euparliamentmonitor 0.8.43 → 0.8.44

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "euparliamentmonitor",
-  "version": "0.8.43",
+  "version": "0.8.44",
   "type": "module",
   "description": "European Parliament Intelligence Platform - Monitor political activity with systematic transparency",
   "main": "scripts/index.js",

package/scripts/aggregator/analysis-aggregator.d.ts

@@ -165,4 +165,13 @@ export declare function renderAnalysisIndex(included: readonly IncludedArtifact[
  * @returns {@link AggregatedRun} describing the rendered document
  */
 export declare function aggregateAnalysisRun(options: AggregateOptions): AggregatedRun;
+/**
+ * Extract a `YYYY-MM-DD` date from a path like
+ * `analysis/daily/2026-01-15/run`. Falls back to the epoch date when no
+ * ISO date is embedded in the path.
+ *
+ * @param runDirRelPath - Repo-relative path of the run directory
+ * @returns ISO date string in `YYYY-MM-DD` form
+ */
+export declare function guessDateFromRunDir(runDirRelPath: string): string;
 //# sourceMappingURL=analysis-aggregator.d.ts.map

package/scripts/aggregator/analysis-aggregator.js

@@ -119,6 +119,24 @@ export function discoverTradecraftFiles(repoRoot) {
     }
     return result.sort();
 }
+/**
+ * Return `true` when a `.md` filename should be excluded from the run
+ * artifact set. Keeps the walk closure under the cognitive-complexity limit.
+ *
+ * Excluded names:
+ *  - `article.md` and translated variants (`article.sv.md`, etc.) — these are
+ *    outputs of the aggregator, not inputs.
+ *  - `README.md` (case-insensitive) — required for the analysis gate but not
+ *    relevant to the published article.
+ *
+ * @param name - Bare filename (no directory prefix)
+ * @returns `true` when the file should be skipped
+ */
+function isExcludedArtifact(name) {
+    if (name.toLowerCase() === 'readme.md')
+        return true;
+    return name.startsWith('article.') && name.endsWith('.md');
+}
 /**
  * Walk the run directory and return every `.md` file as a run-relative
  * POSIX path, excluding files under `data/` (raw MCP payloads, not meant
@@ -137,11 +155,13 @@ function collectRunArtifacts(runDir) {
             const full = path.join(dir, entry.name);
             const rel = prefix ? `${prefix}/${entry.name}` : entry.name;
             if (entry.isDirectory()) {
-                if (entry.name === 'data' || entry.name === 'runs')
+                // Skip raw payloads, legacy run snapshots, and Pass-1 work-in-progress
+                // snapshots so they are not rendered as supplementary artifacts.
+                if (entry.name === 'data' || entry.name === 'runs' || entry.name === 'pass1')
                     continue;
                 walk(full, rel);
             }
-            else if (entry.isFile() && entry.name.endsWith('.md')) {
+            else if (entry.isFile() && entry.name.endsWith('.md') && !isExcludedArtifact(entry.name)) {
                 result.push(rel);
             }
         }
@@ -471,7 +491,7 @@ export function aggregateAnalysisRun(options) {
  * @param runDirRelPath - Repo-relative path of the run directory
  * @returns ISO date string in `YYYY-MM-DD` form
  */
-function guessDateFromRunDir(runDirRelPath) {
+export function guessDateFromRunDir(runDirRelPath) {
     const match = /(\d{4}-\d{2}-\d{2})/.exec(runDirRelPath);
     return match ? (match[1] ?? '1970-01-01') : '1970-01-01';
 }

package/scripts/aggregator/article-generator.d.ts

@@ -38,6 +38,12 @@ export interface CliOptions {
 export interface GenerateResult {
     /** Repo-relative path of the English source Markdown that was written. */
     readonly sourceMarkdownRelPath: string;
+    /**
+     * Repo-relative path of the `article.md` written directly into the
+     * analysis run directory — canonical Markdown source that lives alongside
+     * the artifacts that produced it (riksdagsmonitor pattern).
+     */
+    readonly runArticleMdRelPath: string;
     /** Filenames written under `outDir`, relative to `outDir`. */
     readonly writtenFiles: readonly string[];
     /** Metadata from {@link aggregateAnalysisRun}. */

