@agent-native/core-corpus 0.214.0 → 0.217.0

package/corpus/templates/content/acceptance/round-trip-corpus.ts

@@ -0,0 +1,1397 @@
+import type { PMNode } from "../shared/nfm";
+import type { NfmFidelityLossKind } from "../shared/nfm-fidelity";
+
+/**
+ * SF-04's round-trip corpus: Markdown an agent or the editor writes, with the
+ * canonical form it is stored in. L1 owns this fixture and adds every
+ * difference it fixes and every shape the census finds. L1's unit and
+ * property tests check every sample; the e2e round trip drives each sample
+ * that says how it renders. An id never changes meaning, so censuses can
+ * cite it.
+ */
+export type RoundTripSample = {
+  /** Stable id, used in test titles. */
+  id: string;
+  /** Who writes this shape: an agent through the API, or the editor's save. */
+  origin: "agent" | "editor";
+  /** The Markdown exactly as it is sent. */
+  markdown: string;
+  /**
+   * What a reader sees in the editor when the page opens, after a save and
+   * after export and import: an element inside the editor and its exact text.
+   * A sample stored with loss fails here even when its bytes look canonical.
+   * Samples without it are checked by the unit tests only, since each browser
+   * trip costs a page.
+   */
+  renders?: ReadonlyArray<{ selector: string; text: string }>;
+  note?: string;
+} & (
+  | {
+      /** The bytes the stored form holds: what the editor's live copy saves. */
+      canonical: string;
+      unrepresentable?: never;
+      storedToday?: never;
+    }
+  | {
+      /**
+       * Why the stored form cannot hold this sample. A sample with a reason
+       * must be refused at write time with a typed fidelity error; one without
+       * must be stored, so refusing it fails SF-04 like losing it would.
+       */
+      unrepresentable: string;
+      /** The losses canonicalizing reports, which the refusal names. */
+      losses: readonly NfmFidelityLossKind[];
+      canonical?: never;
+      storedToday?: never;
+    }
+  | {
+      /**
+       * The bytes stored today with no loss reported. Every word survives,
+       * but not in the structure the author meant, so they are not canonical:
+       * the fix for `followUp` changes them and moves the sample to
+       * `canonical`.
+       */
+      storedToday: string;
+      /** The recorded follow-up that changes `storedToday`. */
+      followUp: string;
+      canonical?: never;
+      unrepresentable?: never;
+    }
+);
+
+export const ROUND_TRIP_CORPUS: readonly RoundTripSample[] = [
+  {
+    id: "blank-line-paragraphs",
+    origin: "agent",
+    markdown: "First paragraph.\n\nSecond paragraph.",
+    canonical: "First paragraph.\nSecond paragraph.",
+    renders: [
+      { selector: "p", text: "First paragraph." },
+      { selector: "p", text: "Second paragraph." },
+    ],
+    note: "The editor writes one block per line.",
+  },
+  {
+    id: "escaped-underscore",
+    origin: "agent",
+    markdown: "Call snake\\_case once.",
+    canonical: "Call snake_case once.",
+    renders: [{ selector: "p", text: "Call snake_case once." }],
+  },
+  {
+    id: "underscore-bold",
+    origin: "agent",
+    markdown: "Some __bold__ text.",
+    canonical: "Some **bold** text.",
+    renders: [{ selector: "strong", text: "bold" }],
+  },
+  {
+    id: "underscore-italic",
+    origin: "agent",
+    markdown: "Some _italic_ text.",
+    canonical: "Some *italic* text.",
+    renders: [{ selector: "em", text: "italic" }],
+  },
+  {
+    id: "dollar-amount",
+    origin: "agent",
+    markdown: "It costs $60.",
+    canonical: "It costs \\$60.",
+    renders: [{ selector: "p", text: "It costs $60." }],
+  },
+  {
+    id: "four-space-nested-list",
+    origin: "agent",
+    markdown: "- Parent item\n    - Nested item",
+    canonical: "- Parent item\n\t- Nested item",
+    renders: [{ selector: "li li", text: "Nested item" }],
+  },
+  {
+    id: "two-space-nested-list",
+    origin: "agent",
+    markdown: "- Parent item\n  - Nested item",
+    canonical: "- Parent item\n\t- Nested item",
+    renders: [{ selector: "li li", text: "Nested item" }],
+  },
+  {
+    // The editor's own nested-list form, so a regression in the control shows.
+    id: "tab-nested-list",
+    origin: "editor",
+    markdown: "- Parent item\n\t- Nested item",
+    canonical: "- Parent item\n\t- Nested item",
+    renders: [{ selector: "li li", text: "Nested item" }],
+  },
+  {
+    id: "table-cell-edge-spaces",
+    origin: "agent",
+    markdown: "| Name | Price |\n| --- | --- |\n|  padded  | $5 |",
+    canonical:
+      '<table header-row="true">\n<tr>\n<td>Name</td>\n<td>Price</td>\n</tr>\n<tr>\n<td>padded</td>\n<td>\\$5</td>\n</tr>\n</table>',
+    renders: [
+      { selector: "td, th", text: "padded" },
+      { selector: "td, th", text: "$5" },
+    ],
+  },
+  {
+    id: "tilde-fence",
+    origin: "agent",
+    markdown: "~~~py\nprint(1)\n~~~",
+    canonical: "```py\nprint(1)\n```",
+    renders: [{ selector: "pre", text: "print(1)" }],
+  },
+  {
+    id: "html-table-th-header-row",
+    origin: "agent",
+    markdown:
+      "<table>\n<tr>\n<th>A</th>\n<th>B</th>\n</tr>\n<tr>\n<td>1</td>\n<td>2</td>\n</tr>\n</table>",
+    canonical:
+      '<table header-row="true">\n<tr>\n<td>A</td>\n<td>B</td>\n</tr>\n<tr>\n<td>1</td>\n<td>2</td>\n</tr>\n</table>',
+    // The editor draws a header cell as a td with this class, never as a th.
