{ "openapi": "3.1.0", "info": { "title": "LexSelect External API", "description": "External API for the LexSelect platform. All endpoints are mounted under `/api`.\n\n**[Examples & Quick Start (GitHub)](https://github.com/LexSelect/lexselect-api-demo)** — Go CLI, Python notebook, and TypeScript examples.\n\n## Authentication\n\nAll documented endpoints require a Bearer API key in the `Authorization` header:\n\n```\nAuthorization: Bearer lxs_your_api_key\n```\n\nAPI keys require `ProcessingAPI` scope to access document and processing endpoints.\n\n## Client Environment\n\nThis is a server-to-server API **by design**. API keys must not be embedded\nin browser applications, and cross-origin browser requests are not supported.\nThe API does not emit CORS preflight response headers, and a preflight\n`OPTIONS` request is answered with `401` — this is intentional, not a bug.\nBrowser-based integrations are out of scope for this release.\n\n## Versioning\n\nPass the `X-API-Version` header to pin to a specific API version. Omitting the header\ndefaults to the latest version (`2026-09-12`). Unsupported versions return `400`.\n\nVersion `2026-09-12` and later treat a transcript's printed **line numbers as\nfurniture**: the gutter numerals leave `text`, `blocks`, `paragraphs`, the\nrenders and the default-grain `tree` (`value.text`), and are stated once, on\neach `transcript_line` node's `value.transcript_line_number`; `tokens=true`\nis unchanged. `/parse` and each of its pages carry `transcript.line_numbered`.\nTranscript nodes carry `meta.transcript` (`line_number_source`,\n`paragraph_break`, `turn`). Class 13 is `speaker_turn` — a named speaker's or\nexaminer's turn, with `value.speaker` / `value.turn_number` / `value.role` —\nand a turn continued from the previous page carries\n`meta.continuation.continues_prev`. Earlier pinned versions are byte-identical\nto before.\n\nVersion `2026-09-03` and later report each page's OCR-detected **language**\n(`{code, confidence}` or `null`) and reading **direction** (`rtl` / `ltr`) on\n`/parse` pages and `/processing/latest/pages` rows, the document-level majority\n`language` + `direction` on `/parse` and `/processing/latest`, and a majority\n`language` per sub-document. HTML renders carry `lang` / `dir` attributes.\n\nVersion `2026-06-07` and later report processing progress as **facts**:\nper-scope `progress_details` plus `pages_done` / `pages_total` / `total_known` /\n`stage` on `GET /documents/{id}/processing/latest` — and drops the legacy\n`processing_progress` fraction (everywhere) and the legacy progress\n`page_count` (on `/processing/latest`). The `/parse` response's `page_count` /\n`total_pages` are response-batch metadata, present on all versions, and are\nunaffected. Integrators that still need the legacy shape pin\n`X-API-Version: 2026-03-06`, the only version that carries it.\n\n## Rate Limiting\n\nAuthenticated requests are subject to per-key, per-user aggregate, and optional\nper-key concurrency limits. Authenticated responses include the active\n`X-RateLimit-*` budget headers. When a per-minute rate limit is exceeded,\na `429` is returned with a `Retry-After` header indicating seconds to wait.\nWhen an optional concurrency cap is exceeded, a `429` is returned with\ndetail text such as `Too many concurrent requests. Maximum: 10.`\n\n| Header | Description |\n|--------|-------------|\n| `X-RateLimit-Limit` | Requests allowed per minute for this key tier |\n| `X-RateLimit-Remaining` | Requests remaining in the current window |\n| `X-RateLimit-Reset` | Unix timestamp (seconds) when the window resets |\n| `Retry-After` | Seconds to wait before retrying. Present on per-minute rate-limit `429` responses; omitted for concurrency-limit `429` responses. |\n\n## Usage and Spend\n\n`GET /usage` returns a current-period snapshot — pages processed,\ndocuments stored, the remaining page balance, and spend to date with the\nper-page unit price — so cost and volume can be tracked and alerted on\nprogrammatically rather than read off the developer portal.\n`GET /usage/daily` returns the same counters as a zero-filled per-day\nseries, for forecasting and for alerting when a batch parses more pages\nthan expected.