package/scripts/aggregator/article-generator.js

@@ -310,6 +310,7 @@ function writeLanguageVariant(lang, slug, aggregated, englishHtml, chromeOptions
         sourceMarkdownRelPath: chromeOptions.sourceMarkdownRelPath,
         toc: aggregated.sectionToc,
         articleCount: chromeOptions.articleCount,
+        isBasedOn: aggregated.includedArtifacts.map((a) => `https://github.com/Hack23/euparliamentmonitor/blob/main/${a.repoRelPath}`),
     });
     const filename = getArticleFilename(slug, lang);
     fs.writeFileSync(path.join(opts.outDir, filename), html, 'utf8');
@@ -396,18 +397,31 @@ export function generateArticle(opts, runSuffix, articleCountOverride) {
     const effectiveMetadata = opts.title || opts.description
         ? applyCliOverrides(resolvedMetadata, opts.title, opts.description)
         : resolvedMetadata;
-    // Write source Markdown under <outDir>/<slug>.en.md for transparency.
+    // Write article.md INTO the analysis run directory — canonical Markdown
+    // source that lives alongside the artifacts that produced it.
+    // This mirrors the riksdagsmonitor pattern where `article.md` is committed
+    // inside `analysis/daily/<date>/<type>/` so every run has a browsable,
+    // version-controlled Markdown source in its own directory.
+    const runArticleMdAbs = path.join(opts.runDir, 'article.md');
+    fs.writeFileSync(runArticleMdAbs, aggregated.markdown, 'utf8');
+    const runArticleMdRelPath = path
+        .relative(opts.repoRoot, runArticleMdAbs)
+        .split(path.sep)
+        .join('/');
+    // Also write source Markdown under <outDir>/<slug>.en.md for search
+    // indexing and backwards compatibility with existing news-index scripts.
     ensureDir(opts.outDir);
     const sourceMdFilename = `${slug}.en.md`;
     const sourceMdAbs = path.join(opts.outDir, sourceMdFilename);
     fs.writeFileSync(sourceMdAbs, aggregated.markdown, 'utf8');
-    const sourceMdRelPath = path.relative(opts.repoRoot, sourceMdAbs).split(path.sep).join('/');
     const written = [sourceMdFilename];
     if (!opts.markdownOnly) {
         const rendered = renderMarkdown(aggregated.markdown);
         const chromeOptions = {
             metadata: effectiveMetadata,
-            sourceMarkdownRelPath: sourceMdRelPath,
+            // Point the "View source Markdown" link at the canonical run-directory
+            // article.md so readers can trace the HTML back to the analysis tree.
+            sourceMarkdownRelPath: runArticleMdRelPath,
             articleCount: articleCountOverride ?? countPublishedArticles(opts.repoRoot),
         };
         for (const lang of opts.langs) {
@@ -415,7 +429,12 @@ export function generateArticle(opts, runSuffix, articleCountOverride) {
             written.push(filename);
         }
     }
-    return { sourceMarkdownRelPath: sourceMdRelPath, writtenFiles: written, aggregated };
+    return {
+        sourceMarkdownRelPath: runArticleMdRelPath,
+        runArticleMdRelPath,
+        writtenFiles: written,
+        aggregated,
+    };
 }
 /**
  * Walk `analysis/daily/` recursively and return every subdirectory that

package/scripts/aggregator/article-html.d.ts

@@ -44,6 +44,11 @@ export interface WrapArticleOptions {
      * the line is omitted.
      */
     readonly articleCount?: number;
+    /**
+     * Optional: URLs of source artifacts included in the aggregated article.
+     * Emitted as `isBasedOn` in the JSON-LD `NewsArticle` schema for provenance.
+     */
+    readonly isBasedOn?: readonly string[];
 }
 /**
  * Build the canonical filename for an article in a given language. English