+    renders: [
+      { selector: "td.content-table-header-cell", text: "A" },
+      { selector: "td:not(.content-table-header-cell)", text: "2" },
+    ],
+  },
+  {
+    id: "space-nested-paragraph-after-blank",
+    origin: "agent",
+    markdown: "- a\n\n  more about a\n- b",
+    canonical: "- a\n\tmore about a\n- b",
+    renders: [{ selector: "li p", text: "more about a" }],
+  },
+  {
+    id: "bare-currency",
+    origin: "agent",
+    markdown: "costs $60 and $20,000",
+    canonical: "costs \\$60 and \\$20,000",
+  },
+  {
+    id: "github-inline-math",
+    origin: "agent",
+    markdown: "area $`\\pi r^2`$",
+    canonical: "area $\\pi r^2$",
+  },
+  {
+    id: "bare-inline-math",
+    origin: "agent",
+    markdown: "$x+y$ is math",
+    canonical: "$x+y$ is math",
+  },
+  {
+    id: "refuse-over-limit-inline-math",
+    origin: "agent",
+    markdown: `$${"x".repeat(1_001)}$`,
+    unrepresentable:
+      "The inline math matcher cannot represent a closed expression beyond its length limit.",
+    losses: ["dropped-formatting"],
+  },
+  {
+    id: "refuse-inline-image",
+    origin: "agent",
+    markdown: "see ![x](y) here",
+    unrepresentable:
+      "The editor supports image blocks, not images inside a paragraph.",
+    losses: ["dropped-formatting"],
+  },
+  {
+    id: "literal-image-prefix-beside-link",
+    origin: "agent",
+    markdown: String.raw`see \![x](y) here`,
+    canonical: String.raw`see \![x](y) here`,
+  },
+  {
+    id: "crlf-line-endings",
+    origin: "agent",
+    markdown: "A\r\nB\rC",
+    canonical: "A\nB\nC",
+  },
+  {
+    id: "self-closing-break",
+    origin: "agent",
+    markdown: "a<br/>b",
+    canonical: "a<br>b",
+  },
+  {
+    id: "star-plus-bullets",
+    origin: "agent",
+    markdown: "* a\n+ b",
+    canonical: "- a\n- b",
+  },
+  {
+    id: "paren-ordered-list",
+    origin: "agent",
+    markdown: "7) a\n9) b",
+    canonical: "7. a\n8. b",
+    note: "Only the first number is kept, as the list's start.",
+  },
+  {
+    id: "page-edge-newlines",
+    origin: "agent",
+    markdown: "\nText\n",
+    canonical: "Text",
+  },
+  {
+    id: "blank-line-runs",
+    origin: "agent",
+    markdown: "a\n\n\n\nb",
+    canonical: "a\nb",
+  },
+  {
+    id: "star-and-underscore-dividers",
+    origin: "agent",
+    markdown: "a\n\n***\n\nb\n\n___\n\nc",
+    canonical: "a\n---\nb\n---\nc",
+  },
+  {
+    id: "pipe-table",
+    origin: "agent",
+    markdown: "| A | B |\n|---|:-:|\n|  1 | 2  |\n| 3 |",
+    canonical:
+      '<table header-row="true">\n<tr>\n<td>A</td>\n<td align="center">B</td>\n</tr>\n<tr>\n<td>1</td>\n<td align="center">2</td>\n</tr>\n<tr>\n<td>3</td>\n<td align="center"></td>\n</tr>\n</table>',
+    note: "Cells are trimmed and ragged rows padded, as GFM reads them.",
+  },
+  {
+    id: "pipe-table-short-delimiter",
+    origin: "agent",
+    markdown: "| a |\n|-|\n| 1 |",
+    canonical:
+      '<table header-row="true">\n<tr>\n<td>a</td>\n</tr>\n<tr>\n<td>1</td>\n</tr>\n</table>',
+    note: "GFM needs one hyphen in a delimiter cell; main already took one when a colon aligned it.",
+  },
+  {
+    id: "unnecessary-escape-underscore",
+    origin: "agent",
+    markdown: "snake\\_case and 1\\. not a list",
+    canonical: "snake_case and 1. not a list",
+  },
+  {
+    id: "unnecessary-escape-punctuation",
+    origin: "agent",
+    markdown: "a\\-b \\# c \\! d",
+    canonical: "a-b # c ! d",
+  },
+  {
+    id: "escaped-ordered-marker",
+    origin: "agent",
+    markdown: "1\\. x",
+    canonical: "\\1. x",
+  },
+  {
+    id: "underscore-bold-italic",
+    origin: "agent",
+    markdown: "__bold__ and _it_ and ___both___",
+    canonical: "**bold** and *it* and ***both***",
+  },
+  {
+    id: "underscore-intraword",
+    origin: "agent",
+    markdown: "snake_case_name and foo__bar__ and _a_b_",
+    canonical: "snake_case_name and foo__bar__ and *a_b*",
+  },
+  {
+    id: "underscore-dunder",
+    origin: "agent",
+    markdown: "__init__ method",
+    canonical: "**init** method",
+    note: "CommonMark reads `__init__` as bold; code spans keep it literal.",
+  },
+  {
+    id: "underscore-nested-in-bold",
+    origin: "agent",
+    markdown: "**bold _nested_ text**",
+    canonical: "**bold *****nested***** text**",
+  },
+  {
+    id: "underscore-in-code-math-link",
+    origin: "agent",
+    markdown: "`a_b` $a_b$ [a_b](https://example.com/a_b)",
+    canonical: "`a_b` $a_b$ [a_b](https://example.com/a_b)",
+  },
+  {
+    id: "underscore-literal-editor-text",
+    origin: "editor",
+    markdown: "\\_\\_init\\_\\_",
+    canonical: "\\_\\_init\\_\\_",
+  },
+  {
+    id: "nested-marks-editor-text",
+    origin: "editor",
+    markdown: "**bold *****nested***** text**",
+    canonical: "**bold *****nested***** text**",
+  },
+  {
+    id: "special-characters",
+    origin: "agent",
+    markdown: "a > b {x} ^ | ~",
+    canonical: "a \\> b \\{x\\} \\^ \\| \\~",
+  },
+  {
+    id: "autolink-stays-text",
+    origin: "agent",
+    markdown: "<https://example.com>",
+    canonical: "\\<https://example.com\\>",
+  },
+  {
+    id: "github-alert-stays-quote",
+    origin: "agent",
+    markdown: "> [!TIP]\n> text",
+    canonical: "> \\[!TIP\\]\n> text",
+  },
+  {
+    id: "setext-heading-stays-text",
+    origin: "agent",
+    markdown: "Title\n===",
+    canonical: "Title\n===",
+  },
+  {
+    id: "setext-dash-heading-stays-text",
+    origin: "agent",
+    markdown: "Title\n--",
+    canonical: "Title\n--",
+  },
+  {
+    id: "setext-long-dash-heading-reads-as-divider",
+    origin: "agent",
+    markdown: "Title\n----",
+    canonical: "Title\n---",
+    note: "GFM reads a heading; Content reads a paragraph and a divider, as main does.",
+  },
+  {
+    id: "indented-code-stays-text",
+    origin: "agent",