\n\nWhat an account gets depends on how it is billed:\n\n- **Personal accounts** are metered by pages and carry no price. `spend`\n is `null`, `spend_note` says so, and `period` reports the account's own\n pages and remaining allowance. The daily series covers the key owner's\n own processing runs.\n- **Enterprise workspaces** report the workspace's pooled pages against\n the plan's included pages, plus overage. `spend` — subscription,\n overage, unit price, estimated total — is returned **only to the\n workspace billing owner**; every other member gets `spend: null` with\n a `spend_note` saying so, and their daily series is scoped to their\n own runs. Workspaces on `custom` commercial terms report usage\n normally, but their money lives on the invoice, so `spend` stays\n `null`.\n\nBoth endpoints are read-only, require the `ProcessingAPI` scope, and are\nunaffected by `X-API-Version`.\n\n## Idempotency\n\nMutating requests (POST, PUT, PATCH, DELETE) may include an `Idempotency-Key` header.\nDuplicate requests with the same key return the cached response for 24 hours.\n\n## Errors\n\nAll error responses use [RFC 9457 Problem Details](https://www.rfc-editor.org/rfc/rfc9457)\nwith `Content-Type: application/problem+json`.\n\n```json\n{\n \"type\": \"about:blank\",\n \"title\": \"Unauthorized\",\n \"status\": 401,\n \"detail\": \"Missing or invalid Authorization header.\"\n}\n```\n\nEvery response includes an `X-Request-Id` header (UUID) for tracing.\n\n## Upload Flow\n\nUpload a file with a single request to `POST /documents/upload`. Send a\n`multipart/form-data` body containing the file `name`, its `size` in bytes,\nand the `file` part. In one HTTP round-trip the server stores the bytes,\nverifies their integrity, and queues the document for processing; the\nresponse is the created document object.\n\n**Requirements:**\n- Maximum file size: 250 MB.\n- Allowed extensions: `doc`, `docx`, `eml`, `html`, `msg`, `odt`, `pdf`, `rtf`.\n- The request `Content-Type` must be `multipart/form-data`.\n- Metadata fields (`name`, `size`, `content_type`, `parent_id`,\n `strip_file_name`) must precede the `file` part.\n- If `parent_id` is omitted, the file is placed in the user's Default Project.\n\n**File names.** A folder cannot hold two items with the same name. By\ndefault (`strip_file_name=true`) the stored document name is a generated\nUUIDv4 with the original extension — e.g. `report.pdf` is stored as\n`3f2a9c1e-…-b7d4.pdf` — so repeated uploads of the same file name never\ncollide, and the name you sent is used only to validate the extension and\ninfer `content_type`. The returned document object carries the stored\nname. Send `strip_file_name=false` to keep your file name; a name already\npresent in the target folder is then stored with a ` (n)` suffix.\n\nSee `POST /documents/upload` below for the full field reference.\n\n## Webhooks\n\nConfigure webhook endpoints via the UI to receive real-time notifications\nwhen document processing state changes. Payloads are signed with HMAC-SHA256.\n\nSubscribable event types:\n\n| Event | Description |\n|-------|-------------|\n| `document.processing.queued` | Document added to processing queue |\n| `document.processing.started` | Processing has begun |\n| `document.processing.page_completed` | A page finished processing |\n| `document.processing.completed` | Processing finished successfully |\n| `document.processing.failed` | Processing failed |\n| `*` | All events (includes future event types) |\n", "version": "2026-06-07", "contact": { "name": "LexSelect API Support" } }, "servers": [ { "url": "/api", "description": "API server (relative path, environment-agnostic)" } ], "security": [ { "ApiKeyAuth": [] } ], "tags": [ { "name": "Files", "description": "File storage — upload, list, retrieve, and delete files and folders." }, { "name": "Processing", "description": "Processing pipeline — track status and progress of document processing." }, { "name": "Results", "description": "Processing results — page metadata, extracted data, and evaluations." }, { "name": "Usage", "description": "Account usage and spend — pages processed, documents stored, remaining balance, and cost to date." } ], "paths": { "/folders/{folder_id}/sub-documents": { "get": { "operationId": "listFolderSubDocuments", "summary": "List sub-documents across a folder", "description": "**API version `2026-08-05` and later.