Vision", "Goals", "Programs", "Teams"] }
    json
    { "layout": "matrix", "title": "Four markets", "items": ["Enterprise", "Mid-market", "Consumer", "Public sector"] }

    The arrangement of a region

    A layout record's region takes arrangement: "pyramid" or "matrix" (layout templates). When the region holds one list of one level (no nested entries, no descriptions, no icons, no numbering), composition draws it as that diagram (SlideComposition.arrangements, core arrangements.ts), in the first of these designs that holds every entry:

    Design (design)Pyramid (1 to 8 entries)Matrix
    PowerPoint's (smartart), where the PPTX export can write it as live SmartArt (the domain)Basic Pyramid: level i of n is H/n tall and W·i/n wide, the apex a triangle, the entries centered on their levelsBasic Matrix, for exactly four entries: four rounded squares in a centered square over a diamond, the entries centered
    OPF's own (opf), where PowerPoint's does not hold the entries at the list's size or lies outside the domaina stepped pyramid of rounded bars at the start of the region (level i is i/n of the steps' width), each entry beside its level in one column, start-alignedfour rounded quadrants filling the region two by two (rectangular where the region is), the entries centered
    A grid of cards (cards), for a matrix of other than four entries—one card per entry in the card neutral, in the grid that holds the text largest (one column of wide cards, or a balanced grid), start-aligned; reported as opf/arrangement-adapted (info): a matrix needs four entries

    When none of them holds the entries, or the list has a shape the arrangement does not draw (nested entries, descriptions, icons, numbering, more than 8 levels or 8 cards), the list draws as a list and opf/arrangement-adapted (info) says why. A list that no design holds still fills the region (SH-18 review, for every arrangement): its text takes the largest size from 1.5 times the role size down to the size it has at which the whole list fits the region, and it sits in the middle of the region's height (LH-48's body growth leaves a list alone once it covers half the body).

    List blocks as entries (SH-18 review). A region whose arrangement holds two or more list blocks draws each block as an entry, its list inside, the same way for every count: a cycle's blocks on cards on OPF's ring or loop (whichever holds them larger) with an arrow from each to the next, a matrix's on a grid of cards, a pyramid's beside the levels of OPF's stepped pyramid. Every list keeps its markers, is start-aligned and vertically centered in its card, and all are set at one size, the largest from 1.5 times the role size down to the lists' own size at which every block fits without a broken word. Where a pyramid's levels or a cycle's ring and loops do not hold the lists, the blocks are a grid of cards, with opf/arrangement-adapted; lists that overflow their region keep its composition, which pagination splits. The cards are native shapes (a SmartArt node does not hold a list), and a region with an arrangement keeps its whole cell when LH-48 centers an under-filled body, so the diagram fills it. A matrix never draws an empty quadrant.

    PowerPoint's geometry was measured through COM for its Basic Pyramid (pyramid1) and Basic Matrix (matrix3) layouts:

    PyramidMatrix
    Framethe region's cell, at most 2.5 times as wide as tall (a wider cell centers such a frame)the region's cell
    Nodeslevel i of n (from the top) is H/n tall and W·i/n wide, centered; the sides continue one slope, so the top level is a triangle and every other a trapezoid (DrawingML trapezoid, adj = W/(2n) over the smaller side)a square of side min(W, H) centered in the frame holds a diamond and four rounded squares (roundRect at its default corner) 0.39 of its side, 0.095 in from its edges and 0.03 apart, in the order top left, top right, bottom left, bottom right
    Text rectanglethe top level's whole box; 0.65 of a lower level's width, centered, at its full heightthe rounded square's own text rectangle
    Text margin0.1 of the font size on every side0.3 of the font size on every side
    Lookevery level in the primary color (accent1) with a 1.5 pt outline in the background color (bg1)the quadrants as the levels, over a diamond in the primary lightened 80% (lumMod 20000, lumOff 80000)

    OPF's designs use the same paint (the cards use the card neutral and its border, as every OPF card); text on a node is in whichever of the background and the text color reads better on the primary, text beside a level or on a card in the slide's text color.

    The text of every node is one size, in every design: the largest from 1.5 times the list's role size (36 pt on a 16:9 slide) down to the role size (24 pt) at which every entry fits its rectangle without breaking a word, and the ink of every line stays inside its node's outline (the top of a pyramid level is narrower than its base). The list's own size counts when it is smaller (a crowded slide), never a size below it. The arrangement fills its region, so the growth a lone list takes when it fills too little of the body (LH-48) is not the size it keeps. Centered lines are centered without the space they broke at, as PowerPoint wraps them.

    Trees and set overlap (SH-21)

    Use org-chart for reporting lines, hierarchy for a classification or work breakdown, and venn for overlapping sets. Their geometry is shared between the preview and editable native PowerPoint shapes. Tree connectors follow the authored parents. Hierarchy uses paired rounded cards with a 90% opaque foreground; Venn uses 50% opaque circles so overlaps remain visible.

    A tree is one list of up to eight short entries in preorder. The first entry has level zero (explicitly or by default); each child supplies its zero-based level. A node may return to any earlier parent depth, but may not skip a depth or introduce another level-zero root. For example:

    json
    { "layout": "org-chart", "title": "Give each team a clear owner", "items": [
      { "text": "Lead", "level": 0 },
      { "text": "Build", "level": 1 },
      { "text": "Web", "level": 2 },
      { "text": "App", "level": 2 },
      { "text": "Ship", "level": 1 }
    ] }

    An org chart's assistant (a chief of staff, an executive assistant) is an entry with "assistant": true, written among its manager's reports at their level. As in PowerPoint's organization chart, it hangs off the line below its manager, one row down and to the right of that line, joined to it by a short elbow; the reports start a row lower, so an assistant costs the height of one more level. A manager may have several assistants, one a row. An assistant has no reports of its own, the root is never one, and it counts toward the eight nodes. hierarchy has no assistant position: a hierarchy, or any other arrangement, with an assistant keeps the complete list with opf/arrangement-adapted.

    json
    { "layout": "org-chart", "title": "Give the pilot lead a chief of staff", "items": [
      { "text": "Pilot lead", "level": 0 },
      { "text": "Chief of staff", "level": 1, "assistant": true },
      { "text": "Delivery", "level": 1 },
      { "text": "Evidence", "level": 1 }
    ] }

    Use venn with one flat list of one to seven set labels. One to four circles hold their labels inside. Five to seven sets use enlarged editable circles with outside labels fitted from the actual font measurements. This OPF adaptation preserves circle overlaps and set order while filling more of the body at the previous type size or larger. Basic Venn native measurements remain a separate, unqualified live reference. When the overlap cannot keep every label readable, the complete set labels use opaque cards or a list with opf/arrangement-adapted. Venn fallback lists try a bounded whole-word reading measure of70characters at the retained type size. If the complete list cannot fit that measure in the body, opf/text-overflow reports the refusal; strict composition rejects it. Explicit pagination or shorter labels resolves that limit. Entry order identifies the sets; the diagram implies no numeric overlap values.

    json
    { "layout": "venn", "title": "Three conditions for a useful pilot", "items": ["Clear scope", "Evidence", "Shared goal"] }

