` renders **real** Chart.js canvases backed by live IMF / SCB values (with World Bank residue for non-economic context only). 2. Every article carries a 2–4 sentence **AI commentary** paragraph that cites concrete numbers. 3. A deterministic quality gate fails the PR if any of **data**, **chart**, or **commentary** is missing. --- ## Artefact: `economic-data.json` Every agentic workflow MUST write: ``` analysis/daily/YYYY-MM-DD/{analysisSubfolder}/economic-data.json ``` where `{analysisSubfolder}` maps from the kebab article-type slug via `scripts/analysis-references.ts` → `ARTICLE_TYPE_TO_ANALYSIS_SUBFOLDER`: | Article type slug | analysisSubfolder | |-----------------------|---------------------| | `committee-reports` | `committeeReports` | | `propositions` | `propositions` | | `motions` | `motions` | | `interpellations` | `interpellations` | | `evening-analysis` | `evening-analysis` | | `realtime-monitor` | `realtime-monitor` | | `breaking` | `breaking` | | `week-ahead` | `week-ahead` | | `month-ahead` | `month-ahead` | | `weekly-review` | `weekly-review` | | `monthly-review` | `monthly-review` | | `deep-inspection` | `deep-inspection` | **Schema**: `analysis/schemas/economic-data.schema.json`. **Shape** (v2.0 — additive over v1): ```jsonc { "version": "2.0", "articleType": "committee-reports", "date": "2026-04-20", "policyDomains": ["fiscal policy", "labor market"], "dataPoints": [ // IMF (WEO projections permitted for look-ahead article types) { "countryCode": "SWE", "countryName": "Sweden", "indicatorId": "NGDP_RPCH", "date": "2025", "value": 1.9, "provider": "imf", "projection": false }, { "countryCode": "SWE", "countryName": "Sweden", "indicatorId": "GGXWDG_NGDP", "date": "2027", "value": 32.4, "provider": "imf", "projection": true, "projectionVintage": "WEO-2026-04" }, // World Bank (governance / environment / social residue) { "countryCode": "SWE", "countryName": "Sweden", "indicatorId": "CC.EST", "date": "2023", "value": 2.1, "provider": "worldBank", "projection": false }, // SCB (Swedish primary source) { "countryCode": "SWE", "countryName": "Sweden", "indicatorId": "TAB1291", "date": "2025-02", "value": 405.2, "provider": "scb", "projection": false } ], "commentary": "IMF projects Sweden's general government gross debt at 32.4 % of GDP in 2027 (WEO Apr-2026, GGXWDG_NGDP), giving SkU-2025/26:12 fiscal headroom absent in 2022 (38 %). Control-of-corruption remains strong at 2.1 (WGI 2023).", "source": { "worldBank": ["CC.EST"], "scb": ["TAB1291"], "imf": ["WEO:NGDP_RPCH", "WEO:GGXWDG_NGDP"] } } ``` **Hard rules**: - `dataPoints` MUST be non-empty unless `skip: true` (and the article type is on the allow-list). - `commentary` MUST cite 2–3 concrete numeric values that appear in `dataPoints` (the validator enforces only a word count; the human review + multi-dim quality score enforces the citation). - At least one of `source.imf` / `source.scb` MUST be non-empty for every economic-context article. `source.worldBank` is the appropriate field for governance / environment / social / defence-historicals / crime data points. - Every `dataPoint` whose policy domain is in `{macro, fiscal, monetary, external, trade, inflation, commodity, exchangeRate}` carries `provider: "imf"` (or `provider: "scb"` for Swedish-specific ground truth). The validator confirms each row in those domains uses an IMF or SCB provider. - When any `dataPoint.projection === true`, it MUST carry a `projectionVintage` tag (e.g. `"WEO-2026-04"`, `"FM-2026-04"`). Projection values MAY only be cited in article commentary for the **look-ahead allow-list**: `week-ahead`, `month-ahead`, `weekly-review`, `monthly-review`. Daily types (`committee-reports`, `propositions`, `motions`, `interpellations`, `evening-analysis`) MAY include projection values in `dataPoints` for context but MUST NOT quote them as definitive forward-looking claims in commentary. - File MUST validate against `analysis/schemas/economic-data.schema.json`. --- ## MCP tool contract Step 2.6 of every `news-*.md` workflow MUST perform these MCP calls **before** writing `economic-data.json`. Provider precedence is: 1. **IMF** — every economic claim. Macro (GDP, inflation, unemployment), fiscal (debt, deficit, revenue, expenditure, COFOG functional decomposition), monetary, external sector, trade, commodity, FX. WEO projections extend to T+5 and MUST be used for look-ahead article types. 2. **SCB** — Swedish-specific ground truth that IMF does not carry (household consumption patterns, regional data, high-frequency monthly tables, AKU labour-force survey). 