package/scripts/aggregator/article-html.js

@@ -134,6 +134,11 @@ export function wrapArticleHtml(options) {
             name: 'EU Parliament Monitor',
             url: BASE_URL,
         },
+        ...(options.isBasedOn && options.isBasedOn.length > 0
+            ? {
+                isBasedOn: options.isBasedOn.map((url) => ({ '@type': 'CreativeWork', url })),
+            }
+            : {}),
     };
     const jsonLdString = JSON.stringify(jsonLd).replace(/</g, '\\u003c');
     const pageTitle = `${options.title} — ${siteTitle}`;

package/scripts/aggregator/clean-artifact.d.ts

@@ -35,6 +35,8 @@ export interface CleanArtifactResult {
     readonly strippedH1s: number;
     /** Banner/metadata lines removed. */
     readonly strippedBannerLines: number;
+    /** Operational metadata preamble lines removed (e.g. **Run:** / **Window:** blocks). */
+    readonly strippedMetaLines: number;
     /** Mermaid blocks deduplicated as a reference to a previous occurrence. */
     readonly dedupedMermaidBlocks: number;
 }
@@ -126,6 +128,28 @@ export declare function dedupMermaid(md: string, seen: Set<string>): {
     md: string;
     deduped: number;
 };
+/**
+ * Strip the operational metadata preamble that agent pipelines prepend to
+ * artifacts. These are lines of the form `**Run:** …`, `**Window:** …`,
+ * `**Methodology:** …`, etc., followed optionally by a standalone `---`
+ * horizontal rule. They are agent-operational metadata that should not appear
+ * in the published article.
+ *
+ * Algorithm:
+ *  1. Skip leading blank lines (they don't count as metadata).
+ *  2. If the first non-blank line does NOT match the metadata pattern, return
+ *     the document unchanged (`lines: 0`).
+ *  3. Otherwise consume all metadata lines and interspersed blank lines.
+ *  4. If the next non-blank line is a standalone `---`, consume that too.
+ *  5. Return the stripped Markdown and the count of lines removed.
+ *
+ * @param md - Markdown source (after banner/heading passes)
+ * @returns `{ md, lines }` — stripped Markdown and number of lines removed
+ */
+export declare function stripArtifactMetadataPreamble(md: string): {
+    md: string;
+    lines: number;
+};
 /**
  * Apply all cleanup passes and return the normalised Markdown plus
  * simple counters for telemetry/tests.