** Lists every detected sub-document across the folder's direct file children as a flat, paginated browse list — the subdocument-first navigation surface: filter by `type` (\"every invoice in this folder\") and open the parent document at the returned anchor without listing or opening files. Rows come from the engine's sub-document registry (rebuilt on every reprocess); `kind` is `subdocument` today, with `section` reserved for a future stratum.\n", "tags": [ "Files" ], "parameters": [ { "$ref": "#/components/parameters/XApiVersion" }, { "name": "folder_id", "in": "path", "required": true, "description": "Folder or project ObjectId (direct file children are scanned).", "schema": { "type": "string" } }, { "name": "type", "in": "query", "required": false, "description": "Case-insensitive exact match on the open-vocabulary `document_type` (e.g. `Notice`, `Invoice`).", "schema": { "type": "string" } }, { "name": "kind", "in": "query", "required": false, "description": "Row class filter.", "schema": { "type": "string", "enum": [ "subdocument", "section" ] } }, { "name": "limit", "in": "query", "required": false, "description": "Rows per page, 1-200 (default 50).", "schema": { "type": "integer", "default": 50 } }, { "name": "offset", "in": "query", "required": false, "description": "Rows to skip (default 0).", "schema": { "type": "integer", "default": 0 } } ], "responses": { "200": { "description": "Paginated sub-document rows in per-document reading order.", "content": { "application/json": { "schema": { "type": "object", "required": [ "folder_id", "total", "limit", "offset", "sub_documents" ], "properties": { "folder_id": { "type": "string" }, "total": { "type": "integer", "description": "Total matching rows across the folder (pre-pagination)." }, "limit": { "type": "integer" }, "offset": { "type": "integer" }, "sub_documents": { "type": "array", "items": { "$ref": "#/components/schemas/FolderSubDocumentRow" } } } } } } }, "400": { "description": "Invalid id, filter, or pagination parameter." }, "404": { "description": "Folder not found (or not owned by the caller)." }, "422": { "description": "Folder exceeds the 5000-file listing cap." }, "503": { "description": "Sub-document registry temporarily unavailable." } } } }, "/documents": { "get": { "operationId": "listDocuments", "summary": "List files", "description": "Lists documents and folders within a parent. Without `parent_id`, returns root-level items (projects). Use the project ID as `parent_id` to list files within it.\n", "tags": [ "Files" ], "parameters": [ { "$ref": "#/components/parameters/XApiVersion" }, { "name": "parent_id", "in": "query", "required": false, "description": "Parent folder or project ObjectId. Omit for root-level listing (projects).", "schema": { "type": "string" } }, { "name": "limit", "in": "query", "required": false, "description": "Maximum items per page — an integer from 1 to 100 (default 25). Out-of-range or non-integer values are rejected with `400` (not clamped).", "schema": { "type": "integer", "default": 25, "minimum": 1, "maximum": 100 } }, { "name": "cursor", "in": "query", "required": false, "description": "Cursor from a previous response's `next_cursor` field.", "schema": { "type": "string" } }, { "name": "sort_by", "in": "query", "required": false, "description": "Field to sort by.", "schema": { "type": "string", "enum": [ "name", "size", "created_at", "modified_at" ], "default": "name" } }, { "name": "sort_direction", "in": "query", "required": false, "schema": { "type": "string", "enum": [ "asc", "desc" ], "default": "asc" } }, { "name": "type", "in": "query", "required": false, "description": "Filter by item type.", "schema": { "type": "string", "enum": [ "file", "folder", "project" ] } }, { "name": "include_deleted", "in": "query", "required": false, "description": "Include soft-deleted items.", "schema": { "type": "boolean", "default": false } } ], "responses": { "200": { "description": "List of documents.", "headers": { "X-Request-Id": { "$ref": "#/components/headers/X-Request-Id" } }, "content": { "application/json": { "schema": { "$ref": "#/components/schemas/DocumentListResponse" } } } }, "400": { "description": "Invalid query parameters.", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "401": { "$ref": "#/components/responses/Unauthorized" }, "403": { "$ref": "#/components/responses/Forbidden" }, "404": { "description": "Parent folder not found.", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "429": { "$ref": "#/components/responses/TooManyRequests" }, "500": { "$ref": "#/components/responses/InternalServerError" }, "504": { "$ref": "#/components/responses/GatewayTimeout" } } } }, "/documents/upload": { "post": { "operationId": "uploadDocument", "summary": "Upload file", "description": "Upload a file in a single `multipart/form-data` request. The server handles storage upload and the processing trigger internally, and returns the created document object.