    To name where all the sets meet, which is usually the point of the slide, add one last entry with "overlap": true. It labels the common overlap of two to seven sets and is not a set itself: it is drawn on an opaque pill in the slide's background color with the primary outline, over the circles, at the size of the set labels. The pill sits on the center of the common overlap; where a set label is in the way, it moves up or down the shortest distance that clears every set label while it still reaches into the overlap, and its text wraps onto more lines if that is what keeps it clear. When no position holds it at the list's size or larger, the complete list is kept with opf/arrangement-adapted; a Venn of five to seven sets with an overlap label never falls back to cards, since a card holds a set, not their overlap. More than one overlap label, one that is not last, one with fewer than two sets, or one in another arrangement keeps the complete list with opf/arrangement-adapted too.

    json
    { "layout": "venn", "title": "Where the pilot earns its keep", "items": [
      "Clear scope", "Evidence", "Shared goal",
      { "text": "Ready to scale", "overlap": true }
    ] }

    The arrangement keeps the previous fitted list size or grows. A malformed tree, unsupported forest, nested Venn, descriptions, icons, numbering, too many entries or labels that do not fit keeps the complete list with opf/arrangement-adapted; no entry is dropped or silently reparented. Use a diagram for relationships outside the single-root model. These three layouts currently export as editable native shapes; live SmartArt remains outside their qualified domain. An assistant is a rectangle like the other nodes and its elbow two line segments; the overlap label is a rounded rectangle (roundRect, fully rounded) with its text in a text box over it. Each node's tag records assistant or overlap, so fromPptx puts the field back on its entry. Import recognizes native Organization Chart (orgChart1), Hierarchy (hierarchy1) and Basic Venn (venn1), preserving tree depths. Native assistants, forests or labeled edges use the reported Mermaid hierarchy adaptation instead. A drawing with additional labels beyond the data entries keeps its overlay drawing and reports the adaptation, preserving that extra text.

    The trees and sets example contains the three layouts, an org chart with an assistant and a Venn with an overlap label.

    Chevrons and flows

    Three more layouts draw a list of stages or steps in order (SH-17): chevrons (a row of chevrons, each leading into the next), flow (boxes joined by arrows in one row) and vertical-flow (the same as a column, beside the title). Each is a region's arrangement of the same name, with the same rules as a pyramid's: one list of one level, 1 to 8 short entries, one size for every node, never below the list's role size.

    json
    { "layout": "chevrons", "title": "How we work", "items": ["Plan", "Build", "Test", "Ship"] }
    Chevrons (Basic Chevron Process, chevron1)Flow (Basic Process, process1)Vertical flow (Vertical Process, process2)
    Framethe region's cellthe region's cellthe region's cell
    Nodesn chevrons of width w = s(n) W, height min(0.4 w, H), each 0.9 w after the last (the point of one in the notch of the next), the row centeredn boxes of width w = s(n) W, height min(0.6 w, H), each 1.4 w after the last, centeredn boxes of height h = s(n) H, width min(1.8 h, W), each 1.5 h below the last, centered
    ShapeDrawingML chevron at its default adjustment (the point half the height deep)roundRect at adjustment 10000the same
    Arrowsnonea rightArrow (stem 0.6 of its height, the default head) 0.212 w long and 0.248 w tall, 0.1 w after the box before it, on the frame's middlea rightArrow turned 90 degrees, 0.45 h wide and 0.375 h long, in the middle of the gap
    Text rectanglethe chevron's own, between its notch and its pointthe rounded rectangle's ownthe same
    Text margin0.315 of the font size at the start, 0.105 on the other sides0.3 of the font size on every sidethe same
    Lookevery chevron in the primary with a 1.5 pt outline in the background colorthe boxes as the chevrons; the arrows in the primary lightened 40% (lumMod 60000, lumOff 40000), without an outlinethe same

    PowerPoint fits these layouts with a binary search on one scale and keeps the last step that fits, so a node is not the closed form W / (0.9 n + 0.1) wide but a multiple of 1/8192 of the frame that depends only on the node count: s(n) (core SMART_ART_SCALES) is, in 1024ths of the frame, 1024, 538, 365.5, 276.5, 222.5, 186, 160 and 140.25 for 1 to 8 chevrons, 1024, 426.5, 269, 196.75, 155, 128, 108.875 and 94.75 for boxes in a row, and 1024, 409.5, 256, 186, 146.25, 120.375, 102.375 and 89 (of the frame's height) for boxes in a column. Two chevrons at 800 pt are 420.31 pt wide where the closed form gives 421.05. One node squeezed by the other side of the frame (a chevron in a frame less than 0.4 times as tall as wide) is 1023/1024 of the frame and as tall as it. The COM sweep measured 1 to 10 nodes of each in 20 frames from 250 x 120 to 800 x 540 pt; core places every node and arrow within 0.0001 pt of PowerPoint in all of them but those where PowerPoint's 8 pt test letter did not fit a node (a node a few points tall), whose text made the node grow.

    PowerPoint's design is drawn where it holds the entries at the list's size and lies in the domain; else OPF's own design of the arrangement, filling the region:

    OPF's design keeps PowerPoint's proportions, so the two read as one family: a chevron 0.9 widths after the last (its point at most 0.2 of its width deep), a box 0.4 widths after the last with PowerPoint's arrow (0.212 x 0.248 of a box's width) in the room, a vertical box half its height below the last with a down arrow 0.45 x 0.375 of its height. The entries go into one to four rows (a vertical flow: columns), whichever holds the text largest (one line on a tie), as evenly as possible, the longer lines first. The text is one size for every node, the largest from 1.5 times the list's role size down to the role size at which every node holds its entry; then every node takes the height that text needs with its margins (0.15 em above and below a box's text; PowerPoint's proportion at least: a chevron 0.4 of its width, a box 0.6, a vertical box 1/1.8), and the whole is centered in the region. Only where none of these holds the entries at the list's size does OPF draw a compact form of the same design (rooms of two gaps, the arrow 0.6 of the room, a chevron's point 0.12 of its width deep, no least height), so a list is left a list only when no design holds it. Every row spans the region: a row of fewer entries widens them (a column of fewer lengthens them).

    • chevrons: each row a run of chevrons, left to right; a wrapped row breaks after a chevron's point and the next row starts again at the left, full width, so the points carry the order;
    • flow: the rows snake: the second runs right to left under the first, joined by a down arrow under the end where the first stops, the third runs left to right again, and so on;
    • vertical flow: the columns snake the same way: the second runs bottom to top, joined by a right arrow at the bottom.

    A right-to-left deck mirrors the designs: the first chevron or box is at the right, the chevrons point left (flipH) and the arrows turn.

    Two or more list blocks in the region are the entries themselves, as SH-18 draws them for a cycle: each block on a card of OPF's chevrons or boxes (the card neutral and its border), with the arrows between them, its list kept with its markers, start-aligned and at one size for all, the largest from 1.5 times the role size at which every card holds its list without breaking a word; the cards then take the height of the tallest list. Where no row or column holds them, a grid of cards does (opf/arrangement-adapted), and where nothing does, the blocks keep the region's composition, which pagination splits. A SmartArt node holds no list, so these are native shapes, with pptx/smartart-unavailable saying so.