package/scripts/aggregator/clean-artifact.js

@@ -578,6 +578,63 @@ function hashString(input) {
     }
     return (h >>> 0).toString(16).padStart(8, '0');
 }
+/**
+ * Pattern matching an operational metadata line at the start of an artifact.
+ * Examples: `**Run:** breaking-run-123`, `**Window:** 2026-04-24 00:00Z — 05:49Z`.
+ * The pattern requires the line to start with `**<Word>**` followed by a colon
+ * or whitespace so ordinary bold prose is not mistakenly treated as metadata.
+ */
+const METADATA_LINE_PATTERN = /^\*\*[A-Za-z][^*\n]*\*\*[:\s]/;
+/**
+ * Strip the operational metadata preamble that agent pipelines prepend to
+ * artifacts. These are lines of the form `**Run:** …`, `**Window:** …`,
+ * `**Methodology:** …`, etc., followed optionally by a standalone `---`
+ * horizontal rule. They are agent-operational metadata that should not appear
+ * in the published article.
+ *
+ * Algorithm:
+ *  1. Skip leading blank lines (they don't count as metadata).
+ *  2. If the first non-blank line does NOT match the metadata pattern, return
+ *     the document unchanged (`lines: 0`).
+ *  3. Otherwise consume all metadata lines and interspersed blank lines.
+ *  4. If the next non-blank line is a standalone `---`, consume that too.
+ *  5. Return the stripped Markdown and the count of lines removed.
+ *
+ * @param md - Markdown source (after banner/heading passes)
+ * @returns `{ md, lines }` — stripped Markdown and number of lines removed
+ */
+export function stripArtifactMetadataPreamble(md) {
+    const lines = md.split('\n');
+    let i = 0;
+    // Skip purely blank lines at the very head
+    while (i < lines.length && (lines[i] ?? '').trim() === '')
+        i++;
+    // If the first real line is not a metadata line, return unchanged
+    if (i >= lines.length || !METADATA_LINE_PATTERN.test(lines[i] ?? '')) {
+        return { md, lines: 0 };
+    }
+    // Consume the metadata block (metadata lines + interspersed blank lines)
+    let metaEnd = i;
+    while (metaEnd < lines.length) {
+        const line = lines[metaEnd] ?? '';
+        if (METADATA_LINE_PATTERN.test(line) || line.trim() === '') {
+            metaEnd++;
+        }
+        else {
+            break;
+        }
+    }
+    // If the next non-blank line is a standalone HR, absorb it
+    let scanAhead = metaEnd;
+    while (scanAhead < lines.length && (lines[scanAhead] ?? '').trim() === '')
+        scanAhead++;
+    if (scanAhead < lines.length && /^\s*---\s*$/.test(lines[scanAhead] ?? '')) {
+        metaEnd = scanAhead + 1;
+    }
+    const removed = metaEnd;
+    const stripped = lines.slice(removed).join('\n').replace(/^\n+/, '');
+    return { md: stripped, lines: removed };
+}
 /**
  * Apply all cleanup passes and return the normalised Markdown plus
  * simple counters for telemetry/tests.
@@ -594,6 +651,8 @@ export function cleanArtifact(source, options) {
     md = mdAfterBanners;
     const { md: mdAfterHeadings, h1Count } = demoteHeadings(md);
     md = mdAfterHeadings;
+    const { md: mdAfterMeta, lines: strippedMetaLines } = stripArtifactMetadataPreamble(md);
+    md = mdAfterMeta;
     md = rewriteLinks(md, options.artifactRelPath);
     const { md: mdAfterMermaid, deduped } = dedupMermaid(md, seen);
     md = mdAfterMermaid;
@@ -603,6 +662,7 @@ export function cleanArtifact(source, options) {
         markdown: md,
         strippedH1s: h1Count,
         strippedBannerLines,
+        strippedMetaLines,
         dedupedMermaidBlocks: deduped,
     };
 }