\n\nMaximum file size: 250 MB. Allowed extensions: doc, docx, eml, html, msg, odt, pdf, rtf.\n\nIf `parent_id` is omitted the file is placed in the user's Default Project.\n\nBy default (`strip_file_name=true`) the document is stored under a generated UUIDv4 name that keeps the original extension; `name` is then used only for extension validation and content-type inference. Send `strip_file_name=false` to keep the original file name.\n", "tags": [ "Files" ], "parameters": [ { "$ref": "#/components/parameters/XApiVersion" }, { "$ref": "#/components/parameters/IdempotencyKey" } ], "requestBody": { "required": true, "content": { "multipart/form-data": { "schema": { "type": "object", "required": [ "name", "size", "file" ], "properties": { "name": { "type": "string", "description": "File name including extension. Stored verbatim only when `strip_file_name=false`; by default it is replaced with a generated UUIDv4 (extension preserved) and used only to validate the extension and infer `content_type`.\n", "example": "report.pdf" }, "size": { "type": "integer", "example": 0, "description": "File size in bytes. Must equal the file's exact byte count, or the upload is rejected. Maximum 250 MB.\n" }, "file": { "type": "string", "format": "binary", "description": "The file to upload (max 250 MB)." }, "content_type": { "type": "string", "description": "MIME type override. Inferred from file name if omitted." }, "parent_id": { "type": "string", "description": "Parent folder or project ObjectId. Defaults to the user's Default Project." }, "strip_file_name": { "type": "boolean", "default": true, "description": "When `true` (the default) the stored document name is a generated UUIDv4 plus the original extension (`.pdf`), so re-uploading the same file name never collides with an existing item. When `false` the `name` field is stored as sent; a duplicate name in the target folder gets a ` (n)` suffix. Accepts `true`/`false`, `1`/`0`, `yes`/`no`; any other value is rejected with `400`.\n" } } } } } }, "responses": { "200": { "description": "File uploaded and processing queued.", "headers": { "X-Request-Id": { "$ref": "#/components/headers/X-Request-Id" } }, "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Document" } } } }, "400": { "description": "Validation error (missing name/file or invalid parent), a non-`multipart/form-data` Content-Type, or a malformed `multipart/form-data` body (missing or invalid boundary).\n", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "401": { "$ref": "#/components/responses/Unauthorized" }, "403": { "$ref": "#/components/responses/Forbidden" }, "413": { "$ref": "#/components/responses/ContentTooLarge" }, "415": { "$ref": "#/components/responses/UnsupportedMediaType" }, "429": { "$ref": "#/components/responses/TooManyRequests" }, "502": { "description": "Storage upload failed.", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "504": { "$ref": "#/components/responses/GatewayTimeout" } } } }, "/documents/{id}": { "parameters": [ { "name": "id", "in": "path", "required": true, "description": "Document ObjectId.", "schema": { "type": "string" } } ], "get": { "operationId": "getDocument", "summary": "Get file", "description": "Retrieve a single document by ID.", "tags": [ "Files" ], "parameters": [ { "$ref": "#/components/parameters/XApiVersion" } ], "responses": { "200": { "description": "Document details.", "headers": { "X-Request-Id": { "$ref": "#/components/headers/X-Request-Id" } }, "content": { "application/json": { "schema": { "$ref": "#/components/schemas/Document" } } } }, "401": { "$ref": "#/components/responses/Unauthorized" }, "403": { "$ref": "#/components/responses/Forbidden" }, "404": { "description": "Document not found or not owned by caller.", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "429": { "$ref": "#/components/responses/TooManyRequests" }, "500": { "$ref": "#/components/responses/InternalServerError" }, "504": { "$ref": "#/components/responses/GatewayTimeout" } } }, "delete": { "operationId": "deleteDocument", "summary": "Delete file", "description": "Permanently deletes a document or folder and everything derived from it. Recursive for folders. This cannot be undone: the file, its previews, page images, extracted data and search records are removed from every store, and the same job id cannot be re-uploaded to restore it. The response returns as soon as the deletion is accepted and queued; removal across stores completes shortly afterwards. Idempotent — deleting an already-deleted item returns success.