    In PowerPoint

    toPptx(deck, { smartArt }) (CLI --smart-art <live|shapes>) takes the first that fits, as the owner decided for SmartArt:

    1. Live SmartArt ("live", the default), for PowerPoint's design where the instance lies in the domain below. The slide holds a p:graphicFrame of the diagram (dgm:relIds) inside an mc:AlternateContent whose mc:Fallback is OPF's picture of it (in other apps), and the package:

      • the data model (ppt/diagrams/dataN.xml): the doc point names the layout (loTypeId pyramid1, matrix3, chevron1, process1 or process2) and OPF's own quick style and colors (qsTypeId urn:openpresentation.org/opf/smartart/quickstyle/flat, csTypeId urn:openpresentation.org/opf/smartart/colors/primary); one point per entry, marked custT="1" so PowerPoint keeps the run size (and sets the text margins from it) instead of fitting the text again; runs with OPF's size, typeface and language; each paragraph centered at OPF's line pitch with no space before or after; one cxn per entry from the doc point, in entry order;
      • OPF's colors and quick style parts, whose style labels cover every label the five layouts use (node0, node1, bgShp, revTx and the transition labels: a label a part leaves out renders black; sibTrans2D1, a flow's arrows, is the primary lightened 40% without a line), in the theme's scheme colors, so a theme change in PowerPoint recolors the SmartArt;
      • a drawing part built from OPF's geometry (ppt/diagrams/drawingN.xml), for readers that show the cached shapes;
      • no layout part for PowerPoint's design: OPF never writes Microsoft's layout definitions. r:lo is empty, and PowerPoint resolves its built-in layout from loTypeId (see the layout relationship). OPF's own designs are live on OPF's own layout definitions, which the package carries;
      • the fallback picture: an SVG (asvg:svgBlip) over its PNG raster, the picture path SH-07's diagrams use.

      Live SmartArt needs that picture, so it needs opf-render (an optional peer) to draw the slide and an SVG rasterizer for the PNG (the default in Node is opf-render's; a browser passes svgRasterizer). Without either the arrangement is written as native shapes and smartart-unavailable says so. SH-19: OPF's own design of an arrangement, and a matrix's card grid, is live SmartArt on OPF's own layout definition of that design where its domain holds (OPF's own layouts).

    2. Native shapes: OPF's design and a card grid outside the domain of OPF's layouts, PowerPoint's design outside its domain, and everything with "shapes". The diamond and the nodes are presets (diamond, trapezoid with core's adjustment, roundRect; SH-17: chevron with core's adjustment, mirrored with flipH in a right-to-left deck, and a flow's arrows as rightArrow with core's two adjustments, turned with rot) at core's boxes, under the list's measured lines, tagged OPF_ARRANGEMENT_V1. OPF's design and PowerPoint's outside the domain report the info finding pptx/smartart-unavailable with the reason (reason in the diagnostic); a card grid reports core's arrangement-adapted.

    Both look the same as the preview: the shapes are exact, and the text is where OPF's native text export puts any text.

    In other apps

    Many apps that open PPTX files do not lay SmartArt out. Each live SmartArt is therefore written as an mc:AlternateContent (ECMA-376 Part 3, markup compatibility): the mc:Choice holds the SmartArt frame and requires the namespace http://schemas.microsoft.com/office/drawing/2016/12/diagram (prefix dgm1612, Office 2016's SmartArt extensions), and the mc:Fallback holds a picture with the same id and name, at the same frame grown by its bleed on every side. PowerPoint understands the namespace and shows the live SmartArt. Every other app, one that does not claim the namespace, takes the fallback and shows OPF's own drawing: the preview of that frame (opf-render's SVG of the slide, cut to the arrangement and its entries, the text as glyph outlines in the deck's fonts when the export has the fonts handle), transparent around the drawing, as an SVG with a PNG raster at 192 dpi of the frame for apps without SVG. The bleed is how far the preview draws outside the frame: half of the nodes' outline, and further at a pyramid level's miter-joined corners (its base corners reach 2.6 px past a 16:9 frame), plus a pixel of antialiasing, so the picture draws everything the preview draws. With typefaces: "safe" the picture is laid out and drawn in the Office-safe families the export writes and lays the nodes out with (Arial for the default Aptos), not the preview's. Its alt text is the outline, one entry a line in entry order (Pyramid of 4 levels, top to bottom: then each level). SH-09's drawing part stays in the package for readers that draw the cached SmartArt shapes.

    The namespace was chosen through COM (PowerPoint for Microsoft 365 on Windows 11, 2026-11-06), on a pyramid and a matrix whose mc:Choice required each candidate in turn: dgm1612 (2016/12 diagram), dgm1611 (2016/11 diagram), dgm14 (2010 diagram), dsp (2008 diagram drawing), dgm (the ECMA diagram namespace) and p14 (PowerPoint 2010) all opened as live SmartArt (HasSmartArt, Basic Pyramid or Basic Matrix, every entry a node, the slide exported as the SmartArt); a namespace PowerPoint does not know took the fallback, which PowerPoint drew as the preview does (mean pixel difference 1.27 and 1.14 of 255 over the slide). The chosen namespace is the narrowest of those: only Office's SmartArt uses it, whereas an app that reads the ECMA diagram markup or the cached drawing, or claims the PowerPoint 2010 extensions, could claim the others without laying SmartArt out. No third-party app was run (LibreOffice and the Open XML SDK were not installed); the pptx tests check the markup with a markup-compatibility reader that does not know the namespace, which takes the fallback picture and finds a well-formed SVG and a PNG.

    Saved again in PowerPoint, the file keeps the live SmartArt (COM: SaveCopyAs, reopened, HasSmartArt and every node) but not the wrapper: PowerPoint writes the bare SmartArt frame and its own drawing and layout parts, and drops the fallback picture and its media. Other apps then see what any PowerPoint-saved SmartArt gives them, typically its cached drawing. Import reads the file as before.

    What PowerPoint does on edit

    PowerPoint lays a SmartArt out again from its data model whenever it opens or changes it; it never shows the cached drawing. A SmartArt OPF wrote opens as live SmartArt (COM: HasSmartArt, the layout, one node per entry in AllNodes), and Change Layout switches it to another SmartArt layout. When a node's text is edited, PowerPoint keeps OPF's explicit size for every node (it does not shrink the text to fit), keeps the node rectangles, and wraps the text in the node with its own line breaker.

    The domain

    The PPTX export writes live SmartArt only for the instances the PowerPoint comparison passed (core SMART_ART_DOMAIN, smartArtUnavailable):

    PyramidMatrix
    Entries1 to 5 (6 to 8 levels never fitted the list's role size in the compared frames)1 to 4
    Frame proportions (width over height)1.1 to 2.3 (the 16:9, 4:3 and 1:1 slides gave 2.26, 1.64 and 1.17)0.8 to 1.2 (16:9 and 4:3 gave 1.13 and 0.83; the 1:1 slide's 0.58 is outside)
    Linesat most 2 for an entry and 6 for all of them, each break holding when the width changes by 4% (core fits every entry at both widths)the same
    TextLatin script, left to right; runs bold, italic, underlined or struck through only (no links, citations, code or sizes of their own); no line breaks or tabsthe same
    Line width (SH-19)no line wider than 500 px: PowerPoint measures a line up to about 0.6% wider or narrower than OPF, and a centered line moves by half the difference (the SH-19 sweep: a one-level pyramid's 537 px line sat 1.74 px from OPF's native text, outside SH-07's 1.5 px; every line of 440 px or less was within it); it applies to every arrangement and to OPF's own layouts, and such an instance draws OPF's designthe same

    Everything else (Arabic, Hebrew and Japanese text, longer or tighter entries, other proportions) is written as native shapes, which look the same as the live SmartArt; the finding names the first reason.

