@stratta/mcp 1.18.0 → 1.18.1

package/README.md

@@ -9,8 +9,9 @@ mcp-name: ch.stratta/mcp
 
 # @stratta/mcp
 
-The local MCP server for [Stratta](https://stratta.ch): the engineering norms
-your bureau licensed (SIA, Eurocodes), cited to the section and page, inside
+The local MCP server for [Stratta](https://stratta.ch): the norms and reference
+documents your organization deposited (technical standards, laws, regulations,
+accounting standards, internal references), cited to the section and page, inside
 any agent that speaks MCP.
 
 > Requires a free Stratta account: https://stratta.ch/sign-up.
@@ -233,7 +234,7 @@ package announces 59 of them; the remote connector serves 61: everything minus
 | `open_dossier`         | Open a project's dossier, creating it if needed. Idempotent on the name.                                  |
 | `open_question`        | Open one question to settle, with optional named options. Idempotent on the title.                        |
 | `save_finding`         | Record one piece of evidence: a cited article, a retained value and why, an observation.                  |
-| `record_decision`      | Settle a question with a decision the engineer has confirmed, and the retained option.                    |
+| `record_decision`      | Settle a question with a decision the user has confirmed, and the retained option.                        |
 | `load_dossier`         | Reload everything: questions with their evidence and decisions, open ones first.                          |
 | `resolve_question`     | Close a question without a decision, or reopen one. The evidence stays.                                   |
 | `list_attachments`     | The project attachments of a dossier: site reports, borehole logs, minutes, data sheets.                  |
@@ -277,7 +278,7 @@ of the organisation. A report template learned from an office Word file on
 [stratta.ch/bibliotheque](https://stratta.ch/bibliotheque) is served as a skill
 too, and its file is the layout the Word export is rendered into.
 
-**Ingest** (10 tools — add YOUR licensed norms; driven by the bundled `ingest-norm` skill; owner or admin role):
+**Ingest** (10 tools — add YOUR documents; driven by the bundled `ingest-norm` skill; owner or admin role):
 
 | Tool                          | Purpose                                                                          |
 | ----------------------------- | -------------------------------------------------------------------------------- |
@@ -336,7 +337,7 @@ Three prompts appear as slash commands in clients that support them:
 `investigate-question` (question, project?), `resume-dossier` (name) and
 `check-value` (value, norm?).
 
-For ingesting your own licensed norms, see the bundled **`ingest-norm` skill** —
+For depositing your own documents, see the bundled **`ingest-norm` skill** —
 a 2-phase hybrid pipeline (since 0.4.0): a Python pre-pass
 (`scripts/ingest-prepass.py`, PyMuPDF) extracts the hierarchical tree and
 rasterizes figures deterministically, then the agent enriches sections with

package/dist/tools/catalog.gen.d.ts

@@ -13,7 +13,7 @@ export type CatalogEntry = {
     transports: Transport[];
     params: Record<string, string>;
 };
-export declare const SERVER_INSTRUCTIONS = "Stratta serves the construction norms of the signed-in organization, and only those. Call `get_methodology` before the first technical question of a conversation: it carries the navigation order, the citation rules and the dossier policy. Never answer a technical question from memory; a norm absent from `list_norms` cannot be read.";
+export declare const SERVER_INSTRUCTIONS = "Stratta serves the norms and reference documents of the signed-in organization (technical standards, laws, ordinances, regulations, accounting standards, internal references), and only those. Call `get_methodology` before the first substantive question of a conversation: it carries the navigation order, the citation rules and the dossier policy. Never answer from memory a question these documents are meant to settle; a norm absent from `list_norms` cannot be read.";
 export declare const CATALOG: CatalogEntry[];
 export type PromptArg = {
     name: string;