+    markdown: "    code line\n    second",
+    canonical: "    code line\n    second",
+    note: "GFM reads indented code; Content keeps each line as a paragraph, spaces included.",
+  },
+  {
+    id: "indented-code-after-blank-stays-text",
+    origin: "agent",
+    markdown: "Para\n\n    code line",
+    canonical: "Para\n    code line",
+  },
+  {
+    id: "reference-link-stays-text",
+    origin: "agent",
+    markdown: "[text][ref]\n\n[ref]: https://example.com",
+    canonical: "\\[text\\]\\[ref\\]\n\\[ref\\]: https://example.com",
+  },
+  {
+    id: "backslash-hard-break-stays-text",
+    origin: "agent",
+    markdown: "a\\\nb",
+    canonical: "a\\\\\nb",
+    note: "GFM reads a hard break; Content keeps the backslash and starts a new paragraph.",
+  },
+  {
+    id: "toggle-summary-marks-stay-text",
+    origin: "agent",
+    markdown: '## **Toggle** {toggle="true"}\n\tbody',
+    canonical: '## **Toggle** {toggle="true"}\n\tbody',
+    note: "A toggle's summary is plain text, so its asterisks show.",
+  },
+  {
+    id: "italic-closed-by-spaced-star",
+    origin: "agent",
+    markdown: "*a * _b_",
+    canonical: "*a * *b*",
+    note: "Content closes the italic at the spaced `*`, as main does; GFM reads `*a *` as text.",
+  },
+  {
+    id: "space-nested-list-two",
+    origin: "agent",
+    markdown: "- a\n  - b\n  - c",
+    canonical: "- a\n\t- b\n\t- c",
+  },
+  {
+    id: "space-nested-list-four",
+    origin: "agent",
+    markdown: "- a\n    - b\n        - c\n    - d",
+    canonical: "- a\n\t- b\n\t\t- c\n\t- d",
+  },
+  {
+    id: "space-nested-ordered-three",
+    origin: "agent",
+    markdown: "1. a\n   1. b",
+    canonical: "1. a\n\t1. b",
+  },
+  {
+    id: "space-ordered-two-is-sibling",
+    origin: "agent",
+    markdown: "1. a\n  2. b",
+    canonical: "1. a\n2. b",
+    note: "Two spaces do not reach the text column of `1. `, so CommonMark makes a sibling.",
+  },
+  {
+    id: "space-nested-task",
+    origin: "agent",
+    markdown: "- [ ] a\n  - [x] b",
+    canonical: "- [ ] a\n\t- [x] b",
+  },
+  {
+    id: "space-nested-mixed-kinds",
+    origin: "agent",
+    markdown: "1. a\n    - b\n    - c\n2. d",
+    canonical: "1. a\n\t- b\n\t- c\n2. d",
+  },
+  {
+    id: "space-nested-wide-marker",
+    origin: "agent",
+    markdown: "1. a\n10. b\n    - c",
+    canonical: "1. a\n2. b\n\t- c",
+  },
+  {
+    id: "space-nested-fence",
+    origin: "agent",
+    markdown: "- a\n    ```js\n    x = 1\n\n    y = 2\n    ```\n- b",
+    canonical: "- a\n\t```js\n\tx = 1\n\t\n\ty = 2\n\t```\n- b",
+  },
+  {
+    id: "space-nested-fence-unclosed",
+    origin: "agent",
+    markdown: "- a\n  ```\n  x\n\nafter",
+    canonical: "- a\n  \\`\\`\\`\n  x\nafter",
+    note: "An unclosed fence runs to the page's end in the tab form, so one whose item ends first is not nested. It reads as on main.",
+  },
+  {
+    id: "space-nested-fence-to-page-end",
+    origin: "agent",
+    markdown: "- a\n  ```\n  x",
+    canonical: "- a\n\t```\n\tx\n\t```",
+  },
+  {
+    id: "space-nested-quote",
+    origin: "agent",
+    markdown: "- a\n  > q\n- b",
+    canonical: "- a\n\t> q\n- b",
+  },
+  {
+    id: "space-nested-after-blank-lines",
+    origin: "agent",
+    markdown: "- a\n\n\n    - b",
+    canonical: "- a\n\t- b",
+  },
+  {
+    id: "space-lazy-continuation",
+    origin: "agent",
+    markdown: "- a\n  more",
+    canonical: "- a\n  more",
+    note: "Without a blank line CommonMark joins this to the item's text; it stays its own paragraph.",
+  },
+  {
+    id: "space-nested-under-tab-nested",
+    origin: "agent",
+    markdown: "- a\n\t- b\n  - c",
+    canonical: "- a\n\t- b\n\t- c",
+  },
+  {
+    id: "space-nested-in-details",
+    origin: "agent",
+    markdown: "<details>\n<summary>S</summary>\n- a\n    - b\n</details>",
+    canonical: "<details>\n<summary>S</summary>\n\t- a\n\t\t- b\n</details>",
+  },
+  {
+    id: "space-indented-first-item",
+    origin: "agent",
+    markdown: "  - leading spaces item",
+    canonical: "- leading spaces item",
+  },
+  {
+    id: "editor-paragraph-with-spaced-marker",
+    origin: "editor",
+    markdown: "\\  - not a list",
+    canonical: "\\  - not a list",
+  },
+  {
+    id: "tab-indented-spaced-marker-stays-text",
+    origin: "agent",
+    markdown: "Intro\n\t  - item\n\tafter",
+    canonical: "Intro\n\t\\  - item\n\tafter",
+    note: "A list one tab in keeps no indent, so the indented line after it would join the item on the next read.",
+  },
+  {
+    id: "untabbed-container-body",
+    origin: "agent",
+    markdown: "<details>\n<summary>S</summary>\nBody\n</details>",
+    canonical: "<details>\n<summary>S</summary>\n\tBody\n</details>",
+  },
+  {
+    id: "untabbed-callout-body",
+    origin: "agent",
+    markdown: "Hello\n<callout>\nOnly text\n</callout>",
+    canonical: "Hello\n<callout>\n\tOnly text\n</callout>",
+    renders: [
+      { selector: "[data-notion-callout-content] p", text: "Only text" },
+    ],
+  },
+  {
+    id: "untabbed-callout-body-with-attributes",
+    origin: "agent",
+    markdown:
+      '<callout icon="💡" color="blue_bg">\nFirst line\n- one\n- two\n</callout>',
+    canonical:
+      '<callout icon="💡" color="blue_bg">\n\tFirst line\n\t- one\n\t- two\n</callout>',
+  },
+  {
+    id: "callout-body-partly-tabbed",
+    origin: "agent",
+    markdown: '<callout icon="i">\n\tkept\nalso kept\n</callout>',
+    canonical: '<callout icon="i">\n\tkept\n\talso kept\n</callout>',
+  },
+  {
+    id: "untabbed-callout-code-block",
+    origin: "agent",
+    markdown: "<callout>\n```js\nconst a = 1;\n```\n</callout>",
+    canonical: "<callout>\n\t```js\n\tconst a = 1;\n\t```\n</callout>",
+  },
+  {
+    id: "untabbed-nested-callouts",
+    origin: "agent",