    The gate

    The domain comes from a PowerPoint sweep through COM (PowerPoint for Microsoft 365 on Windows 11, read-only, visible and minimized, quit after each batch): a pyramid of 1 to 8 entries and a matrix of 1 to 4, with short, medium and long entries, on 16:9, 4:3 and 1:1 slides, and an Arabic and a Japanese deck of each, 112 decks. The Latin decks use Source Sans 3, which OPF bundles and the comparison machine has installed, so the preview and PowerPoint draw one font. Each deck was exported twice, live (smartArt: "live") and as native shapes ("shapes"), and both files were drawn by PowerPoint at the canvas's pixel size. An instance passes when:

    1. PowerPoint opens it as live SmartArt of the right layout with every entry as a node;
    2. every node rectangle PowerPoint lays out (GroupItems) is within 0.1 pt of core's;
    3. every node's font size in PowerPoint equals OPF's;
    4. the shapes show no visible difference from the preview (no 16 px block of the shapes' area outside the text whose mean differs by more than 24/255);
    5. PowerPoint draws the live SmartArt as it draws OPF's native export of the same slide, within SH-07's native gate (mean absolute difference at most 1 at a quarter size and 2 at full size, at most 0.5% of blocks visibly different, the text's ink within 1.5 px).

    The domain was set from a first pass, in which 38 decks drew PowerPoint's design live: in all 38 PowerPoint laid every node out where core does (largest difference 0.000 pt) at core's size, with no visible difference in the shapes, and 33 passed every check. The five that did not are the two Arabic decks (their text 6 px from where the native export draws it), the Japanese matrix (10 px) and two four-quadrant matrices whose entries wrap (mean differences 1.1 and 1.22), which is why the domain leaves out right-to-left and Chinese or Japanese text, entries of more than two lines and more than six lines in all.

    With the domain and OPF's designs in place, the same 112 decks draw: PowerPoint's design as live SmartArt in 20 (every one passes: node rectangles within 0.000 pt, sizes equal, no visible shape difference, the SmartArt within SH-07's gate of the native export), OPF's design in 60, a card grid in 27 (the matrices of 1 to 3 entries) and a list in 5 (pyramids of 6 to 8 long levels on 4:3 and 1:1 slides, which no design holds at 24 pt; opf/arrangement-adapted says so). In every native-shape instance PowerPoint draws the shapes as the preview does (mean difference at most 1.15 inside the shapes and outside the text, no visibly different block).

    The text of a live SmartArt sits where OPF's native export puts any text, within a pixel. Both sit higher in PowerPoint than in the preview, by about 0.15 em in Source Sans 3 and 0.07 em in Aptos (4 to 7 px at the 24 to 36 pt of these nodes): OPF's text model sets every baseline one em below its line's top, and PowerPoint at a height that depends on the face. That difference is the same for every OPF text box and is not SmartArt's; at the sizes of SH-07's diagram labels it is under the 1.5 px gate, at these sizes it is not, so the gate compares the SmartArt with the native export, and the preview offset is recorded in the gate table. The evidence (the gate table, the preview | PowerPoint | native export sheets and the COM records) is in the SH-09 build folder of the shapes program.

    The domain and the gate of chevrons and flows

    The SH-17 sweep measured PowerPoint's own layout first (AddSmartArt through COM, 1 to 10 nodes in 20 frames per layout, the fitted model above), then ran the gate above on OPF's exports: chevrons, a flow and a vertical flow of 1 to 8 entries, short, medium and long, on 16:9, 4:3 and 1:1 slides, and an Arabic and a Japanese deck of each, 222 decks, each exported live and as native shapes and drawn by PowerPoint. The domain (core SMART_ART_DOMAIN) is what passed:

    ChevronsFlowVertical flow
    Entries1 to 6 (7 and 8 never fitted the list's size in PowerPoint's design)1 to 61 to 4
    Frame proportions1.1 to 2.3 (the 16:9, 4:3 and 1:1 slides gave 2.26, 1.64 and 1.17)the same0.55 to 1.2 (the region beside the title: 1.13, 0.81 and 0.58)
    Fillno entry's lines more than 95% of its node's text height (less margins): PowerPoint lays a box nearly full of text out again, in the first pass at 0.969 and 0.996 of the height (node rectangles 0.22 and 5.3 pt off), while every box at 0.95 or less matchedthe samethe same
    Inkthe entries' text (each line's width times the size) at most 8% of the frame: PowerPoint sets a SmartArt's lines up to a pixel from where it sets OPF's native text (the same text in both), which in a frame dense with large text passes SH-07's mean-difference gate no longer (every first-pass failure of that check was at 0.082 or more)the samethe same
    Lines, textas the pyramid's: at most 2 lines an entry and 6 in all, breaks that hold within 4% of the width, Latin left-to-right text with bold, italic, underline or strike onlythe samethe same

    In the first pass, before the fill and ink limits, 88 decks drew PowerPoint's design live; every node lay exactly where core lays it (0.000 pt) at core's size in 81, and 74 passed every check. The seven laid out again had nearly full text (the fill limit), and the seven others that drew the SmartArt more than a mean of 1 from the native export were dense with large text (the ink limit). With the domain in place: 63 decks draw live SmartArt, every one passing (node rectangles and arrows within 0.000 pt of core, sizes equal, no visible difference in the shapes, within SH-07's gate of the native export), 151 draw OPF's own design and 8 stay a list (7 or 8 long chevrons and 8 long boxes on the 1:1 slide, and vertical flows of 6 to 8 long entries on the 4:3 and 1:1 slides, which no design holds at the list's size; opf/arrangement-adapted says so). Every native-shape instance PowerPoint draws as the preview does: no visibly different block in the shapes, a mean difference of at most 2.7 outside the text (the outlines' antialiasing). The Arabic and Japanese decks are OPF's design, as for the pyramid and matrix.

    COM also showed, on two live instances of each (short and medium entries): HasSmartArt, the layout's name, every entry a node of AllNodes; a node's text edited in PowerPoint keeps OPF's size in every node and every node rectangle; Change Layout to Basic Block List keeps the nodes. With its mc:Choice requiring an unknown namespace, PowerPoint draws SH-16's fallback picture of each (mean difference 1.3 from the preview over the slide, about what it is for the live SmartArt). Import reads a Basic Chevron Process, Basic Process or Vertical Process that PowerPoint itself wrote back to the chevrons, flow or vertical-flow layout and its list, from the data model.