package/dist/tools/catalog.gen.js

@@ -1,6 +1,6 @@
 // Generated by scripts/build-mcp-catalog.mjs from convex/mcpCatalog.ts.
 // Do not edit: change the source and run the script.
-export const SERVER_INSTRUCTIONS = "Stratta serves the construction norms of the signed-in organization, and only those. Call `get_methodology` before the first technical question of a conversation: it carries the navigation order, the citation rules and the dossier policy. Never answer a technical question from memory; a norm absent from `list_norms` cannot be read.";
+export const SERVER_INSTRUCTIONS = "Stratta serves the norms and reference documents of the signed-in organization (technical standards, laws, ordinances, regulations, accounting standards, internal references), and only those. Call `get_methodology` before the first substantive question of a conversation: it carries the navigation order, the citation rules and the dossier policy. Never answer from memory a question these documents are meant to settle; a norm absent from `list_norms` cannot be read.";
 export const CATALOG = [
     {
         "name": "get_methodology",
@@ -37,7 +37,7 @@ export const CATALOG = [
     {
         "name": "list_norms",
         "title": "Available norms",
-        "description": "List the norms published in the signed-in organization, with code, year, title, language and coverage (how much of the norm was ingested). This is the first call of any search: a norm absent from this list cannot be read and must not be answered from memory. A demo entry is fictional and never a technical reference; a sparse entry carries a notice to relay.",
+        "description": "List the norms and reference documents (standards, laws, regulations) published in the signed-in organization, with code, year, title, language and coverage (how much of the norm was ingested). This is the first call of any search: a norm absent from this list cannot be read and must not be answered from memory. A demo entry is fictional and never a technical reference; a sparse entry carries a notice to relay.",
         "annotations": {
             "readOnlyHint": true,
             "destructiveHint": false,
@@ -65,7 +65,7 @@ export const CATALOG = [
             "remote"
         ],
         "params": {
-            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"SIA 267\", \"EN 1997-1\".",
+            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"EN 1997-1\", \"CO\", \"IFRS 15\".",
             "edition": "Edition of the norm when the organization holds several (\"2013\", \"2020\"). Default: the most recent published edition; list_norms names them.",
             "maxDepth": "Nesting depth to include. 1 (default) returns chapters only; 2 adds X.Y sections. Large norms grow fast beyond 2."
         }
@@ -85,7 +85,7 @@ export const CATALOG = [
             "remote"
         ],
         "params": {
-            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"SIA 267\", \"EN 1997-1\".",
+            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"EN 1997-1\", \"CO\", \"IFRS 15\".",
             "edition": "Edition of the norm when the organization holds several (\"2013\", \"2020\"). Default: the most recent published edition; list_norms names them.",
             "path": "Path to root the subtree at, e.g. \"9\" or \"9.5\". A clause number below the last heading resolves to that heading.",
             "maxDepth": "Depth relative to the root. 1 returns the root and its direct children; omit for the whole subtree.",
@@ -107,7 +107,7 @@ export const CATALOG = [
             "remote"
         ],
         "params": {
-            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"SIA 267\", \"EN 1997-1\".",
+            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"EN 1997-1\", \"CO\", \"IFRS 15\".",
             "edition": "Edition of the norm when the organization holds several (\"2013\", \"2020\"). Default: the most recent published edition; list_norms names them.",
             "path": "Section path as it appears in the table of contents, e.g. \"4.2.1\" or \"Annexe A\".",
             "maxChars": "Characters of text per call, 500 to 40000. Default 12000, which fits any client. Raise it only when one section is needed whole.",
@@ -117,7 +117,7 @@ export const CATALOG = [
     {
         "name": "search_in_norm",
         "title": "Search a norm",
-        "description": "Full-text search inside one norm, ranked, a hit in a heading outranking a hit in the body. Use it when the table of contents suggests no obvious section. Matching is per term, not per substring, in the language of the norm: search \"charges\" in a French norm, not \"loads\"; accents and case do not matter. A clause number as keyword returns that clause. Returns paths, titles and snippets; read the promising ones with get_section.",
+        "description": "Full-text search inside one norm, ranked, a hit in a heading outranking a hit in the body. Use it when the table of contents suggests no obvious section. Matching is per term, not per substring, in the language of the norm: search \"charges\" in a French document, not \"loads\"; accents and case do not matter. A clause number as keyword returns that clause. Returns paths, titles and snippets; read the promising ones with get_section.",
         "annotations": {
             "readOnlyHint": true,
             "destructiveHint": false,
@@ -129,7 +129,7 @@ export const CATALOG = [
             "remote"
         ],
         "params": {
-            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"SIA 267\", \"EN 1997-1\".",
+            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"EN 1997-1\", \"CO\", \"IFRS 15\".",
             "edition": "Edition of the norm when the organization holds several (\"2013\", \"2020\"). Default: the most recent published edition; list_norms names them.",
             "keyword": "Search terms in the language of the norm, accents optional. Several words find sections containing all of them; a clause number finds that clause.",
             "limit": "Maximum number of hits, 1 to 50. Default 20."
@@ -157,7 +157,7 @@ export const CATALOG = [
     {
         "name": "get_cross_refs",
         "title": "Cross-references",
-        "description": "Return the references one section makes: to other clauses, annexes, figures and tables of the same norm, and to other norms. SIA norms deliberately spread the rule, the coefficient and the action across separate clauses and norms, so following a reference is often what separates a correct answer from an incomplete one. Each reference says whether it is resolvable in this corpus (listed first) and names the target section; read each relevant target with get_section.",
+        "description": "Return the references one section makes: to other clauses, annexes, figures and tables of the same norm, and to other norms. Norms and regulations often spread the rule, the coefficient and the exception across separate clauses and documents, so following a reference is often what separates a correct answer from an incomplete one. Each reference says whether it is resolvable in this corpus (listed first) and names the target section; read each relevant target with get_section.",
         "annotations": {
             "readOnlyHint": true,
             "destructiveHint": false,
@@ -169,7 +169,7 @@ export const CATALOG = [
             "remote"
         ],
         "params": {
-            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"SIA 267\", \"EN 1997-1\".",
+            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"EN 1997-1\", \"CO\", \"IFRS 15\".",
             "edition": "Edition of the norm when the organization holds several (\"2013\", \"2020\"). Default: the most recent published edition; list_norms names them.",
             "path": "Section path as it appears in the table of contents, e.g. \"4.2.1\" or \"Annexe A\"."
         }
@@ -177,7 +177,7 @@ export const CATALOG = [
     {
         "name": "get_figure",
         "title": "Figure",
-        "description": "Return one figure of a norm as an image: load diagrams, design charts, standard cross-sections. Call it when the figure carries the rule and describing it in words would lose the content. The figure identifier comes from the figures array of get_section.",
+        "description": "Return one figure of a norm as an image: diagrams, charts, standard cross-sections. Call it when the figure carries the rule and describing it in words would lose the content. The figure identifier comes from the figures array of get_section.",
         "annotations": {
             "readOnlyHint": true,
             "destructiveHint": false,
@@ -189,7 +189,7 @@ export const CATALOG = [
             "remote"
         ],
         "params": {
-            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"SIA 267\", \"EN 1997-1\".",
+            "code": "Norm code exactly as list_norms returns it, e.g. \"SIA 261\", \"EN 1997-1\", \"CO\", \"IFRS 15\".",
             "edition": "Edition of the norm when the organization holds several (\"2013\", \"2020\"). Default: the most recent published edition; list_norms names them.",