\n", "tags": [ "Files" ], "parameters": [ { "$ref": "#/components/parameters/XApiVersion" }, { "$ref": "#/components/parameters/IdempotencyKey" } ], "responses": { "200": { "description": "Document deleted.", "headers": { "X-Request-Id": { "$ref": "#/components/headers/X-Request-Id" } }, "content": { "application/json": { "schema": { "$ref": "#/components/schemas/DocumentDeleteResponse" } } } }, "401": { "$ref": "#/components/responses/Unauthorized" }, "403": { "$ref": "#/components/responses/Forbidden" }, "404": { "description": "Document not found or not owned by caller.", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "429": { "$ref": "#/components/responses/TooManyRequests" }, "500": { "$ref": "#/components/responses/InternalServerError" }, "504": { "$ref": "#/components/responses/GatewayTimeout" } } } }, "/documents/{id}/parse": { "parameters": [ { "name": "id", "in": "path", "required": true, "description": "Document ObjectId.", "schema": { "type": "string" } } ], "get": { "operationId": "getDocumentParse", "summary": "Get parsed document structure", "description": "Returns one or more projections of the document's latest processing output for every page. The default response includes the full hierarchical `tree` and the flat `kvps` map per page. Use the `include` query parameter to request alternative projections (plain `text`, extracted `tables`, or a flat list of `blocks`) instead of — or in addition to — the raw tree.\n\n**Slices:**\n\n- `tree` — hierarchical layout tree produced by the engine's\n tree-builder stage. From API version `2026-08-01` (the default)\n nodes are the curated public shape (`class` names, whitelisted\n `value`, closed `meta` object — see `ParsedTreeNode` /\n `ParsedNodeMeta`); word-level TOKEN nodes are included only with\n `tokens=true`. Pinned versions return the legacy verbatim engine\n shape.\n- `kvps` — flat `{key: string, value: string}` map of extracted\n key-value pairs. Only populated when a partner-specific extraction\n workflow has run; otherwise `null` on every page.\n- `text` — plain text in reading order, one line per LINE node,\n joined with `\\n`.\n- `tables` — every TABLE subtree projected to\n `{headers: [...], rows: [[...]]}`.\n- `blocks` — flat pre-order list of content blocks (HEADER, FOOTER,\n SECTION, LINE, TABLE, SUB_DOCUMENT), each with `{id, class, text}`.\n- `paragraphs` — reading-order logical paragraphs with section and\n heading attribution; continuation-linked runs merged within each\n page, with seam flags for cross-page stitching.\n- `clauses` — whole-document nested outline (headings + numbered\n clauses), stitched across pages; returned once at the response\n top level, not per page.\n\n\n**Filtering:**\n- `pages` limits which pages are returned (1-indexed). - `exclude_node_types` strips subtrees whose class matches before\n any projection runs, applying across `tree`, `text`, `tables`,\n and `blocks` — with one exception from `2026-08-01`: `token` in\n the list is ignored for the `tree` slice (the `tokens` parameter\n is the sole TOKEN control there; derived slices still honor it).\n- `strip_geometry=true` removes per-node `coordinates` and per-page\n `width`/`height` from the `tree` slice (no effect on other\n slices). On pinned versions it also removes `engine_meta` and\n `tokens` — the legacy all-or-nothing behavior.\n- `tokens=true` (from `2026-08-01`) includes word-level TOKEN nodes\n in the `tree` slice — per-word geometry and `meta.styling`.\n\n\nFor in-progress processings, only pages with `status=SUCCESS` are included.\n", "tags": [ "Results" ], "parameters": [ { "$ref": "#/components/parameters/XApiVersion" }, { "name": "pages", "in": "query", "required": false, "description": "Page subset, 1-indexed, inclusive ranges. Examples: `1-10`, `1,3,5`, `1,3-5,10`. Omitted or empty returns **every page** of the document.