    The layout relationship

    The SH-09 spike found that PowerPoint resolves its built-in layout from the data model's loTypeId only when the package has no layout part; a stub layout part with the built-in uniqueId is not resolved. The frame's dgm:relIds still names four relationships, and there are three ways to point r:lo at nothing. All three open as live SmartArt in PowerPoint with the layout resolved, a node edit and Change Layout working (the build checked each on a pyramid and a matrix):

    FormWhat a validator seesChosen
    r:lo=""the attribute the schema requires is present (ST_RelationshipId is a string), and an empty id can name no relationship (relationship ids are non-empty)yes
    r:lo="rId99", no such relationshipa reference to a relationship the part does not have: a broken package referenceno
    no r:lo attributethe schema's required attribute is missingno

    The empty id is the only form that is both schema-valid and names nothing that is missing. The Open XML SDK validator and LibreOffice were not installed on the comparison machine, so neither was run.

    Cycles (SH-18)

    The cycle layout (a region's arrangement: "cycle") draws its lone list of one level as the steps of a loop: each entry a node on a ring, the first at the top and the rest clockwise, with an arrow from each to the next and from the last back to the first. It draws 1 to 8 entries (two make a pair with an arrow each way; more than 8 stay a list, with opf/arrangement-adapted).

    json
    { "layout": "cycle", "title": "How each round feeds the next", "items": ["Plan", "Pilot", "Measure", "Adjust"] }

    The designs are tried in order, each at one text size for every node, from 1.5 times the list's role size down to the role size, never below it:

    Design (design)Drawing
    PowerPoint's (smartart), written as live SmartArt inside the domainBasic Cycle (cycle2): circles on a ring with an arrow after each, in a frame of the cycle's own proportion centered in the region
    OPF's (opf), where the circles do not hold the entries at the list's size or the instance lies outside the domainthe larger-text of two designs that fill the region: rounded rectangles on an ellipse (two entries side by side, or one above the other in a tall region), or a loop of two or three rows (columns in a tall region), clockwise from the top left, with a node on every place of its edge: two rows hold ceil(n/2) nodes on top and the rest below, three rows a node at each end of the middle row too, and a row of fewer nodes widens them so every row spans the region (no place is ever empty); the arrows on the lines between the nodes' centers, all one size

    Every node is filled in the primary color with the 1.5 pt background-colored outline of the other arrangements; the arrows are the primary lightened 40% (lumMod 60000, lumOff 40000), without an outline; the entries are centered on their nodes in the color that reads on the primary. A right-to-left deck mirrors the cycle, so its steps run counterclockwise.

    PowerPoint's Basic Cycle

    PowerPoint lays Basic Cycle out from its frame alone: through COM (PowerPoint for Microsoft 365 on Windows 11, 2026-11-06) the circles and arrows were the same for short, medium and long text and at every font size, and exactly proportional to the frame at a fixed width over height (the largest difference across frames 230 to 920 pt wide was 0.0004 pt). At other proportions the result is not a closed form: PowerPoint searches the circle size as a ten-bit binary fraction of the frame's width and fits the ring in a second step, so the circles move by up to 0.4 pt between frames a few points apart. OPF therefore gives the SmartArt frame one proportion per count, 1.8 : 1 for two entries and 1.15 : 1 otherwise (the cycle's own proportion, a little wider so its height decides the size), the largest such frame centered in the region, and reproduces PowerPoint's geometry from the frame's width W (core cycleFrame, cycleNodes):

    Basic Cycle
    CirclesDrawingML ellipse, diameter CYCLE_DIAMETER[n]·W: 890/1024, 818/2048, 774/2048, 570/2048, 538/2048, 890/4096, 830/4096 and 723/4096 for 1 to 8 entries
    Ringthe centers at radius CYCLE_RADIUS[n]·W (0.3002, 0.32775, 0.29552, 0.33524, 0.32594, 0.3506 and 0.34595 for 2 to 8), the first at the top and the rest clockwise 360/n degrees apart; two entries left and right; the ring centered in the frame, an odd count's moved down by half of r·(1 − cos(180°/n)) so the circles' bounds are centered
    Arrowsafter each circle a DrawingML rightArrow (adj1 60000, adj2 50000) 0.3375 of a diameter thick, on the line between two centers, 0.53 of the gap between the circles long and starting 0.22 of it after the first circle, turned to point along the line; two entries: one over and one under the gap, 0.2491·W long, their centers 0.1888·W off the line and 0.4883 of the way along it
    Text rectanglethe ellipse's own (the inscribed square, as PowerPoint's txXfrm), inset by 0.1 of the font size
    Stackingeach arrow after its circle (two entries: an arrow's head runs under the next circle)

    The model matches every circle and arrow PowerPoint drew within 0.0001 pt in the frames measured. In the live SmartArt, the arrows are the data model's transition points (sibTrans, one after each entry, the last closing the loop), styled by OPF's sibTrans2D1 label.

    In PowerPoint (COM), a cycle OPF wrote opens as live SmartArt (HasSmartArt, Basic Cycle, one node per entry); a node edit keeps OPF's size on every node and the circles where they were, Change Layout switches it to another layout with every entry, and a file saved again by PowerPoint (which writes its own layout part) imports back as the same list on the cycle layout, as does a Basic Cycle PowerPoint authored. With the mc:Choice requiring a namespace PowerPoint does not know, PowerPoint draws the fallback picture as the preview does (mean difference 0.94 to 1.05 of 255 over the slide, two, four and six entries).

    The cycle domain

    Cycle
    Entries1 to 8
    Frame proportionsexactly cycleAspect's (1.8 : 1 for two entries, 1.15 : 1 otherwise), which OPF always gives it
    Linesat most 2 for an entry, each break holding when the width changes by 4%; no limit in all (each circle holds its own entry: eight entries of one line and four of two matched)
    TextLatin script, left to right, with the run formatting of the other arrangements

    Right-to-left decks stay native shapes: PowerPoint draws Basic Cycle clockwise from the data model whatever the deck's direction, so its circles are not where OPF's mirrored cycle draws them. Japanese text sat 11 px off in the comparison. Entries of three lines or more drew within the shapes but outside SH-07's native gate (mean difference 1.03 to 1.62).

    The cycle gate

    The same comparison as SH-09's, plus every arrow's rotation: a cycle of 1 to 8 entries with short, medium and long entries on 16:9, 4:3 and 1:1 slides, and an Arabic and a Japanese deck of 3, 4 and 5 entries, 78 decks, each exported live and as native shapes and drawn by PowerPoint. A first pass with the domain open to every count, line count and script drew 35 decks live: PowerPoint laid every circle and arrow out where core does (largest difference 0.000 pt, rotations equal) at core's size in all of them, and 26 passed every check; the nine that did not are the three right-to-left decks (PowerPoint's clockwise layout), the three Japanese decks (text 11 px off) and the three entries of three or four lines. With the domain above the 78 decks draw: PowerPoint's design as live SmartArt in 25 (every one passes: circles and arrows within 0.000 pt, rotations equal, sizes equal, no visible shape difference, the SmartArt within SH-07's gate of the native export: mean difference at most 0.84 at a quarter size and 1.29 at full size, no visibly different block, the text's ink within 1.03 px), OPF's design in 44 (PowerPoint draws the shapes as the preview does: mean difference at most 1.04 inside the shapes and outside the text, no visibly different block) and a list in 9, filling the region (five to eight long entries on 1:1 slides, six to eight on 4:3, eight on 16:9, and eight medium entries on a 1:1 slide; opf/arrangement-adapted says so). In each of the nine, every design breaks one of the two rules that keep text readable at the list's role size: a word wider than its node (OPF never breaks a word: "Customers" is 143 px at 32 px, a node of the 4 + 4 loop on 16:9 109 px wide) or entries taller than their node (OPF never sets an arrangement's text below the list's role size, or below the list's fitted size when the list itself fits smaller: the 3 + 1 + 1 + 3 loop on 16:9 is 16 px short). The evidence is in the SH-18 folder of the shapes program.