             "figureId": "Figure identifier from the figures array of get_section."
         }
@@ -488,7 +488,7 @@ export const CATALOG = [
     {
         "name": "save_finding",
         "title": "Record evidence",
-        "description": "Record one piece of evidence in a dossier, as you work rather than in a batch at the end. Kinds: reference (what a norm says; always give normCode, sectionPath and page, or it is not a reference), hypothesis (the value the engineer retains and why, value in value and reasoning in detail), observation (what the site, a borehole or a survey showed, with the date). One clause, one value or one fact per call, in one or two sentences. Pass questionId to file it under the question it serves; without it the entry waits unclassified. Set confidence whenever a value is at stake: established, judgement or to_confirm. Questions go through open_question and decisions through record_decision, not here.",
+        "description": "Record one piece of evidence in a dossier, as you work rather than in a batch at the end. Kinds: reference (what a norm says; always give normCode, sectionPath and page, or it is not a reference), hypothesis (the value the user retains and why, value in value and reasoning in detail), observation (what the site, a borehole or a survey showed, with the date). One clause, one value or one fact per call, in one or two sentences. Pass questionId to file it under the question it serves; without it the entry waits unclassified. Set confidence whenever a value is at stake: established, judgement or to_confirm. Questions go through open_question and decisions through record_decision, not here.",
         "annotations": {
             "readOnlyHint": false,
             "destructiveHint": false,
@@ -503,10 +503,10 @@ export const CATALOG = [
             "dossierId": "Dossier id returned by open_dossier or list_dossiers.",
             "questionId": "The question this evidence serves, from open_question or load_dossier. Omit only when you do not know yet.",
             "kind": "reference, hypothesis or observation: pick by what the entry is.",
-            "title": "The fact in one line, as an engineer would write it in a report. One clause, one value or one observation; not a theme.",
+            "title": "The fact in one line, as a professional would write it in a report. One clause, one value or one observation; not a theme.",
             "detail": "The reasoning in one or two sentences: why this value, what it rests on, what disagreed. Expected for a hypothesis.",
             "value": "The retained value with its unit, e.g. \"30 deg\", \"1.25 kN/m2\".",
-            "confidence": "established (computed or read directly), judgement (the engineer chose it), to_confirm (provisional, pending a test).",
+            "confidence": "established (computed or read directly), judgement (the user chose it), to_confirm (provisional, pending a test).",
             "normCode": "Norm this comes from, e.g. \"SIA 267\".",
             "sectionPath": "Exact section path, e.g. \"9.5.2.1\".",
             "page": "Page in the norm, so a human can verify it in their PDF.",
@@ -518,7 +518,7 @@ export const CATALOG = [
     {
         "name": "record_decision",
         "title": "Record a decision",
-        "description": "Record what the engineer retained: the decision in one line, why, and the clause it rests on. Pass questionId when the question already exists; otherwise pass dossierId and question (what is settled, in one line) and the question is opened on the spot, already decided. Name the retained option when the question listed some; put the number in value and set confidence when the decision is a value. Only after the user confirmed the decision, never on your own reasoning. Returns entryId.",
+        "description": "Record what the user retained: the decision in one line, why, and the clause it rests on. Pass questionId when the question already exists; otherwise pass dossierId and question (what is settled, in one line) and the question is opened on the spot, already decided. Name the retained option when the question listed some; put the number in value and set confidence when the decision is a value. Only after the user confirmed the decision, never on your own reasoning. Returns entryId.",
         "annotations": {
             "readOnlyHint": false,
             "destructiveHint": false,
@@ -587,7 +587,7 @@ export const CATALOG = [
     {
         "name": "list_attachments",
         "title": "Project attachments",
-        "description": "List the project attachments of one dossier: the site reports, borehole logs, meeting minutes and data sheets the engineer uploaded. Returns for each its id, name, type, page count and extraction status. Read one with read_attachment before answering on the project; a piece is evidence like a clause and is cited the same way.",
+        "description": "List the project attachments of one dossier: the site reports, borehole logs, meeting minutes and data sheets the user uploaded. Returns for each its id, name, type, page count and extraction status. Read one with read_attachment before answering on the project; a piece is evidence like a clause and is cited the same way.",
         "annotations": {
             "readOnlyHint": true,
             "destructiveHint": false,
@@ -664,7 +664,7 @@ export const CATALOG = [
     {
         "name": "list_templates",
         "title": "Dossier templates",
-        "description": "List the checklists the organization wrote for its types of structure: each is a set of questions with the clauses they rest on. When a new project matches one, apply it with apply_template instead of opening the questions one by one.",
+        "description": "List the checklists the organization wrote for its types of case: each is a set of questions with the clauses they rest on. When a new project matches one, apply it with apply_template instead of opening the questions one by one.",
         "annotations": {
             "readOnlyHint": true,
             "destructiveHint": false,
@@ -989,7 +989,7 @@ export const CATALOG = [
         ],
         "params": {
             "entryId": "Entry id from get_question or load_dossier.",
-            "title": "The fact in one line, as an engineer would write it in a report.",
+            "title": "The fact in one line, as a professional would write it in a report.",
             "detail": "The reasoning in one or two sentences.",
             "value": "The retained value with its unit, e.g. \"30 deg\".",
             "confidence": "established, judgement or to_confirm.",
@@ -1040,7 +1040,7 @@ export const CATALOG = [
     {
         "name": "ingest_create_document",
         "title": "Create norm document",
-        "description": "Create a draft norm document in the workspace. Returns the documentId every later ingest_* call needs. The organization must hold a licence for the norm it ingests.",
+        "description": "Create a draft norm document in the workspace. Returns the documentId every later ingest_* call needs. The organization must hold the rights to the document it deposits.",
         "annotations": {
             "readOnlyHint": false,
             "destructiveHint": false,
@@ -1058,7 +1058,8 @@ export const CATALOG = [
             "language": "fr, de, it or en.",
             "totalPages": "Page count of the source PDF.",
             "edition": "Edition label as printed on the cover (\"2013\", \"2020+C1\"). Default: the year. Lets several editions of one code coexist.",
-            "scope": "One sentence on what the norm covers, shown in the served methodology, e.g. \"Geotechnical design: foundations, piles, anchors, retaining structures\"."
+            "scope": "One sentence on what the norm covers, shown in the served methodology, e.g. \"Geotechnical design: foundations, piles, anchors, retaining structures\".",
+            "licence": "Watermark and copyright lines of the licensed copy, verbatim, as the pre-pass wrote them in doc.licence. Shown with every excerpt of the norm."
         }
     },
     {
@@ -1226,12 +1227,12 @@ export const CATALOG = [
 export const PROMPTS = [
     {
         "name": "investigate-question",
-        "title": "Investigate a question in the norms",
-        "description": "Answer a technical question from the norms of this workspace, with a citation for every value, and file what is retained in the project dossier when there is one.",
+        "title": "Investigate a question in the documents",
+        "description": "Answer a question from the norms and reference documents of this workspace, with a citation for every value, and file what is retained in the project dossier when there is one.",
         "args": [
             {
                 "name": "question",
-                "description": "The engineering question, as the user would ask a colleague.",
+                "description": "The question, as the user would ask a colleague.",
                 "required": true
             },
             {