package/scripts/mcp/imf-mcp-client.d.ts

@@ -9,21 +9,25 @@
  * because the upstream project is a Python git-URL package (not npm) and
  * could not be pinned to an integrity hash per the ISMS Secure Development
  * Policy §7. This module replaces the Python transport with a direct,
- * typed HTTP client — the public API is preserved so callers
- * (`src/utils/imf-data.ts`, validator fingerprints, workflow probes) are
- * untouched.
+ * typed HTTP client. The public API (`IMFMCPClient`, five tool methods,
+ * `MCPToolResult`-shaped envelope) is stable across the migration. The
+ * earlier companion module `src/utils/imf-data.ts` (SDMX-JSON parser,
+ * indicator/country maps, HTML builders) was purged in the April-2026
+ * aggregator-pipeline migration — callers now consume the raw SDMX-JSON
+ * envelope returned by {@link IMFMCPClient.fetchData} directly.
  *
  * ## Public API (unchanged from the MCP-backed iteration)
  *
  * - {@link IMFMCPClient} — class with semantic wrappers for five "tools".
  * - {@link IMF_MCP_TOOLS} — stable virtual tool-name list used by the
- *   content-validator fingerprint and the workflow probe.
+ *   Stage-C editorial fingerprint and the workflow probe. Drift-guarded
+ *   by `test/integration/mcp/imf-mcp.test.js`.
  * - {@link getIMFMCPClient} / {@link closeIMFMCPClient} — singleton lifecycle.
  *
  * The return envelope of every method is {@link MCPToolResult}
- * (`{ content: [{ type: "text", text: "<json>" }] }`) so downstream code
- * that already calls `parseSDMXJSON(response.content[0]?.text)` continues
- * to work unmodified.
+ * (`{ content: [{ type: "text", text: "<json>" }] }`). The `text` payload
+ * is the raw SDMX-JSON document returned by the IMF REST endpoint;
+ * downstream code parses it with any standard SDMX-JSON reader.
  *
  * ## Transport
  *
@@ -50,7 +54,11 @@ import type { MCPToolResult, MCPClientOptions } from '../types/index.js';
  * longer talks to an MCP server, but the tool-name list is preserved so
  * it continues to serve as:
  *
- * 1. The content-validator fingerprint source (`IMF_STRONG_FINGERPRINTS`).
+ * 1. The Stage-C editorial fingerprint source for "IMF is cited" (see
+ *    `analysis/imf/indicator-catalog.md §6` — the earlier runtime
+ *    fingerprint table `IMF_STRONG_FINGERPRINTS` in
+ *    `src/utils/content-validator.ts` was purged in the April-2026
+ *    aggregator-pipeline migration).
  * 2. The workflow probe's heartbeat identifiers.
  * 3. A drift guard against method additions: if a new helper method lands
  *    here, `test/integration/mcp/imf-mcp.test.js` fails unless the list
@@ -169,8 +177,10 @@ export declare class IMFMCPClient {
      * Fetch a time-series slice from an IMF dataflow as SDMX-JSON.
      *
      * Virtual tool: `imf-fetch-data`. The response is already in SDMX-JSON
-     * format, so {@link parseSDMXJSON} (`src/utils/imf-data.ts`) can
-     * consume `response.content[0]?.text` directly without reshaping.
+     * format; callers read the series under `data.dataSets[0].series`
+     * using any standard SDMX-JSON reader. (The earlier helper
+     * `parseSDMXJSON` in `src/utils/imf-data.ts` was purged in the
+     * April-2026 aggregator-pipeline migration.)
      *
      * @param options - Fetch parameters.
      * @param options.databaseId - IMF dataflow ID (`"WEO"`, `"IFS"`, ...).

package/scripts/mcp/imf-mcp-client.js

@@ -14,7 +14,11 @@ const IMF_FALLBACK = {
  * longer talks to an MCP server, but the tool-name list is preserved so
  * it continues to serve as:
  *
- * 1. The content-validator fingerprint source (`IMF_STRONG_FINGERPRINTS`).
+ * 1. The Stage-C editorial fingerprint source for "IMF is cited" (see
+ *    `analysis/imf/indicator-catalog.md §6` — the earlier runtime
+ *    fingerprint table `IMF_STRONG_FINGERPRINTS` in
+ *    `src/utils/content-validator.ts` was purged in the April-2026
+ *    aggregator-pipeline migration).
  * 2. The workflow probe's heartbeat identifiers.
  * 3. A drift guard against method additions: if a new helper method lands
  *    here, `test/integration/mcp/imf-mcp.test.js` fails unless the list
@@ -375,8 +379,10 @@ export class IMFMCPClient {
      * Fetch a time-series slice from an IMF dataflow as SDMX-JSON.
      *
      * Virtual tool: `imf-fetch-data`. The response is already in SDMX-JSON
-     * format, so {@link parseSDMXJSON} (`src/utils/imf-data.ts`) can
-     * consume `response.content[0]?.text` directly without reshaping.
+     * format; callers read the series under `data.dataSets[0].series`
+     * using any standard SDMX-JSON reader. (The earlier helper
+     * `parseSDMXJSON` in `src/utils/imf-data.ts` was purged in the
+     * April-2026 aggregator-pipeline migration.)
      *
      * @param options - Fetch parameters.
      * @param options.databaseId - IMF dataflow ID (`"WEO"`, `"IFS"`, ...).