    OPF's own layouts (SH-19)

    PowerPoint's built-in layouts draw PowerPoint's designs. OPF's own designs (the second design of each arrangement: the stepped pyramid with the entries beside it, the four quadrants, the card grid of a matrix, the ring and the two- and three-row loops of a cycle, the wrapped and snaking rows of chevrons and flows, the snaking columns of a vertical flow, and their compact forms) have a SmartArt layout definition of their own, which OPF writes (ECMA-376 Part 1, §21.4, dgm:layoutDef) from the specification and OPF's design, never from PowerPoint's parts. Where the instance lies in the domain the export writes the design as live SmartArt on that definition, so it is editable SmartArt in PowerPoint, with SH-16's fallback picture as every live SmartArt has. Elsewhere it stays native shapes, with pptx/smartart-unavailable and the reason.

    A design inside a built-in layout's domain keeps PowerPoint's built-in layout, which is better for the user (it is PowerPoint's own).

    What OPF writes

    The definition is a table of explicit constraints, the one form PowerPoint lays out with no algorithm of its own (the SH-19 spike, through COM):

    • a composite root with one if branch per entry count, 1 to 12 (func="cnt"; SH-23); each branch places every node with l, t, w and h constraints as shares of the frame's width and height, so the design scales with the frame when it is resized in PowerPoint;
    • a hidden full-frame child (shape type="rect" hideGeom="1") in every branch: without one, PowerPoint moves a drawing whose shapes do not reach the frame's edges to the frame's center;
    • every entry a forEach axis="ch" ptType="node" st="i" cnt="1" with its node: a tx node in the design's preset (roundRect with the design's corner, chevron with its point, ellipse), or, where the text is not in the preset's own text rectangle (beside a stepped pyramid's level), a shape and a text-only node (revTx, no fill or line) at the entry's text rectangle;
    • every arrow the transition after its node (forEach axis="followSib" ptType="sibTrans" cnt="1", hideLastTrans="0" for a loop's closing arrow): an axis-aligned arrow a rightArrow turned by rot (0, 90, 180 or 270 degrees, which PowerPoint keeps); any other arrow a conn connector between the two node centers (begPts/endPts ctr) with its pads and thickness in millimeters;
    • every entry's text at OPF's size: the data model's run sz (with custT="1", so PowerPoint never fits it again) and a primFontSz constraint at the same size for an entry added in PowerPoint; the margins as shares of the size (lMarg and the others, refType="primFontSz"); the composed count's entries anchored at the top (txAnchorVert t) with the top margin where OPF's composition starts the entry's first line, so PowerPoint sets the lines where OPF's native text export does (centering the block, PowerPoint's line metrics put it up to a pixel off);
    • the tables of every other count, from the same design at the same frame (core opfLayoutGeometry), so a node added or deleted in PowerPoint keeps OPF's design. OPF draws at most 8 entries, but the definition carries tables up to 12 (OPF_LAYOUT_TABLES, SH-23), so a ninth to twelfth entry a PowerPoint user adds is drawn in OPF's design at the same frame; more than 12 take the 12-entry table under names of their own (x..., the entries past 12 not drawn until one is removed). A design whose larger tables do not hold (a node outside the frame) keeps tables up to 8.

    Every name of a layout node or forEach is unique in the definition: PowerPoint rejects a definition that repeats one (it opens the SmartArt on Basic Block List instead). The identity is OPF's: the uniqueId is urn:openpresentation.org/smartart/layout/<design>/<digest> (the design's name, -compact for a compact form, and a 64-bit digest of the definition's body), the title and description are the design's (OPF Stepped Pyramid, OPF Card Grid, OPF Snaking Flow, OPF Chevron Rows, OPF Snaking Vertical Flow, OPF Cycle Ring, OPF Cycle Loop, OPF Cycle Loop of Three Rows, OPF Quadrants), in the category of the arrangement (pyramid, list, process, cycle, matrix). The digest is there because PowerPoint keeps the first definition it loads for a uniqueId for the rest of its session: in the sweep, a deck opened after another with the same uniqueId and other tables was drawn with the first deck's tables. Two exports of one design at one frame share a uniqueId; within one file, two frames with one uniqueId and different content each use their own part. The data model's loTypeId names the definition, and the frame's r:lo names the layout part. The style labels are OPF's colors part's: the nodes node1 (a card alignNode1, the card neutral with its border, the outline's width on each entry's spPr), the text-only nodes revTx (the slide's text color) and fgAcc1 (the color that reads on the primary), the arrows sibTrans2D1.

    The export evaluates the definition with OPF's interpreter (core evaluateLayoutDefinition, layoutArrangement) and writes it only when the result is the composed design within 0.1 pt: the preview, the definition and PowerPoint then draw the same nodes.

    OPF's interpreter

    Core's smart-art-layout.ts evaluates the subset of §21.4 OPF writes, as PowerPoint does: the layout node tree, forEach over the entries and their transitions (st, cnt, hideLastTrans), choose/if on the entry count, composite with l/t/w/h constraints (shares of the root's w and h, or millimeters), PowerPoint's centering of a drawing that does not reach its frame's edges, sp, tx and conn (between centers), and shapes with their presets, adjustments and turns. A turned shape is PowerPoint's: the largest-area rectangle at that angle inside its constraint box (at 45 degrees always a square), which is why OPF writes turned arrows as connectors. It returns nodes and arrows in the shape of a composed arrangement (with the preset's text rectangle, or a text-only node's box), so render and pptx have no second geometry path. It runs on the edge runtimes (no Node built-ins). Its unit test replays what PowerPoint drew through COM for OPF-written definitions (177 shapes: composite placement with and without the full-frame child, shapes turned by rot, connectors, the closing transition) within 0.001 pt.

    The spike measured the other algorithms in OPF-written definitions too. Each evaluates, but none is a closed form: lin scales each axis on its own and centers with a 0.1 to 0.4 pt offset; snake picks its own grid (up to 343 pt from the best-grid form); cycle searches the node size (215 pt from the form its parameters describe); pyra keeps the levels to 0.67 of the width. OPF's definitions use none of them.