+    markdown: "<callout>\n<callout>\nInner\n</callout>\nOuter\n</callout>",
+    canonical:
+      "<callout>\n\t<callout>\n\t\tInner\n\t</callout>\n\tOuter\n</callout>",
+  },
+  {
+    id: "untabbed-columns",
+    origin: "agent",
+    markdown:
+      "<columns>\n<column>\nLeft\n</column>\n<column>\nRight\n</column>\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\tLeft\n\t</column>\n\t<column>\n\t\tRight\n\t</column>\n</columns>",
+  },
+  {
+    id: "column-body-partly-tabbed",
+    origin: "agent",
+    markdown:
+      "<columns>\n\t<column>\n\t\tkept\n\talso kept\n\t</column>\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\tkept\n\t\talso kept\n\t</column>\n</columns>",
+  },
+  {
+    id: "untabbed-synced-block-body",
+    origin: "agent",
+    markdown:
+      '<synced_block url="https://www.notion.so/abc">\nShared text\n</synced_block>',
+    canonical:
+      '<synced_block url="https://www.notion.so/abc">\n\tShared text\n</synced_block>',
+  },
+  {
+    id: "untabbed-callout-nested-list",
+    origin: "agent",
+    markdown: "<callout>\n- parent\n\t- child\n</callout>",
+    canonical: "<callout>\n\t- parent\n\t\t- child\n</callout>",
+    note: "Lines nested under an untabbed child stay nested under it.",
+  },
+  {
+    id: "callout-math-block-with-container-tag",
+    origin: "agent",
+    markdown:
+      "<callout>\n\t$$\n\t<callout>\n\tx\n\t</callout>\n\t$$\n</callout>",
+    canonical:
+      "<callout>\n\t$$\n\t<callout>\n\tx\n\t</callout>\n\t$$\n</callout>",
+    note: "A container tag inside an equation is equation source.",
+  },
+  {
+    id: "untabbed-callout-math-block-with-container-tag",
+    origin: "agent",
+    markdown: "<callout>\n$$\n<callout>\n\tx\n</callout>\n$$\n</callout>",
+    canonical:
+      "<callout>\n\t$$\n\t<callout>\n\t\tx\n\t</callout>\n\t$$\n</callout>",
+  },
+  {
+    id: "untabbed-callout-code-block-with-container-tag",
+    origin: "agent",
+    markdown: "<callout>\n```\n<callout>\nx\n</callout>\n```\n</callout>",
+    canonical:
+      "<callout>\n\t```\n\t<callout>\n\tx\n\t</callout>\n\t```\n</callout>",
+  },
+  {
+    id: "untabbed-nested-callout-code-block",
+    origin: "agent",
+    markdown:
+      "<callout>\n\t<callout>\n\t```\nx\n\t```\n\t</callout>\n</callout>",
+    canonical:
+      "<callout>\n\t<callout>\n\t\t```\n\t\tx\n\t\t```\n\t</callout>\n</callout>",
+    note: "A code block keeps the lines it reads shallower than itself, and its close.",
+  },
+  {
+    id: "untabbed-nested-callout-math-block",
+    origin: "agent",
+    markdown: "<callout>\n\t<callout>\n\t$$\nx\n\t$$\n\t</callout>\n</callout>",
+    canonical:
+      "<callout>\n\t<callout>\n\t\t$$\n\t\tx\n\t\t$$\n\t</callout>\n</callout>",
+  },
+  {
+    id: "untabbed-nested-callout-meeting-notes",
+    origin: "agent",
+    markdown:
+      "<callout>\n\t<callout>\n\t<meeting-notes>\nx\n\t</meeting-notes>\n\t</callout>\n</callout>",
+    canonical:
+      "<callout>\n\t<callout>\n\t\t<meeting-notes>\n\t\tx\n\t\t</meeting-notes>\n\t</callout>\n</callout>",
+  },
+  {
+    id: "untabbed-callout-three-deep",
+    origin: "agent",
+    markdown:
+      "<callout>\n\t<callout>\n\t<callout>\nx\n\t</callout>\n\t</callout>\n</callout>",
+    canonical:
+      "<callout>\n\t<callout>\n\t\t<callout>\n\t\t\tx\n\t\t</callout>\n\t</callout>\n</callout>",
+  },
+  {
+    id: "untabbed-column-list-code-block",
+    origin: "agent",
+    markdown:
+      "<columns>\n\t<column>\n\t- a\n\t\t```\nx\n\t\t```\n\t</column>\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\t- a\n\t\t\t```\n\t\t\tx\n\t\t\t```\n\t</column>\n</columns>",
+  },
+  {
+    id: "untabbed-callout-fence-close-below-opener",
+    origin: "agent",
+    markdown: "<callout>\n- x\n\t```\ncode\n```\nafter\n</callout>",
+    storedToday:
+      "<callout>\n\t- x\n\t\t````\n\t\tcode\n\t\t```\n\t\tafter\n\t\t````\n</callout>",
+    followUp:
+      "A close shallower than its fence is code text, as in the same lines outside a callout, so `after` reads as code and the fence ends at `</callout>`.",
+  },
+  {
+    id: "untabbed-details-with-tabbed-summary-in-callout",
+    origin: "agent",
+    markdown:
+      "<callout>\n\tkept\n<details>\n\t<summary>S</summary>\n\tD\n</details>\n</callout>",
+    storedToday:
+      "<callout>\n\tkept\n\t<details>\n\t<summary></summary>\n\t\t\\<summary\\>S\\</summary\\>\n\t\tD\n\t</details>\n</callout>",
+    followUp:
+      "A summary one tab deeper than its opener is body text, as on main, so the toggle has no title.",
+  },
+  {
+    id: "untabbed-callout-ordered-list-continues",
+    origin: "agent",
+    markdown: "<callout>\n\t1. a\n1. b\n</callout>",
+    canonical: "<callout>\n\t1. a\n\t2. b\n</callout>",
+    note: "An untabbed item continues the list above it.",
+  },
+  {
+    id: "untabbed-column-task-list-continues",
+    origin: "agent",
+    markdown:
+      "<columns>\n\t<column>\n\t\t- [ ] a\n\t- [x] b\n\t</column>\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\t- [ ] a\n\t\t- [x] b\n\t</column>\n</columns>",
+  },
+  {
+    id: "columns-child-before-any-column",
+    origin: "agent",
+    markdown: "<columns>\n\tstray\n\t<column>\n\t\ta\n\t</column>\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\tstray\n\t</column>\n\t<column>\n\t\ta\n\t</column>\n</columns>",
+    note: "A child before any column gets a column of its own.",
+  },
+  {
+    id: "columns-child-after-a-column",
+    origin: "agent",
+    markdown: "<columns>\n\t<column>\n\t\ta\n\t</column>\n\tafter\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\ta\n\t\tafter\n\t</column>\n</columns>",
+    note: "A child after a column joins it.",
+  },
+  {
+    id: "columns-list-after-a-column-continues",
+    origin: "agent",
+    markdown: "<columns>\n\t<column>\n\t\t- a\n\t</column>\n\t- b\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\t- a\n\t\t- b\n\t</column>\n</columns>",