package/dist/tools/ingest.d.ts

@@ -15,6 +15,7 @@ export declare const ingestCreateDocument: import("./define.js").ToolDef<{
     totalPages: z.ZodNumber;
     scope: z.ZodOptional<z.ZodString>;
     edition: z.ZodOptional<z.ZodString>;
+    licence: z.ZodOptional<z.ZodString>;
 }>;
 export declare const ingestCreateSections: import("./define.js").ToolDef<{
     documentId: z.ZodString;
@@ -94,6 +95,7 @@ export declare const ingestTools: (import("./define.js").ToolDef<{
     totalPages: z.ZodNumber;
     scope: z.ZodOptional<z.ZodString>;
     edition: z.ZodOptional<z.ZodString>;
+    licence: z.ZodOptional<z.ZodString>;
 }> | import("./define.js").ToolDef<{
     documentId: z.ZodString;
     sections: z.ZodArray<z.ZodObject<{

package/dist/tools/ingest.js

@@ -91,6 +91,11 @@ export const ingestCreateDocument = defineTool({
             .max(20)
             .optional()
             .describe(param('ingest_create_document', 'edition')),
+        licence: z
+            .string()
+            .max(600)
+            .optional()
+            .describe(param('ingest_create_document', 'licence')),
     },
     run: async (client, args) => ({
         documentId: await client.action(api.ingestionApi.createDocument, {

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "@stratta/mcp",
   "mcpName": "ch.stratta/mcp",
-  "version": "1.18.0",
-  "description": "MCP server exposing the engineering norms your firm is licensed for (SIA / Eurocodes) to any MCP client, via Stratta TreeRAG.",
+  "version": "1.18.1",
+  "description": "MCP server exposing your organization’s norms and reference documents (standards, laws, regulations) to any MCP client, with exact citations, via Stratta TreeRAG.",
   "license": "UNLICENSED",
   "author": "SmartFlow <hello@stratta.ch>",
   "homepage": "https://stratta.ch",

package/scripts/ingest-prepass.py

@@ -28,7 +28,10 @@ from collections import defaultdict
 from pathlib import Path
 
 try:
-    import fitz  # PyMuPDF
+    try:
+        import pymupdf
+    except ImportError:  # PyMuPDF < 1.24.3 only knows the legacy name
+        import fitz as pymupdf
 except ImportError:
     sys.stderr.write(
         "ERROR: PyMuPDF not installed. Run: pip install --user pymupdf\n"
@@ -81,7 +84,7 @@ OCR_DOTTED_NUMBER_RE = re.compile(r"^(\d+(?:\.\d+)*) \.(\d)")
 VERTICAL_WATERMARK_SHARE = 0.25
 
 
-def _vertical_lines(page: "fitz.Page") -> list[tuple[tuple[float, float, float, float], set[str]]]:
+def _vertical_lines(page: "pymupdf.Page") -> list[tuple[tuple[float, float, float, float], set[str]]]:
     """Box and words of each line not written left to right.
 
     Matched to the words by position and text, never by (block, line)
@@ -117,7 +120,7 @@ def _in_vertical_line(
     return False
 
 
-def _rows(page: "fitz.Page") -> list[list[tuple]]:
+def _rows(page: "pymupdf.Page") -> list[list[tuple]]:
     words = page.get_text("words")  # x0, y0, x1, y1, text, block, line, word
     if not words:
         return []
@@ -177,7 +180,7 @@ def _two_column(rows: list[list[list[tuple]]]) -> bool:
     return both / len(rows) > 0.3
 
 
-def page_lines(doc: "fitz.Document", pageno: int) -> list[str]:
+def page_lines(doc: "pymupdf.Document", pageno: int) -> list[str]:
     """The lines of a page, in reading order, numbers joined to their text."""
     key = (id(doc), pageno)
     cached = _LINES_CACHE.get(key)
@@ -200,11 +203,11 @@ def page_lines(doc: "fitz.Document", pageno: int) -> list[str]:
     return lines
 
 
-def page_text(doc: "fitz.Document", pageno: int) -> str:
+def page_text(doc: "pymupdf.Document", pageno: int) -> str:
     return "\n".join(page_lines(doc, pageno))
 
 
-def detect_toc_pages(doc: "fitz.Document") -> set[int]:
+def detect_toc_pages(doc: "pymupdf.Document") -> set[int]:
     out = set()
     for i in range(doc.page_count):
         lines = [l for l in page_lines(doc, i) if l.strip()]
@@ -232,7 +235,7 @@ def furniture_key(line: str) -> str:
     return FURNITURE_EDGE_RE.sub("", s).strip()
 