    In PowerPoint

    Through COM (PowerPoint for Microsoft 365 on Windows 11): PowerPoint opens the SmartArt live on OPF's layout (HasSmartArt, the layout's name and description, category as written, every entry a node of AllNodes); Application.SmartArtLayouts does not list it (the Change Layout gallery itself was not inspected: COM only). A node's text edited keeps OPF's size in every node and every node rectangle; an added node takes the next count's table and OPF's size (primFontSz; without it PowerPoint's 36 pt), deleted nodes the smaller count's; Change Layout to Basic Block List and back (to the layout object read before) restores OPF's layout with every node where it was. A file saved again by PowerPoint keeps the layout part, rewritten but the same definition (PowerPoint adds empty sample, style and color data with useDef, leaves out attributes at their schema defaults, reorders attributes and writes the constraint factors to four decimals, a change of at most 0.00005 of the frame; normalized, the part equals OPF's), the same uniqueId and the same node rectangles on reopening (five designs checked); it drops SH-16's wrapper as for every SmartArt. OPF imports such a file back to the same arrangement and list. With the mc:Choice requiring a namespace PowerPoint does not know, PowerPoint draws SH-16's fallback picture (mean difference 1.3 to 1.6 from the preview over the slide, as for the other live SmartArt). A definition PowerPoint cannot evaluate (a connector to a node of another name, in a build of this item) made PowerPoint take the mc:Fallback picture.

    The domain of OPF's layouts

    The export writes OPF's layout only for the instances the PowerPoint comparison passed (core SMART_ART_OPF_DOMAIN, opfSmartArtUnavailable):

    Passed
    Designs and entriesthe stepped pyramid 1 to 8, the card grid 1 to 6, chevron rows 6 to 8, the snaking flow 6 to 8, the snaking vertical flow 5 to 6
    Text densitythe entries' text (each line's width times the size) at most 7.5% of the frame
    Linesat most 2 an entry, each break holding when the width changes by 4% (core fits every entry at both widths, as for PowerPoint's designs)
    TextLatin script, left to right, with the run formatting of the other arrangements
    Line widthno line wider than 500 px, as for PowerPoint's designs

    Everything else is native shapes, which look the same: OPF's cycles and quadrants (every instance of the sweep held its entries only above the text density), the compact forms, right-to-left and Chinese or Japanese text, a row of chevrons pointing left (a layout definition has no mirrored shape), and list blocks (a SmartArt node holds no list).

    The gate

    The sweep of SH-09, SH-17 and SH-18 for all six arrangements: 1 to 8 entries, short, medium and long, on 16:9, 4:3 and 1:1 slides, an Arabic and a Japanese deck of 3, 4 and 6 entries and decks of 2 and 4 list blocks of each, 468 decks, each exported live and as native shapes and drawn by PowerPoint; an instance passes on SH-09's checks (every node rectangle within 0.1 pt, arrows turned the same, equal sizes, no visible shape difference from the preview, within SH-07's native gate of the native export).

    With the domain open (every OPF design live), 310 decks drew OPF's layout live. In the first pass PowerPoint laid 261 of them out where OPF's interpreter does; the other 49 were drawn with the tables of an earlier deck of the same uniqueId (the session's cache, hence the digest). With the digest every one of the 310 laid out within 0.1 pt of the interpreter (largest difference 0.000 pt, arrows turned the same) at OPF's size. The pixel checks then found what the domain keeps out: with the text centered in its node, PowerPoint set the lines up to a pixel off (mean difference from the native export up to 6 in dense frames); anchoring the text at OPF's first line halved that, and of the 292 Latin instances every one at a text density of 7.5% or less with stable breaks passed (30), while above it some did not (a mean difference of 1.01 at 8%, up to 3.3 in the densest), and every Japanese instance did not (text 11 px off). With the domain (and the 500 px line limit, which moved five built-in decks with a single long entry to OPF's design), 30 decks draw OPF's layout live and every one passes (node rectangles within 0.000 pt of the interpreter, sizes equal, no visible shape difference, within SH-07's gate of the native export: a mean difference of at most 0.8 at a quarter size and 0.94 at full size, the text's ink within 1.08 px); PowerPoint's built-in layouts draw 102 live decks, every one passing; OPF's design draws 271 decks and a card grid 43 as native shapes, and 32 stay a list. The gate's shapes-versus-preview check leaves out each node's text rectangle grown by 0.2 em up and down, so the preview's known text offset is not counted as a shape difference. The evidence (the gate table, sheets of preview, PowerPoint live and PowerPoint's drawing of the native export, labelled with the design and the layout that drew it, and the COM records) is in the SH-19 folder of the shapes program.

    Import

    A SmartArt OPF wrote (in full provenance it carries OPF_SMARTART_V1: the slide, the list path, the arrangement, the authored list and the text of every node) imports as its list on its layout again when its node texts are unchanged. An edited one, and a Basic Pyramid, Basic Matrix, Basic Cycle, Basic Chevron Process, Basic Process or Vertical Process that PowerPoint wrote, import their data model's nodes: the points in cxn order (nested entries keep their level, SH-08), their text and run properties (bold, italic, underline, strike), as a list on the pyramid, matrix, cycle, chevrons, flow or vertical-flow layout (the one of the same arrangement; SH-19: a SmartArt on one of OPF's own layouts by its uniqueId, urn:openpresentation.org/smartart/layout/<design>/..., also after PowerPoint saved it again), with smartart-edited for an edited OPF SmartArt. The diagram's place on the slide is its frame's; the nodes' geometry is the layout's again, as OPF draws it, so nothing of PowerPoint's layout is kept. Import never reads or stores a SmartArt layout part. A wrapped frame is read from its mc:Choice. A file in which only the fallback picture survived (an app without SmartArt saved it) imports the same way in full provenance: the picture carries OPF_SMARTART_V1 too, with the media keys of its SVG and PNG, so a picture still named for its list whose SVG or PNG is the one written is that list on its layout again; a replaced picture reports invalid-smartart-provenance and imports as a picture, as does any fallback without the record. Native-shape arrangements (OPF_ARRANGEMENT_V1 on every node) are the layout's drawing: import consumes them and reads the measured lines as the list. Every other SmartArt imports as the next section says.

    SmartArt PowerPoint users made (SH-08)

    Any SmartArt PowerPoint wrote imports with its text (src/smart-art-import.js in opf-pptx). Import reads the data model, not the layout part:

    • The outline: the parOf connections from the doc point, in srcOrd order, as a tree of entries (nodes and an organization chart's assistants, type="asst"). An entry left as a placeholder (no text and no explicit authored-text marker) gives its place to its children; an explicitly authored blank parent retains its children and level. A run's bold, italic, underline and strike are kept, a line break (a:br) is a newline and a field is its text. The paragraphs of one node are one entry, except on the list and process layouts, where each is an item of its own at the node's level: PowerPoint holds a Vertical Bullet List's or an Accent Process's bullets as the paragraphs of one child node (its object model adds no second child there). A node's title box is its entry's description: in a Name and Title Organization Chart PowerPoint stores the title on the sibling transition of the node's connection and presents it (presOf); a presentation point with text of its own (custT) is read the same way.
    • The layout: the doc point's loTypeId, then its category (loCatId).