+  },
+  {
+    id: "math-block-open-at-column-close",
+    origin: "agent",
+    markdown: "<columns>\n\t<column>\n\t\t$$\n\t</column>\n\traw\n</columns>",
+    canonical:
+      "<columns>\n\t<column>\n\t\t$$\n\t\t\n\t\t$$\n\t\traw\n\t</column>\n</columns>",
+    note: "The equation ends at its column's close; the line after it joins that column.",
+  },
+  {
+    id: "math-block-open-at-columns-close",
+    origin: "agent",
+    markdown: "<columns>\n\t$$\n</columns>\nAfter\n$$",
+    canonical:
+      "<columns>\n\t<column>\n\t\t$$\n\t\t\n\t\t$$\n\t</column>\n</columns>\nAfter\n$$\n\n$$",
+  },
+  {
+    id: "math-block-open-at-callout-close",
+    origin: "agent",
+    markdown: "<callout>\n\t$$\n\tx = 1\n</callout>\nAfter\n$$",
+    canonical: "<callout>\n\t$$\n\tx = 1\n\t$$\n</callout>\nAfter\n$$\n\n$$",
+    note: "A block inside a container ends at the container's closing tag.",
+  },
+  {
+    id: "table-open-at-callout-close",
+    origin: "agent",
+    markdown:
+      "<callout>\n\t<table>\n\t<tr>\n\t<td>a</td>\n\t</tr>\n</callout>\nAfter\n\t</table>",
+    canonical:
+      "<callout>\n\t\\<table\\>\n\t\\<tr\\>\n\t\\<td\\>a\\</td\\>\n\t\\</tr\\>\n</callout>\nAfter\n\t\\</table\\>",
+    note: "A table without its closing tag inside the container stays as text.",
+  },
+  {
+    id: "unclosed-container",
+    origin: "agent",
+    markdown: '<callout icon="x">\nHello',
+    canonical: '\\<callout icon="x"\\>\nHello',
+    note: "The tags stay as text.",
+  },
+  {
+    id: "quoted-boolean-open",
+    origin: "agent",
+    markdown: '<details open="">\n<summary>S</summary>\n\tB\n</details>',
+    canonical: "<details open>\n<summary>S</summary>\n\tB\n</details>",
+  },
+  {
+    id: "mention-raw-ampersand",
+    origin: "agent",
+    markdown: '<mention-page url="https://example.com/p">R&D</mention-page>',
+    canonical:
+      '<mention-page url="https://example.com/p">R&amp;D</mention-page>',
+  },
+  {
+    id: "multi-line-quote",
+    origin: "agent",
+    markdown: "> a\n> b",
+    canonical: "> a\n> b",
+    note: "Each `> ` line is its own quote block.",
+  },
+  {
+    id: "loose-list",
+    origin: "agent",
+    markdown: "- a\n\n- b",
+    canonical: "- a\n- b",
+  },
+  {
+    id: "empty-emphasis",
+    origin: "agent",
+    markdown: "x*** and ** ** and ~~~~",
+    canonical: "x**\\* and ** \\*\\* and \\~\\~\\~\\~",
+    note: "Delimiters around nothing stay text.",
+  },
+  {
+    id: "long-backtick-fence",
+    origin: "agent",
+    markdown: "``````ts\ncode\n``````",
+    canonical: "```ts\ncode\n```",
+  },
+  {
+    id: "tilde-fence-holds-backticks",
+    origin: "agent",
+    markdown: "~~~\n```\ninner\n```\n~~~",
+    canonical: "````\n```\ninner\n```\n````",
+  },
+  {
+    id: "code-language-with-backtick",
+    origin: "agent",
+    markdown: "~~~a`b\nx\n~~~",
+    canonical: "~~~a`b\nx\n~~~",
+    note: "A backtick fence's language cannot hold a backtick.",
+  },
+  {
+    id: "code-language-with-backtick-after-tilde",
+    origin: "agent",
+    markdown: "~~~ ~a`\nx\n~~~",
+    canonical: "~~~ ~a`\nx\n~~~",
+    note: "A tilde starting the language would lengthen the fence.",
+  },
+  {
+    id: "long-code-span-at-line-start",
+    origin: "editor",
+    markdown: "``` `` ```",
+    canonical: "``` `` ```",
+    note: "A backtick fence's language holds no backtick, so this is a code span.",
+  },
+  {
+    id: "emphasis-around-code-span",
+    origin: "editor",
+    markdown: "*`*args`*",
+    canonical: "*`*args`*",
+    note: "Code spans bind tighter than emphasis.",
+  },
+  {
+    id: "emphasis-around-code-span-in-bold",
+    origin: "agent",
+    markdown: "**a `*` b** and *`a_b`*",
+    canonical: "**a ****`*`**** b** and *`a_b`*",
+  },
+  {
+    id: "bold-inside-italic",
+    origin: "agent",
+    markdown: "*Run **npm** install* and *a **b***",
+    canonical: "*Run ****npm**** install* and *a ****b***",
+    note: "A `**` that can only open is bold inside the italic.",
+  },
+  {
+    id: "italic-space-opening-block",
+    origin: "editor",
+    markdown: "\\* a*",
+    canonical: "\\* a*",
+    note: "The editor escapes a star that opens italics at the start of a block.",
+  },
+  {
+    id: "link-around-code-with-bracket",
+    origin: "editor",
+    markdown: "[`a[`](https://example.com)",
+    canonical: "[`a[`](https://example.com)",
+    note: "Code spans bind tighter than link brackets.",
+  },
+  {
+    id: "math-before-digit",
+    origin: "editor",
+    markdown: "$`x`$1 and $`a`$$b$",
+    canonical: "$`x`$1 and $`a`$$b$",
+    note: "Plain math does not close before a digit or another `$`.",
+  },
+  {
+    id: "html-table-th-header-column",
+    origin: "agent",
+    markdown:
+      "<table>\n<tr>\n<th>R1</th>\n<td>1</td>\n</tr>\n<tr>\n<th>R2</th>\n<td>2</td>\n</tr>\n</table>",
+    canonical:
+      '<table header-column="true">\n<tr>\n<td>R1</td>\n<td>1</td>\n</tr>\n<tr>\n<td>R2</td>\n<td>2</td>\n</tr>\n</table>',
+  },
+  {
+    id: "html-table-thead-one-line-rows",
+    origin: "agent",
+    markdown:
+      "<table>\n<thead>\n<tr><th>A</th><th>B</th></tr>\n</thead>\n<tbody>\n<tr><td>1</td><td>2</td></tr>\n</tbody>\n</table>",
+    canonical:
+      '<table header-row="true">\n<tr>\n<td>A</td>\n<td>B</td>\n</tr>\n<tr>\n<td>1</td>\n<td>2</td>\n</tr>\n</table>',
+  },
+  {
+    id: "html-table-multi-line-cell",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>\nfirst\nsecond\n</td>\n</tr>\n</table>",
+    canonical: "<table>\n<tr>\n<td>first second</td>\n</tr>\n</table>",
+  },
+  {
+    id: "html-table-cells-one-line",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a</td><td>b</td>\n</tr>\n</table>",
+    canonical: "<table>\n<tr>\n<td>a</td>\n<td>b</td>\n</tr>\n</table>",