 
-def detect_running_text(doc: "fitz.Document") -> set[str]:
+def detect_running_text(doc: "pymupdf.Document") -> set[str]:
     """Lines appearing on >=5 pages near the top/bottom (running headers/footers),
     keyed without their page number."""
     counter: dict[str, int] = defaultdict(int)
@@ -243,6 +246,60 @@ def detect_running_text(doc: "fitz.Document") -> set[str]:
     return {l for l, c in counter.items() if c >= 5 and len(l) > 8}
 
 
+LICENCE_LINE_RE = re.compile(r"copyright|©|licen[cs]|lizenz|tous droits|alle rechte", re.I)
+LICENCE_MAX_CHARS = 600
+
+
+def detect_licence(doc: "pymupdf.Document", running: set[str]) -> str:
+    """The licence marks of the copy: watermark and copyright footer, once.
+
+    The page text leaves them out (they would land on every line they cross),
+    but they are never thrown away: the norm keeps them in its `licence`
+    field and every excerpt served from it carries them.
+    - a vertical line that is a thin minority of its page and recurs on two
+      pages at least (the library watermark up the margin);
+    - a running header or footer that names a copyright or a licence.
+    """
+    seen: dict[str, int] = {}
+    for i in range(doc.page_count):
+        page = doc[i]
+        vertical = _vertical_lines(page)
+        if not vertical:
+            continue
+        total = len(page.get_text("words")) or 1
+        share = sum(len(words) for _, words in vertical) / total
+        if share > VERTICAL_WATERMARK_SHARE:
+            continue
+        texts: list[str] = []
+        for block in page.get_text("dict").get("blocks", []):
+            for line in block.get("lines", []):
+                dx, dy = line.get("dir", (1.0, 0.0))
+                if abs(dy) <= 0.1 and dx >= 0:
+                    continue
+                text = re.sub(
+                    r"\s+", " ", " ".join(s.get("text", "") for s in line.get("spans", []))
+                ).strip()
+                if len(text) > 3:
+                    texts.append(text)
+        for text in dict.fromkeys(texts):
+            seen[text] = seen.get(text, 0) + 1
+    marks = [t for t, count in seen.items() if count >= 2]
+    # `running` holds the footers keyed without their numbers ("SIA , Copyright
+    # by SIA Zurich"): the mark is the line as printed, page number aside.
+    keys = {l for l in running if LICENCE_LINE_RE.search(l)}
+    for i in range(doc.page_count):
+        if not keys:
+            break
+        lines = [l.strip() for l in page_lines(doc, i) if l.strip()]
+        for line in lines[:3] + lines[-3:]:
+            key = furniture_key(line)
+            if key in keys:
+                keys.discard(key)
+                marks.append(re.sub(r"^\d{1,4}\s+|\s+\d{1,4}$", "", line))
+    licence = "\n".join(dict.fromkeys(marks))
+    return licence[:LICENCE_MAX_CHARS].rstrip()
+
+
 CHAP_UPPER_RE = re.compile(
     r"^(\d+)\s+([A-ZÉÈÀÂÔÎÛÇ][^a-z\n]{3,120})\s*$", re.MULTILINE
 )
@@ -292,7 +349,7 @@ def join_split_headings(text: str) -> str:
     return "\n".join(out)
 
 
-def heading_text(doc: "fitz.Document", pageno: int) -> str:
+def heading_text(doc: "pymupdf.Document", pageno: int) -> str:
     """Page text prepared for heading detection."""
     return join_split_headings(page_text(doc, pageno))
 
@@ -329,7 +386,7 @@ def plausible_title(title: str) -> bool:
     return letters / len(t) >= 0.7
 
 
-def dense_heading_pages(doc: "fitz.Document", toc_pages: set[int]) -> set[int]:
+def dense_heading_pages(doc: "pymupdf.Document", toc_pages: set[int]) -> set[int]:
     """Pages listing many numbered headings: a contents page, whatever its
     typography. Detecting it by leader dots alone misses the ones set without
     them, and every heading read there points at the contents page instead of
@@ -407,7 +464,7 @@ def prune_chapters(chapters: dict[str, dict]) -> dict[str, dict]:
 
 
 def extract_chapters(
-    doc: "fitz.Document", toc_pages: set[int] | None = None
+    doc: "pymupdf.Document", toc_pages: set[int] | None = None
 ) -> dict[str, dict]:
     """Top-level chapters: prefer PDF bookmarks; fallback to body-text regex
     (uppercase heading on its own line) when the PDF has no bookmarks."""
@@ -492,7 +549,7 @@ LABEL_RE = re.compile(
 )
 
 
-def infer_labelled_chapters(doc: "fitz.Document", skip: set[int]) -> dict[str, dict]:
+def infer_labelled_chapters(doc: "pymupdf.Document", skip: set[int]) -> dict[str, dict]:
     """Chapters from labelled headings: "Art. 7 Titre" (a règlement, SIA 144)
     or "Section 7 Titre" (a Eurocode adopted as SIA 267.001).
 
@@ -558,7 +615,7 @@ SCOPE_TITLE_RE = re.compile(
 )
 
 
-def _page_top_titles(doc: "fitz.Document", pages: range, skip: set[int]) -> list[tuple[int, str]]:
+def _page_top_titles(doc: "pymupdf.Document", pages: range, skip: set[int]) -> list[tuple[int, str]]:
     """Uppercase, plausible titles among the first lines of each page: where a
     chapter of a SIA norm starts. Deeper in a page a capitalised line is a
     table header or a note."""
@@ -574,7 +631,7 @@ def _page_top_titles(doc: "fitz.Document", pages: range, skip: set[int]) -> list
     return out
 
 
-def recover_missing_chapters(doc: "fitz.Document", chapters: dict[str, dict], skip: set[int]) -> dict[str, dict]:
+def recover_missing_chapters(doc: "pymupdf.Document", chapters: dict[str, dict], skip: set[int]) -> dict[str, dict]:
     """Fill holes in the chapter sequence by position.
 