    Then the first form that fits:

    LayoutOPF
    The known arrangement IDs (pyramid1, matrix3, cycle2, chevron1, process1, process2, orgChart1, hierarchy1, venn1) and OPF's own layoutsThe list on that arrangement's layout, as above
    Inverted Pyramid (pyramid3), widest entry firstThe list on the inverted-pyramid layout, retaining its order (the widest first), with smartart-adapted. The inverted arrangement exports as editable native shapes; it has no qualified live SmartArt domain.
    Another layout of those families: pyramidN, matrixN, cycleN (Text Cycle, Block Cycle...), any chevron layout (chevron2, hChevron3...); a layout of category pyramid, matrix or cycleThe list on the pyramid, matrix, cycle or chevrons layout, with smartart-adapted (info)
    The list family (category list: Basic Block List default, Vertical Bullet List vList2...)The list on the list layout, with smartart-adapted
    The process family (category process, other than the arrangements above: Accent Process, Step Up Process...)The list on the process layout, with smartart-adapted
    Other hierarchies and organization charts, or the known trees with assistants, separate roots or labeled edges (orgChartN, hierarchyN, the 2008 and 2009 organization charts; category hierarchy)A Mermaid flowchart (flowchart TD, flowchart LR for a horizontal hierarchy): a node per entry, its description on a second line, an edge from each parent, dotted (-.->) to an assistant, the connection's own text as the edge's label; the outline as the diagram's alt; smartart-adapted
    Every other layout (other Venn, radial, target, funnel, gear, picture layouts...)The drawing (dsp:drawing, which PowerPoint saves with every SmartArt) as overlay shapes, with smartart-adapted; without a drawing, the list on the list layout

    Nested entries keep their level in every list. Organization charts and hierarchies use those depths as parent links; the other arrangements draw one level, so a pyramid of nested entries draws as a list (opf/arrangement-adapted says so) with all of its text.

    The drawing as overlay shapes. The drawing's shapes are read by SH-05's reader (shapes), in a group that places the drawing's space on the SmartArt's frame, so each is an overlay shape with its preset, fill, line, effects and text, in PowerPoint's z order at the frame's place. PowerPoint sets a SmartArt shape's text in a rectangle of its own (dsp:txXfrm): where it lies inside the shape's own text rectangle, the difference is the shape's padding; where it does not (a Venn circle's text across the overlap, text beside a shape), the shape is drawn without text and a text-only overlay (no fill, no line, style none) holds the text at PowerPoint's rectangle. Text in another color than the slide's text color keeps it (the style's fontRef). The first shape carries the outline as its alt text (SmartArt (venn1): and an entry a line, indented a level by two spaces). The no-op color transforms and empty 3-D element PowerPoint writes on every SmartArt shape are not reported as dropped properties; anything else SH-05 does not keep is (shape-property-dropped).

    Nothing dropped silently. What the OPF form does not keep is a finding naming it:

    FindingWhen
    smartart-text-dropped (with text)Text on a transition (a connection's own text, other than an organization chart's edge label or a presented title box), text on a presentation point that belongs to no entry, or an entry no connection reaches
    smartart-picture-droppedAn entry's picture fill (a picture layout), in every form but the drawing
    smartart-formatting-droppedBold, italic, underlined or struck-through text in a hierarchy (Mermaid labels carry none); the text is kept
    smartart-adapted (with form: an arrangement, list, process, diagram or shapes)A layout imported as anything other than the arrangement OPF writes for it
    smartart-connection-droppedConnections to an entry that already has a parent (or to its own ancestor): each entry is imported once, under the first parent that reaches it
    smartart-limited (with limit: depth or entries)An outline more than 64 points deep or longer than 10,000 entries: the drawing, or without one the entries as a flat list (the first 10,000)
    smartart-unreadableA data model or drawing that cannot be read: the frame imports as any other graphic frame, or the drawing adds no shapes

    Every finding has the layout (layout, the last segment of loTypeId) and the slide's path. A SmartArt with no text and no drawing (only placeholders) imports as nothing.

    Evidence. packages/pptx/test/fixtures/powerpoint-smartart.pptx is PowerPoint-made (COM, PowerPoint for Microsoft 365 on Windows 11, 2026-11-07): 17 layouts (Basic and Inverted Pyramid, Basic Matrix, Basic and Text Cycle, Basic Chevron Process, Basic, Vertical and Accent Process, Basic Block List, Vertical Bullet List, Basic Venn, Basic Radial, Organization Chart, Hierarchy, Horizontal Hierarchy and Name and Title Organization Chart), each added from a seed deck OPF writes (make-smartart-seeds.mjs: this PowerPoint build gives scripts no SmartArtLayouts list, so each layout is a seed graphic's SmartArt.Layout) and filled through SmartArt.Nodes; powerpoint-smartart.json is PowerPoint's own reading of every node. Every node's text imports, with no smartart-text-dropped. A slide whose SmartArt became overlay shapes has only its title besides them. Overlay shapes whose bounds cover at least a fifth of the slide (core overlaysDrawBody, OVERLAY_BODY_SHARE; front layer, placed by fractions) are the slide's body, so its title is a content slide's, on top, not a cover's centered over the shapes; a badge on a cover leaves the cover as it was. This holds for any slide, authored or imported (examples/technical/shape-overlays.opf.json slide 5's title moves to the top).

    SH-23: custom layout text algorithms set their LTR/RTL paragraph alignment defaults, so newly added nodes keep the design's alignment without a script changing their formatting after the edit.

    SH-23 qualification: five in-domain OPF designs retain an added ninth node and a text edit through SaveCopyAs/reopen, with unchanged font sizes, correct paragraph/first-line alignment and maximum geometry error0.0000876pt. The native text algorithms explicitly preserve start or center alignment for newly added nodes, which have no authored paragraph properties. Raw qualification and limits retain the initial visual failure and all corrected witnesses.

    Inverted pyramids

    Use the inverted-pyramid layout to narrow from a broad starting pool to a final choice. A custom pyramid region sets pyramidDirection: "down"; "up" is the existing default. The first entry stays at the top, so a downward pyramid starts wide and ends at the tip. Both the continuous outline and the stepped design use core geometry; neither reverses the text order or reduces the text size. PowerPoint export uses editable native shapes with a located smartart-unavailable finding. Live pyramid3 and downward OPF layout definitions remain unqualified. Import maps an Inverted Pyramid to this layout while reporting its adapted styling. See the example.

    Completed domain audit for 0.34 (SH20/22)

    The expanded diagnostic sweep contains 516 instances, including 465 forced-live observations. Replaying the unchanged gate on the original sealed PPTX, PNG and native probes with the current core interpreter reproduces every original metric and result: 201 full passes and 264 failures. All 465 node rectangles, rotations and font sizes agree; every failure includes native pixels/text, and three dense cases also wrap an extra line outside the node. Current shipped predicates admit 132 of these cases (102 built-in and 30 OPF-layout instances), all passing. These diagnostic measurements do not make every isolated passing case a safe domain.

    There is no additional safe live domain under the current text-density, stable-wrap, two-line and 500 px line-width limits. Ignoring only design/count/compact restrictions still leaves exactly the same 30 OPF-layout candidates. Ring/loop cycles, quadrants and compact forms therefore retain editable native-shape output and pptx/smartart-unavailable; the live domain does not expand in 0.34.

    Chinese/Japanese and Arabic/Hebrew remain outside live eligibility. Their complete measured live scopes are 18 Japanese, 18 Chinese, 15 Arabic and 15 Hebrew cases; the few passing Arabic/Hebrew instances do not qualify their other frames or text densities. Measured font slots agree, but visible baseline/horizontal offsets remain. Single-spacing and paragraph-marker experiments did not fix them and were rejected. The export preserves editable shapes/text at OPF's composed size with the explicit native-SmartArt refusal. No per-language coordinate adjustment, font-policy change or relaxed gate hides the remaining live limitation.