+  },
+  {
+    id: "html-table-cells-one-line-with-prices",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>$5</td><td>a\\b</td>\n</tr>\n</table>",
+    canonical: "<table>\n<tr>\n<td>\\$5</td>\n<td>a\\\\b</td>\n</tr>\n</table>",
+    note: "A dollar sign or a backslash does not hide the close tag after it.",
+  },
+  {
+    id: "html-table-close-tag-in-code",
+    origin: "editor",
+    markdown: "<table>\n<tr>\n<td>`</td>`</td>\n</tr>\n</table>",
+    canonical: "<table>\n<tr>\n<td>`</td>`</td>\n</tr>\n</table>",
+    note: "A close tag in code is the cell's text.",
+  },
+  {
+    id: "html-table-row-closed-on-cell-line",
+    origin: "agent",
+    markdown:
+      "<table>\n<tr>\n<td>a</td></tr>\n<tr>\n<td>b</td>\n</tr>\n</table>",
+    canonical:
+      "<table>\n<tr>\n<td>a</td>\n</tr>\n<tr>\n<td>b</td>\n</tr>\n</table>",
+    note: "A row closed on its last cell's line ends there.",
+  },
+  {
+    id: "html-table-row-closed-after-wrapped-cell",
+    origin: "agent",
+    markdown:
+      "<table>\n<tr>\n<td>a\nb</td></tr>\n<tr>\n<td>c</td>\n</tr>\n</table>",
+    canonical:
+      "<table>\n<tr>\n<td>a b</td>\n</tr>\n<tr>\n<td>c</td>\n</tr>\n</table>",
+    note: "A cell wrapped over lines may close its row on its last line.",
+  },
+  {
+    id: "html-table-wrapped-cell-beside-row-tag",
+    origin: "agent",
+    markdown: "<table>\n<tr><td>first\nsecond</td></tr>\n</table>",
+    canonical: "<table>\n<tr>\n<td>first second</td>\n</tr>\n</table>",
+    note: "A cell may open on its row's line and wrap past it.",
+  },
+  {
+    id: "html-table-rows-one-line",
+    origin: "agent",
+    markdown:
+      '<table>\n<tr><td>a</td></tr><tr color="red"><td>b</td></tr>\n</table>',
+    canonical:
+      '<table>\n<tr>\n<td>a</td>\n</tr>\n<tr color="red">\n<td>b</td>\n</tr>\n</table>',
+  },
+  {
+    id: "html-table-row-break-between-cells",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a</td></tr><tr><td>b</td>\n</tr>\n</table>",
+    canonical:
+      "<table>\n<tr>\n<td>a</td>\n</tr>\n<tr>\n<td>b</td>\n</tr>\n</table>",
+  },
+  {
+    id: "html-table-row-left-open",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a</td>\n<tr>\n<td>b</td>\n</tr>\n</table>",
+    canonical:
+      "<table>\n<tr>\n<td>a</td>\n</tr>\n<tr>\n<td>b</td>\n</tr>\n</table>",
+    note: "HTML ends a row where the next one starts. Main read both rows as one.",
+  },
+  {
+    id: "html-table-row-left-open-one-line",
+    origin: "agent",
+    markdown: "<table>\n<tr><td>a</td><tr><td>b</td></tr>\n</table>",
+    canonical:
+      "<table>\n<tr>\n<td>a</td>\n</tr>\n<tr>\n<td>b</td>\n</tr>\n</table>",
+  },
+  {
+    id: "html-table-row-left-open-at-end",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a</td>\n</table>",
+    canonical: "<table>\n<tr>\n<td>a</td>\n</tr>\n</table>",
+    note: "Main read this table as its literal tags.",
+  },
+  {
+    id: "html-table-thead-row",
+    origin: "agent",
+    markdown:
+      "<table>\n<thead>\n<tr>\n<td>A</td>\n</tr>\n</thead>\n<tbody>\n<tr>\n<td>b</td>\n</tr>\n</tbody>\n</table>",
+    canonical:
+      '<table header-row="true">\n<tr>\n<td>A</td>\n</tr>\n<tr>\n<td>b</td>\n</tr>\n</table>',
+    note: "A `<thead>` around the first row names the header row, as `<th>` cells do. Main read it as a body row.",
+  },
+  {
+    id: "html-table-cell-edge-spaces",
+    origin: "editor",
+    markdown: "<table>\n<tr>\n<td> a </td>\n</tr>\n</table>",
+    canonical: "<table>\n<tr>\n<td> a </td>\n</tr>\n</table>",
+    note: "HTML cells keep their spaces, so a cell the editor holds as ` a ` reads back the same.",
+  },
+  {
+    id: "heading-extra-spaces",
+    origin: "agent",
+    markdown: "#   Title",
+    canonical: "# Title",
+  },
+  {
+    id: "empty-span",
+    origin: "agent",
+    markdown: "<span>plain</span> text",
+    canonical: "plain text",
+  },
+  {
+    id: "unknown-inline-tag-stays-text",
+    origin: "agent",
+    markdown: "a <kbd>Ctrl</kbd> b",
+    canonical: "a \\<kbd\\>Ctrl\\</kbd\\> b",
+  },
+  {
+    id: "code-span-whitespace-only",
+    origin: "editor",
+    markdown: "`\t `",
+    canonical: "`\t `",
+    note: "Padding is stripped only around code that is not all whitespace, so none is added.",
+  },
+  {
+    id: "whitespace-only-text",
+    origin: "editor",
+    markdown: "<span>  </span>",
+    canonical: "<span>  </span>",
+    note: "A neutral span protects the editor's whitespace from block trimming.",
+  },
+  {
+    id: "refuse-invalid-block-color",
+    origin: "agent",
+    markdown: 'text {color="bad"}',
+    unrepresentable:
+      "An unsupported block color must be reported even when the suffix stays literal text.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "compact-table-head-group",
+    origin: "agent",
+    markdown: "<table>\n<thead><tr><th>A</th></tr></thead>\n</table>",
+    canonical: '<table header-row="true">\n<tr>\n<td>A</td>\n</tr>\n</table>',
+  },
+  {
+    id: "compact-table-body-group",
+    origin: "agent",
+    markdown: "<table>\n<tbody><tr><td>x</td></tr></tbody>\n</table>",
+    canonical: "<table>\n<tr>\n<td>x</td>\n</tr>\n</table>",
+  },
+  {
+    id: "empty-code-fence",
+    origin: "agent",
+    markdown: "```\n```",
+    canonical: "```\n\n```",
+  },
+  {
+    id: "colored-paragraph-holding-a-marker",
+    origin: "editor",
+    markdown: '#{color="red"}\n-{color="red"}\n1.{color="red"}',
+    canonical: '#{color="red"}\n-{color="red"}\n1.{color="red"}',
+    note: "With a space before the color, the text would read as a heading or list item.",
+  },
+  {
+    id: "trailing-empty-block-after-heading",
+    origin: "agent",
+    markdown: "# H\n<empty-block/>",
+    canonical: "# H",
+    note: "The editor adds an empty paragraph after a last block of another kind, and its save drops it.",