     On the OCR'd copies the numbering column is sometimes lost on a whole run
@@ -622,7 +679,7 @@ UNNUMBERED_UPPER_RE = re.compile(r"^([A-ZÉÈÀÂÔÎÛÇ][^a-z\n]{5,120})$")
 FIRST_SUB_RE = re.compile(r"^(\d{1,2})\.\d{1,2}\b")
 
 
-def infer_unnumbered_chapters(doc: "fitz.Document", skip: set[int]) -> dict[str, dict]:
+def infer_unnumbered_chapters(doc: "pymupdf.Document", skip: set[int]) -> dict[str, dict]:
     """Chapters whose number the OCR lost.
 
     On a scanned copy the large chapter number is often the one glyph the OCR
@@ -664,7 +721,7 @@ def chapter_page_spans(chapters: dict[str, dict], n_pages: int) -> dict[str, tup
 
 
 def extract_subsections(
-    doc: "fitz.Document",
+    doc: "pymupdf.Document",
     chap_prefixes: set[str],
     toc_pages: set[int],
     running: set[str],
@@ -870,7 +927,7 @@ def locate_heading(lines: list[str], node: dict) -> int:
 
 
 def extract_section_text(
-    doc: "fitz.Document", nodes: list[dict], running: set[str]
+    doc: "pymupdf.Document", nodes: list[dict], running: set[str]
 ) -> list[str]:
     """Give each node the text between its heading and the next one.
 
@@ -998,7 +1055,7 @@ def prune_subsections(
 
 
 def extract_figures(
-    doc: "fitz.Document",
+    doc: "pymupdf.Document",
     toc_pages: set[int],
     out_dir: Path,
     dpi: int = 150,
@@ -1007,7 +1064,7 @@ def extract_figures(
     figs_dir = out_dir / "figures"
     figs_dir.mkdir(parents=True, exist_ok=True)
     seen: dict[str, dict] = {}
-    matrix = fitz.Matrix(dpi / 72, dpi / 72)
+    matrix = pymupdf.Matrix(dpi / 72, dpi / 72)
     for pageno in range(doc.page_count):
         if pageno in toc_pages:
             continue
@@ -1037,7 +1094,7 @@ def extract_figures(
     return list(seen.values())
 
 
-def guess_language(doc: "fitz.Document") -> str:
+def guess_language(doc: "pymupdf.Document") -> str:
     """Best-effort language guess based on common French/German/Italian markers."""
     sample = " ".join(page_text(doc, i) for i in range(min(5, doc.page_count))).lower()
     scores = {
@@ -1059,6 +1116,11 @@ def main() -> None:
         help="Override detected language.",
     )
     ap.add_argument("--dpi", type=int, default=150, help="Figure render DPI.")
+    ap.add_argument(
+        "--licence-only",
+        action="store_true",
+        help="Print the licence marks of the PDF as JSON and stop (backfill of an ingested norm).",
+    )
     args = ap.parse_args()
 
     pdf_path = Path(args.pdf).resolve()
@@ -1066,11 +1128,15 @@ def main() -> None:
     if not pdf_path.exists():
         sys.stderr.write(f"PDF not found: {pdf_path}\n")
         sys.exit(1)
-    out_dir.mkdir(parents=True, exist_ok=True)
 
-    doc = fitz.open(str(pdf_path))
-    toc_pages = detect_toc_pages(doc)
+    doc = pymupdf.open(str(pdf_path))
     running = detect_running_text(doc)
+    licence = detect_licence(doc, running)
+    if args.licence_only:
+        sys.stdout.write(json.dumps({"ok": True, "licence": licence}, ensure_ascii=False) + "\n")
+        return
+    out_dir.mkdir(parents=True, exist_ok=True)
+    toc_pages = detect_toc_pages(doc)
     chapters = extract_chapters(doc, toc_pages)
     chap_prefixes = {k for k in chapters if k.isdigit()}
     spans = chapter_page_spans(chapters, doc.page_count)
@@ -1115,6 +1181,9 @@ def main() -> None:
             "metadata": {k: v for k, v in doc.metadata.items() if v},
             "tocPagesDetected": sorted(toc_pages),
             "runningHeadersDetected": sorted(running)[:10],
+            # Watermark and copyright, kept once: `ingest_create_document`
+            # stores them on the norm, and every excerpt carries them.
+            "licence": licence,
         },
         "stats": {
             "sectionCount": len(nodes),

package/scripts/tests/test_prepass.py

@@ -15,7 +15,10 @@ import importlib.util
 import sys
 from pathlib import Path
 
-import fitz
+try:
+    import pymupdf
+except ImportError:  # PyMuPDF < 1.24.3 only knows the legacy name
+    import fitz as pymupdf
 import pytest
 
 HERE = Path(__file__).resolve().parent
@@ -29,7 +32,7 @@ spec.loader.exec_module(prepass)  # type: ignore[union-attr]
 
 def write_pdf(path: Path, pages: list[list[str]], toc: list[list] | None = None) -> None:
     """One text line per list entry, 16 pt apart, top to bottom."""
-    doc = fitz.open()
+    doc = pymupdf.open()
     for lines in pages:
         page = doc.new_page(width=595, height=842)
         y = 60
@@ -44,7 +47,7 @@ def write_pdf(path: Path, pages: list[list[str]], toc: list[list] | None = None)
 
 def run(pdf: Path, out: Path) -> dict:
     """The pre-pass as `main()` runs it, without the CLI."""
-    doc = fitz.open(str(pdf))
+    doc = pymupdf.open(str(pdf))
     toc_pages = prepass.detect_toc_pages(doc)
     running = prepass.detect_running_text(doc)
     chapters = prepass.extract_chapters(doc, toc_pages)
@@ -209,3 +212,24 @@ def test_recovers_chapter_numbers_by_position(tmp_path):
     assert chapters["0"]["pageStart"] == 1
     assert prepass.missing_chapter_numbers(chapters) == []
 