\n\nOut-of-range values are silently clamped to the document's actual page count. A request like `pages=1-1000` against a 6-page document returns all 6 pages, not an empty list. Use the response's `total_pages` field to see the document's real size.\n\nRanges that would expand to more than **100,000 distinct pages** are rejected with `400`. Cumulative lists are checked against the same cap.\n\nInvalid syntax (non-numeric, zero, negative, reversed range) → `400`.\n", "schema": { "type": "string", "example": "1-10" } }, { "name": "include", "in": "query", "required": false, "description": "Comma-separated list of slices to include per page. Valid values: `tree`, `kvps`, `text`, `tables`, `blocks`, `paragraphs`, `clauses`. Unknown values → 400. Defaults to `tree,kvps`.\n", "schema": { "type": "string", "example": "text,tables" } }, { "name": "exclude_node_types", "in": "query", "required": false, "description": "Comma-separated list of `NodeClass` names (lowercase) whose subtrees should be stripped before projection. Common use: `header,footer` to drop boilerplate. Unknown names → 400. See `ParsedTreeNode.class_id` for the full vocabulary. From API version `2026-08-01`, `token` in this list is ignored for the `tree` slice — TOKEN inclusion is governed solely by the `tokens` parameter (tokens are the text carriers; stripping them here would produce a text-less tree). Derived slices (`text`, `tables`, `blocks`, `paragraphs`) still honor the full list.\n", "schema": { "type": "string", "example": "header,footer" } }, { "name": "strip_geometry", "in": "query", "required": false, "description": "When `true`, removes spatial metadata from the response: per-node `coordinates` in the `tree` slice plus per-page `width` and `height`. From `2026-08-01` it affects ONLY geometry — the curated `meta` is always kept. On pinned versions (`2026-06-07` and earlier) it additionally removes `engine_meta` and `tokens` (the legacy all-or-nothing behavior). Has no effect on `text`, `tables`, `blocks`, or `kvps` slices. Default `false`.\n", "schema": { "type": "boolean", "default": false } }, { "name": "tokens", "in": "query", "required": false, "description": "From `2026-08-01`: when `true`, word-level TOKEN nodes (one per OCR word, with geometry and `meta.styling`) are included in the `tree` slice. Default `false` — line text is carried on each LINE's `value.text` instead, and responses are typically ~5x smaller. No effect on pinned versions (tokens always present there).\n", "schema": { "type": "boolean", "default": false } } ], "responses": { "200": { "description": "Parsed document structure.", "headers": { "X-Request-Id": { "$ref": "#/components/headers/X-Request-Id" } }, "content": { "application/json": { "schema": { "$ref": "#/components/schemas/ParseResponse" } } } }, "400": { "description": "Invalid document id format.", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "401": { "$ref": "#/components/responses/Unauthorized" }, "403": { "$ref": "#/components/responses/Forbidden" }, "404": { "description": "Document not found, not owned by caller, or has no processing records.", "content": { "application/problem+json": { "schema": { "$ref": "#/components/schemas/ProblemDetails" } } } }, "429": { "$ref": "#/components/responses/TooManyRequests" }, "500": { "$ref": "#/components/responses/InternalServerError" }, "504": { "$ref": "#/components/responses/GatewayTimeout" } } } }, "/documents/{id}/render": { "parameters": [ { "name": "id", "in": "path", "required": true, "description": "Document ObjectId.", "schema": { "type": "string" } } ], "get": { "operationId": "getDocumentRender", "summary": "Render document to Markdown or HTML (experimental)", "description": "**EXPERIMENTAL.** Renders the document's parsed tree to a single Markdown or HTML string and returns it as the raw response body (not JSON). Use this to quickly preview what the parser produced, paste into a Markdown viewer, or open in a browser.\n\nRendering rules (best effort, driven entirely by the parsed tree — no LLM):\n\n- `TITLE`/`HEADING` → `# `/`## ` (Markdown) or `

`/`

` (HTML). - `PARAGRAPH`/`ABSTRACT` → one block (child lines grouped, not one\n per line); `LIST_ITEM` runs → a single `