+  },
+  {
+    id: "refuse-colored-trailing-empty-block",
+    origin: "agent",
+    markdown: '# H\n{color="red"}',
+    canonical: '# H\n{color="red"}',
+    note: "The color suffix keeps an authored empty paragraph distinct from the editor's terminal filler.",
+  },
+  {
+    id: "refuse-formatted-math",
+    origin: "agent",
+    markdown: "**$a*b$**",
+    unrepresentable:
+      "Yjs keeps marks only on text, so the editor cannot hold formatting on math.",
+    losses: ["dropped-formatting"],
+  },
+  {
+    id: "refuse-formatted-mention",
+    origin: "agent",
+    markdown:
+      '[<mention-page url="https://example.com/p">P</mention-page>](https://example.com) and <span color="red">*$x$*</span>',
+    unrepresentable:
+      "Yjs keeps marks only on text, so the editor cannot hold formatting on a mention or math.",
+    losses: ["dropped-formatting"],
+  },
+  {
+    id: "refuse-empty-block-trailing-text",
+    origin: "agent",
+    markdown: "<empty-block/> tail",
+    unrepresentable: "An empty block holds no text.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-span-unknown-attribute",
+    origin: "agent",
+    markdown: '<span color="red" extra="retained">t</span>',
+    unrepresentable: "A span keeps only color, background and underline.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-table-colspan",
+    origin: "agent",
+    markdown: '<table>\n<tr>\n<td colspan="2">x</td>\n</tr>\n</table>',
+    unrepresentable: "Table cells do not span columns.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-table-stray-text",
+    origin: "agent",
+    markdown: "<table>\nstray\n<tr>\n<td>x</td>\n</tr>\n</table>",
+    unrepresentable: "A table holds only rows.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-mention-label-edge-spaces",
+    origin: "agent",
+    markdown:
+      'See <mention-page url="https://example.com/p"> Plan </mention-page>.',
+    unrepresentable: "A mention's label has no spaces at its edges.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-table-cell-open-at-row-end",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a\n</tr>\n<tr>\n<td>b</td>\n</tr>\n</table>",
+    unrepresentable: "A cell closes inside its row.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-table-cell-open-before-next-cell",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a\n<td>b</td>\n</tr>\n</table>",
+    unrepresentable: "A cell closes before the next one opens.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-table-cell-open-before-next-cell-on-line",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a <td>b</td>\n</tr>\n</table>",
+    unrepresentable: "A cell closes before the next one opens.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-span-underline-not-boolean",
+    origin: "agent",
+    markdown: '<span color="red" underline="yes">x</span>',
+    unrepresentable: 'A span\'s underline is "true" or "false".',
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-table-thead-two-rows",
+    origin: "agent",
+    markdown:
+      "<table>\n<thead>\n<tr>\n<td>A</td>\n</tr>\n<tr>\n<td>B</td>\n</tr>\n</thead>\n</table>",
+    unrepresentable: "The editor has one header row.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-table-tfoot",
+    origin: "agent",
+    markdown:
+      "<table>\n<tr>\n<td>a</td>\n</tr>\n<tfoot>\n<tr>\n<td>f</td>\n</tr>\n</tfoot>\n</table>",
+    unrepresentable: "The editor has no footer rows.",
+    losses: ["dropped-source"],
+  },
+  {
+    id: "refuse-table-header-flag-not-boolean",
+    origin: "agent",
+    markdown: '<table header-row="yes">\n<tr>\n<td>a</td>\n</tr>\n</table>',
+    unrepresentable: 'A table flag is "true" or "false".',
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-details-open-false",
+    origin: "agent",
+    markdown:
+      '<details open="false">\n<summary>S</summary>\n\tbody\n</details>',
+    unrepresentable:
+      'HTML reads any value of `open` as open, so `open="false"` is a toggle that is open.',
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-table-th-in-body",
+    origin: "agent",
+    markdown: "<table>\n<tr>\n<td>a</td>\n<th>b</th>\n</tr>\n</table>",
+    unrepresentable:
+      "Header cells are a whole first row or a whole first column.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-table-unknown-attribute",
+    origin: "agent",
+    markdown: '<table style="x">\n<tr>\n<td>a</td>\n</tr>\n</table>',
+    unrepresentable: "A table has no style.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-callout-unknown-attribute",
+    origin: "agent",
+    markdown: '<callout icon="x" kind="warn">\n\tbody\n</callout>',
+    unrepresentable: "A callout keeps only its icon and color.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-mention-unquoted-attribute",
+    origin: "agent",
+    markdown: "<mention-page url=abc>Page</mention-page>",
+    unrepresentable: "Only double-quoted attributes are read.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-file-bare-attribute",
+    origin: "agent",
+    markdown: '<file url="x" bogus/>',
+    unrepresentable: "An attribute without a value is not read.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-audio-unknown-color",
+    origin: "agent",
+    markdown: '<audio src="x" color="bad"/>',
+    unrepresentable: "Only the editor's named colors are kept.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-details-line-below-body-after-equation-fence-line",
+    origin: "agent",
+    markdown: "\t\t<details>\n$$\n\t```\n\t$$\ncode\n```\n\t\t</details>",
+    unrepresentable:
+      "The details reader takes the fence line in the equation for a fence, so `code` stays below its body and is not read.",
+    losses: ["dropped-source", "reparse-mismatch"],
+  },
+  {
+    id: "refuse-table-line-after-math-block-open-at-callout-close",