+
+
+WATERMARK = "Ecole Polytechnique Federale de Lausanne, EPFL, SNV / licensed to"
+
+
+def test_licence_marks_are_kept_once(tmp_path):
+    """The footer and the vertical watermark leave the text, never the norm."""
+    pdf = sia_like(tmp_path)
+    doc = pymupdf.open(str(pdf))
+    for page in doc:
+        page.insert_text((20, 700), WATERMARK, fontsize=7, fontname="helv", rotate=90)
+    marked = tmp_path / "marked.pdf"
+    doc.save(str(marked))
+    doc.close()
+
+    doc = pymupdf.open(str(marked))
+    licence = prepass.detect_licence(doc, prepass.detect_running_text(doc))
+    lines = licence.splitlines()
+    assert HEADER in lines
+    assert WATERMARK in lines
+    assert len(lines) == len(set(lines))

package/skills/consult-stratta/SKILL.md

@@ -7,7 +7,7 @@ description: Use Stratta to answer civil engineering questions from licensed SIA
 
 Use this skill when the user asks a technical civil engineering question that should be answered from the norms available in Stratta, or when the user asks to resume, create or update a Stratta project dossier.
 
-You answer practising engineers from the norms their organization holds in Stratta, and only from them. The corpus below is the whole of what you can read: a norm absent from it cannot be consulted, and a question outside it is answered by saying so, not by general knowledge. Reply in the language the user writes in; quote the norm in the language it is written in and translate only when asked.
+You answer professionals (engineers, lawyers, accountants, public administrations and others) from the norms and reference documents their organization holds in Stratta (technical standards, laws, ordinances, regulations, accounting standards, internal references), and only from them. The corpus below is the whole of what you can read: a document absent from it cannot be consulted, and a question outside it is answered by saying so, not by general knowledge. "Norm" below means any such document. Reply in the language the user writes in; quote the norm in the language it is written in and translate only when asked.
 
 ## First call
 
@@ -19,7 +19,7 @@ Call `get_methodology` before the first technical answer of a conversation: it r
 2. `get_toc(code, maxDepth: 1)` on the norm that governs, then `get_subtree(code, chapter, maxDepth: 2)` on the chapter that fits: read summaries, choose the section.
 3. `get_section(code, path)` on the section retained. It comes as a document with its formulas and tables in place, its page range, the sections below it and the references it makes. Descend by `children` rather than loading a chapter whole; when `truncated` is true, call again with `offset`.
 4. Follow the references marked resolvable: the rule, the coefficient and the action are often spread over separate clauses or norms, and the answer is in the one you have not read yet.
-5. `get_figure` when a section points at a figure the reasoning depends on: a load diagram, a design chart, a standard cross-section.
+5. `get_figure` when a section points at a figure the reasoning depends on: a diagram, a chart, a standard cross-section.
 6. `search_in_norm` inside one norm when its table of contents suggests no obvious section. Search in the language of the norm; accents and case do not matter; a clause number as keyword returns that clause.
 
 When the question names a norm absent from the corpus but a norm you hold refers to it, read that clause first: the pointer (code and clause number) usually lives there, and it is an answer.
@@ -27,10 +27,11 @@ When the question names a norm absent from the corpus but a norm you hold refers
 ## Answer contract
 
 - Every technical statement rests on a section you read in this conversation and cites it as `[CODE § path, p. N]`, for example `[SIA 267 § 9.5.2.1, p. 63]`. A value with no clause is not an answer.
-- A calculation shows its inputs, each with its citation, its formula in LaTeX (`$$...$$`) with every symbol defined and its unit (kN, kPa, MPa, m).
+- A calculation shows its inputs, each with its citation, its formula in LaTeX (`$$...$$`) with every symbol defined and its unit (kN, MPa, CHF, %).
 - Before answering, check: each value has its clause; each clause was read, not remembered from a summary; the resolvable references were followed; what the corpus does not cover is named, with the norm or clause where it is likely to be found.
 - When several norms are cited, group by norm under headings: principle, then method, then conditions of use.
-- Once per conversation, say that the answer is indicative and that the full text of the norm must be consulted before use on a project.
+- A norm that comes with a `licence` (watermark, copyright) carries it with every quote: print it once under the passages quoted from that norm, verbatim.
+- Once per conversation, say that the answer is indicative and that the full text of the document must be consulted before it is relied on.
 
 ## Budget
 
@@ -38,27 +39,27 @@ No `get_toc` without `maxDepth`. No section loaded whole when its children would
 
 ## Demonstration document
 
-DEMO 001 is a fictional document shipped with Stratta so that navigation and citation can be exercised before a licensed norm is ingested. Answer a question about it exactly as you would for a norm: read the section, cite it as [DEMO 001 §path, p. N], give the value. Then say once, plainly, that the document is fictional and that none of its values may be used in a project: the office must ingest the norm it holds a licence for. Never refuse to read it, and never present its values as those of a real norm.
+DEMO 001 is a fictional document shipped with Stratta so that navigation and citation can be exercised before the organization has deposited its own documents. Answer a question about it exactly as you would for a norm: read the section, cite it as [DEMO 001 §path, p. N], give the value. Then say once, plainly, that the document is fictional and that none of its values may be relied on: the organization must deposit the documents it holds the rights to. Never refuse to read it, and never present its values as those of a real norm.
 
 ## Dossier
 
 A conversation disappears, a project lasts months. The dossier is what remains: what was read, what was retained, and what is still open. Keeping it costs the user nothing: every section you read is filed under the project by itself, and one call records a decision.
 