package/scripts/types/imf.d.ts

@@ -22,12 +22,15 @@
  * `c-cf/imf-data-mcp` MCP server. That transport was replaced with a
  * native TypeScript HTTP client so the stack remains npm-pure and
  * pinned per ISMS §7. The `IMF_MCP_TOOLS` identifier list is retained
- * as a stable "virtual tool" surface for the content-validator
- * fingerprint and workflow probes.
+ * as a stable "virtual tool" surface for the Stage-C editorial
+ * fingerprint (see `analysis/imf/indicator-catalog.md §6`) and the
+ * workflow probe. The earlier companion module `src/utils/imf-data.ts`
+ * (parsers, indicator/country maps, HTML builders) was purged in the
+ * April-2026 aggregator-pipeline migration.
  *
  * @see {@link https://dataservices.imf.org/REST/SDMX_3.0 | IMF SDMX 3.0 REST API}
- * @see `analysis/methodologies/imf-indicator-mapping.md` for the committee →
- *   IMF indicator mapping enforced by the validator.
+ * @see {@link https://github.com/Hack23/euparliamentmonitor/blob/main/analysis/methodologies/imf-indicator-mapping.md | IMF Indicator Mapping methodology} for the committee →
+ *   IMF indicator mapping enforced at Stage-C editorial review.
  */
 import type { MCPClientOptions } from './mcp.js';
 /**
@@ -144,11 +147,17 @@ export interface IMFForecastPoint {
 /**
  * Curated set of IMF indicator keys used across the EU Parliament Monitor.
  *
- * These map onto specific IMF database + SDMX dimension codes — see
- * `IMF_POLICY_INDICATORS` in `src/utils/imf-data.ts` for the concrete
- * mapping. The key names are chosen to parallel the World Bank
- * `PolicyRelevantIndicators` keys where an equivalent series exists, so
- * article generators and templates can migrate incrementally.
+ * These map onto specific IMF database + SDMX dimension codes. The key
+ * names are chosen to parallel the World Bank `PolicyRelevantIndicators`
+ * keys where an equivalent series exists, so article generators and
+ * templates can migrate incrementally.
+ *
+ * The earlier runtime constant `IMF_POLICY_INDICATORS` in
+ * `src/utils/imf-data.ts` was purged in the April-2026
+ * aggregator-pipeline migration.
+ *
+ * @see {@link https://github.com/Hack23/euparliamentmonitor/blob/main/analysis/imf/indicator-catalog.md#2-policy-domain--imf-indicator-mapping | IMF indicator catalog §2}
+ * @see {@link https://github.com/Hack23/euparliamentmonitor/blob/main/analysis/methodologies/imf-indicator-mapping.md | IMF per-committee indicator mapping}
  */
 export type IMFMacroIndicatorKey = 'gdp' | 'gdpGrowth' | 'gdpPerCapita' | 'inflation' | 'unemployment' | 'population' | 'currentAccount' | 'exportsGdp' | 'govDebt' | 'primaryBalance' | 'structuralBalance' | 'fdiInflow' | 'realEffectiveExchangeRate' | 'policyRate';
 /**
@@ -212,10 +221,11 @@ export interface IMFEconomicContext {
 /**
  * Canonical IMF "virtual tool" names used by the native TypeScript
  * client. These identifiers are preserved from the earlier MCP-backed
- * iteration so the content-validator fingerprint list and workflow
- * probes (`scripts/imf-mcp-probe.sh`) remain stable across the
- * transport swap. Each name maps to a semantic method on
- * {@link IMFMCPClient} rather than to a remote MCP tool call.
+ * iteration so the Stage-C editorial fingerprint list (see
+ * `analysis/imf/indicator-catalog.md §6`) and the workflow probe
+ * (`scripts/imf-mcp-probe.sh`) remain stable across the transport
+ * swap. Each name maps to a semantic method on {@link IMFMCPClient}
+ * rather than to a remote MCP tool call.
  */
 export type IMFMCPToolName = 'imf-list-databases' | 'imf-search-databases' | 'imf-get-parameter-defs' | 'imf-get-parameter-codes' | 'imf-fetch-data';
 //# sourceMappingURL=imf.d.ts.map