+    origin: "agent",
+    markdown: "<callout>\n\t$$\n</callout>\n<table>\nx\n</table>",
+    unrepresentable:
+      "The equation ends at `</callout>`, and the table reader drops `x`, a line that isn't a row, as it does for the same three lines alone.",
+    losses: ["dropped-source", "invalid-structure"],
+  },
+  {
+    id: "refuse-untabbed-callout-paragraph-then-deeper-fence",
+    origin: "agent",
+    markdown: "<callout>\nx\n\t```\ncode\n```\nafter\n</callout>",
+    unrepresentable:
+      "The editor saves a code block one tab under a paragraph at the paragraph's depth, as on main; every line, `after` included, is kept.",
+    losses: ["editor-mismatch"],
+  },
+  {
+    id: "refuse-columns-empty",
+    origin: "agent",
+    markdown: "<columns>\n</columns>",
+    unrepresentable: "Columns need at least one column.",
+    losses: ["invalid-structure"],
+  },
+  {
+    id: "refuse-video-poster",
+    origin: "agent",
+    markdown: '<video src="a.mp4" poster="p.png"></video>',
+    unrepresentable: "The editor's video block has no poster.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-video-color",
+    origin: "agent",
+    markdown: '<video src="a.mp4" color="red"></video>',
+    unrepresentable: "The editor's video block has no color.",
+    losses: ["editor-mismatch"],
+  },
+  {
+    id: "refuse-image-color",
+    origin: "agent",
+    markdown: '![a](b.png) {color="red"}',
+    unrepresentable: "The editor's image block has no color.",
+    losses: ["editor-mismatch"],
+  },
+  {
+    id: "refuse-image-width",
+    origin: "agent",
+    markdown: '![a](b.png) {width="200" color="red"}',
+    unrepresentable: "The editor's image block has no width or color.",
+    losses: ["dropped-attribute", "editor-mismatch"],
+  },
+  {
+    id: "refuse-paragraph-toggle",
+    origin: "agent",
+    markdown: 'text {toggle="true"}',
+    unrepresentable: "Only headings can be toggles.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-paragraph-extra-attribute",
+    origin: "agent",
+    markdown: 'text {color="red" size="2"}',
+    unrepresentable: "A paragraph keeps only its color.",
+    losses: ["dropped-attribute"],
+  },
+  {
+    id: "refuse-heading-toggle-level-five",
+    origin: "agent",
+    markdown: '##### H {toggle="true"}\n\tchild',
+    unrepresentable: "Toggle headings go to level 4.",
+    losses: ["reparse-mismatch"],
+  },
+];
+
+/**
+ * Editor documents whose Markdown reads back as a different document. Each
+ * is a loss main has too; the test fails when one is fixed, so its entry
+ * moves into the corpus instead of going stale.
+ */
+export interface KnownEditorLoss {
+  id: string;
+  doc: PMNode;
+  /** What the editor's live copy saves for `doc`. */
+  markdown: string;
+  note: string;
+}
+
+const text = (value: string, marks?: PMNode["marks"]): PMNode =>
+  marks ? { type: "text", text: value, marks } : { type: "text", text: value };
+const paragraph = (value: string): PMNode => ({
+  type: "paragraph",
+  content: [text(value)],
+});
+const bulletList = (value: string): PMNode => ({
+  type: "bulletList",
+  content: [{ type: "listItem", content: [paragraph(value)] }],
+});
+const mathParagraph = (label: string): PMNode => ({
+  type: "paragraph",
+  content: [
+    {
+      type: "notionInlineAtom",
+      attrs: { tagName: "math", attrsJson: "{}", label },
+    },
+  ],
+});
+
+export const KNOWN_EDITOR_LOSSES: readonly KnownEditorLoss[] = [
+  {
+    id: "heading-leading-space",
+    doc: {
+      type: "doc",
+      content: [
+        { type: "heading", attrs: { level: 2 }, content: [text(" a")] },
+      ],
+    },
+    markdown: "##  a",
+    note: "A heading's text is trimmed, as CommonMark reads it.",
+  },
+  {
+    id: "literal-star-opening-block",
+    doc: {
+      type: "doc",
+      content: [
+        {
+          type: "paragraph",
+          content: [text("* "), text(";", [{ type: "italic" }])],
+        },
+      ],
+    },
+    markdown: "\\* *;*",
+    note: "A literal star is written `\\*`, and so is a star opening italics at the start of a block. Stored pages read it as italics, so this one changes.",
+  },
+  {
+    id: "adjacent-lists-of-one-kind",
+    doc: { type: "doc", content: [bulletList("a"), bulletList("b")] },
+    markdown: "- a\n- b",
+    note: "Two lists of one kind in a row read back as one list. The editor joins them as they are typed.",
+  },
+  {
+    id: "underline-mark",
+    doc: {
+      type: "doc",
+      content: [
+        {
+          type: "paragraph",
+          content: [text("u", [{ type: "underline" }])],
+        },
+      ],
+    },
+    markdown: '<span underline="true">u</span>',
+    note: "Underlined text reads back as an underlined span, which looks the same.",
+  },
+  {
+    id: "formatted-inline-math",
+    doc: {
+      type: "doc",
+      content: [
+        {
+          type: "paragraph",
+          content: [
+            {
+              type: "notionInlineAtom",
+              attrs: { tagName: "math", attrsJson: "{}", label: "x" },
+              marks: [{ type: "bold" }],
+            },
+          ],
+        },
+      ],
+    },
+    markdown: "$x$",
+    note: "The editor can format math or a mention, but Yjs keeps marks only on text, so the live copy never has the formatting to save.",
+  },
+  {
+    id: "math-label-closing-early",
+    doc: { type: "doc", content: [mathParagraph("a`$b")] },
+    markdown: "$a`$b$",
+    note: "Both inline math forms close at a backtick followed by `$`, so neither holds a label with one.",
+  },
+  {
+    id: "math-label-line-break",
+    doc: { type: "doc", content: [mathParagraph("a\nb")] },
+    markdown: "$`a\nb`$",
+    note: "Inline math sits on one line, so a label with a line break splits its paragraph.",
+  },
+];