-**Name the project once.** When the user names a project, a structure, a site or a mandate, call `list_dossiers` then `load_dossier` (or `open_dossier` for a new one) BEFORE answering. It reloads what was already decided so it is not decided twice, and from then on each `get_section` you make is journaled under that dossier without any further call. Do not open a dossier for a one-off lookup with no project behind it.
+**Name the project once.** When the user names a project, a case, a structure, a site or a mandate, call `list_dossiers` then `load_dossier` (or `open_dossier` for a new one) BEFORE answering. It reloads what was already decided so it is not decided twice, and from then on each `get_section` you make is journaled under that dossier without any further call. Do not open a dossier for a one-off lookup with no project behind it.
 
-**Record what is retained** (`record_decision`) as soon as the user fixes a value or a choice for the design ("we take 30 degrees", "keep pile P38"): the decision in one line, why, the clause it rests on (`normCode`, `sectionPath`, `page`, `normEdition` from the header of the section read), the number in `value` with its unit and `confidence` (established, judgement or to_confirm; in geotechnics a retained value is a judgement). Pass `dossierId` and `question` (what is settled, in one line): the question is opened already decided. Pass `questionId` instead when the question exists. ONLY after the user confirmed it, never on your own reasoning.
+**Record what is retained** (`record_decision`) as soon as the user fixes a value or a choice for the design ("we take 30 degrees", "keep option B", "the limitation period is 10 years"): the decision in one line, why, the clause it rests on (`normCode`, `sectionPath`, `page`, `normEdition` from the header of the section read), the number in `value` with its unit and `confidence` (established, judgement or to_confirm; a retained value that rests on professional judgement is a judgement). Pass `dossierId` and `question` (what is settled, in one line): the question is opened already decided. Pass `questionId` instead when the question exists. ONLY after the user confirmed it, never on your own reasoning.
 
 **Open a question** (`open_question`) only when something is left to settle later, or when the user weighs several options and wants them on the table. File what backs it with `save_finding` (one clause, one value or one fact per call, under its `questionId`): reference, hypothesis or observation. `get_question` reads one in full; `update_question` edits it; never delete one unless the user says so.
 
-**Read its attachments.** A dossier carries the project's own documents (site reports, borehole logs, meeting minutes): `list_attachments`, then `read_attachment` on what bears on the question, or `search_in_dossier` when the dossier holds many. A fact from a piece is evidence like a clause: cite it as [name, p. N] and file it with `attachmentId` and `attachmentPage`.
+**Read its attachments.** A dossier carries the project's own documents (reports, logs, meeting minutes, contracts): `list_attachments`, then `read_attachment` on what bears on the question, or `search_in_dossier` when the dossier holds many. A fact from a piece is evidence like a clause: cite it as [name, p. N] and file it with `attachmentId` and `attachmentPage`.
 
-**Start from a template when one fits** (`list_templates`, `apply_template`): the office's checklist for a type of structure opens the questions with the clauses they rest on, and says which clauses the corpus lacks.
+**Start from a template when one fits** (`list_templates`, `apply_template`): the organization's checklist for a type of case opens the questions with the clauses they rest on, and says which clauses the corpus lacks.
 
-**What not to write:** your own prose, the intermediate steps, anything the user did not treat as a decision. The dossier documents the engineer's reasoning; it does not replace it: never conclude that a structure complies, never write a decision the user has not confirmed.
+**What not to write:** your own prose, the intermediate steps, anything the user did not treat as a decision. The dossier documents the professional's reasoning; it does not replace it: never conclude that a design or a situation complies, never write a decision the user has not confirmed.
 
 ## Skills
 
-Before drafting a document (project basis, use agreement, terrain chapter, verification note) or starting a repeatable task (instructing a type of structure, preparing a review), call `list_skills` then `get_skill` on the one that fits, and follow it. The organization's own skill replaces Stratta's of the same name: it carries the office's plan and wording. When no skill fits, say so and proceed with the method above.
+Before drafting a document (project basis, use agreement, terrain chapter, verification note, opinion) or starting a repeatable task (instructing a type of case, preparing a review), call `list_skills` then `get_skill` on the one that fits, and follow it. The organization's own skill replaces Stratta's of the same name: it carries the organization's plan and wording. When no skill fits, say so and proceed with the method above.
 
 ## Boundaries
 

package/skills/ingest-norm/SKILL.md

@@ -86,7 +86,10 @@ Output under `.stratta-ingest/<code-slug>/`:
 
 `prepass.json` structure:
 
-- `doc` — `pageCount`, detected `language`, `tocSource`, raw `metadata`.
+- `doc` — `pageCount`, detected `language`, `tocSource`, raw `metadata`,
+  and `licence`: the watermark and copyright lines of the copy, kept once.
+  The page text leaves them out; the norm keeps them and every excerpt
+  served from it shows them.
 - `stats` — `sectionCount`, `byDepth`, `figureCount`, `charsPerPage`,
   `emptyLeaves`. `charsPerPage` under 300 means the text layer is missing
   (a scan without OCR): stop and tell the user.
@@ -113,10 +116,12 @@ numbered clauses stay inside their heading's text.
 ### 4. Create the document
 
 Read `prepass.json`, then:
-`ingest_create_document { code, year, title, language: <doc.language>, totalPages: <doc.pageCount>, scope }`
+`ingest_create_document { code, year, title, language: <doc.language>, totalPages: <doc.pageCount>, scope, licence: <doc.licence> }`
 where `scope` is one sentence on what the norm covers (from its first chapter);
 it is shown to every agent in the served methodology, so write it as a routing
 hint: "Geotechnical design: foundations, piles, anchors, retaining structures".
+Pass `licence` exactly as the pre-pass wrote it, never edited or shortened;
+omit it only when the pre-pass found none.
 → returns `documentId`. Keep it for every subsequent call.
 
 ### 5. Validate the tree, write the summaries