3. **World Bank** — non-economic residue **only**: governance (WGI: CC.EST, RL.EST, VA.EST, GE.EST, RQ.EST, PV.EST), environment (CO2, forest area, renewables, water), defence historicals (`MS.MIL.XPND.GD.ZS`, `MS.MIL.TOTL.P1`), and long-horizon social/education indicators IMF does not cover. ### 1. Map documents → policy domains → indicators ``` view analysis/economic-indicators-inventory.json ``` Each indicator entry carries a `provider` field (`imf` | `worldBank` | `scb`), plus IMF-specific fields (`imfDatabase`, `imfIndicatorCode`, `imfDimensionFilters`, `projectionHorizon`) when the primary provider is IMF. Select every indicator whose `committees` / `policyAreas` match the day's source documents. The non-economic residue inventory at `analysis/worldbank/indicators-inventory.json` covers governance / environment / social / defence historicals only. ### 2. Query IMF (Sweden + Nordic peers) — PRIMARY for macro/fiscal/monetary Via the repository's pure-TypeScript client `scripts/imf-client.ts`, exposed to agentic workflows through the thin `scripts/imf-fetch.ts` CLI. No Python MCP / `uvx` runtime is involved; all IMF traffic goes directly to `data.imf.org`, `api.imf.org`, and `www.imf.org` on the firewall allowlist. ```bash # 1. WEO time series for one country (default 10 years; --persist caches # the JSON under analysis/data/imf/{indicator}/{country}.json with # projectionVintage provenance). tsx scripts/imf-fetch.ts weo \ --country SWE --indicator NGDP_RPCH --years 15 --persist # 2. Compare the latest WEO value across Sweden + Nordic peers in one call. tsx scripts/imf-fetch.ts compare \ --indicator GGXWDG_NGDP \ --countries SWE,DNK,NOR,FIN,DEU --persist # 3. SDMX 3.0 passthrough for IFS / BOP / FM / GFS / DOTS. tsx scripts/imf-fetch.ts sdmx \ --path "/data/IMF.STA,CPI,5.0.0/SWE.CPI._T.IX.M?startPeriod=2024-01" \ --indicator _T.IX --country SWE --persist # 4. Inspect the built-in WEO + FM indicator catalog (no network call). tsx scripts/imf-fetch.ts list-indicators ``` **Rate-limit discipline** (IMF ~10 req / 5 s): prefer the `compare` subcommand (one batched call across countries), insert a 1 s sleep between separate `imf-fetch.ts` invocations, and rely on the client's built-in 3× retry with exponential back-off (1 s → 2 s → 4 s) for 429 / 5xx. Pre-warm 1 request at workflow start. ### 3. Query World Bank — governance, environment, social residue, defence historicals, crime/justice > **Scope** — World Bank supplies non-economic context: governance (WGI `source=75`), environment (CO₂, forest area, renewables, water), social/education residue (population, life expectancy, school enrolment), defence historicals (`MS.MIL.XPND.GD.ZS`, `MS.MIL.TOTL.P1`), and crime/justice (`VC.IHR.PSRC.P5`). Use `scripts/imf-fetch.ts` for every economic claim; use `scripts/scb-client.ts` for Swedish-specific ground truth. ``` # Governance / WGI (source=75) get-economic-data(countryCode="SE", indicator="CC.EST", years=5) # Control of Corruption get-economic-data(countryCode="SE", indicator="RL.EST", years=5) # Rule of Law get-economic-data(countryCode="SE", indicator="VA.EST", years=5) # Voice & Accountability # Environment get-economic-data(countryCode="SE", indicator="EN.ATM.CO2E.PC", years=10) # CO2 emissions per capita get-economic-data(countryCode="SE", indicator="EG.FEC.RNEW.ZS", years=10) # Renewables share get-economic-data(countryCode="SE", indicator="AG.LND.FRST.ZS", years=10) # Forest cover # Defence historicals (use instead of COFOG 02 when 5+ year trend needed) get-economic-data(countryCode="SE", indicator="MS.MIL.XPND.GD.ZS", years=15) # Education participation get-education-data(countryCode="SE", indicator="SCHOOL_ENROLLMENT", years=10) # Social / demographic context (WB preferred for long-horizon > 15 years) get-social-data(countryCode="SE", indicator="POPULATION", years=10) ``` ### 4. Query SCB (Statistics Sweden) ``` # CRITICAL: language MUST be "sv" or "en". NEVER "no" — SCB returns # HTTP 400 "Unsupported language" for "no". search_tables(query="", language="en") # Then fetch specific tables from the committee→TAB mapping # (scripts/scb-context.ts): # FiU → TAB1291 AU → TAB5765 JuU → TAB1172 # MJU → TAB5404 SoU → health tables UbU → education tables query_table(table_id="", value_codes={"Tid": "top(10)", ...}) ``` ### Committee → primary provider & indicator matrix (v2) | Committee | Primary provider(s) | MUST query (provider:code) | |-----------|---------------------|-----------------------------------------------------------------------------------------------| | FiU | IMF + WB | `imf:WEO:NGDP_RPCH`, `imf:WEO:PCPIPCH`, `imf:WEO:NGDPDPC`, `imf:WEO:GGXWDG_NGDP` | | SkU | IMF + WB | `imf:WEO:GGR_NGDP`, `imf:WEO:GGX_NGDP`, `imf:WEO:GGXCNL_NGDP`, `imf:FM:GGXONLB_NGDP` | | AU | IMF + WB | `imf:WEO:LUR`, `wb:SL.UEM.1524.ZS`, `wb:SL.TLF.CACT.ZS` | | NU / UU | IMF | `imf:WEO:BCA_NGDPD`, `imf:WEO:TX_RPCH` | | SoU | WB | `wb:SH.XPD.CHEX.GD.ZS`, `wb:SH.MED.PHYS.ZS`, `wb:SH.MED.BEDS.ZS`, `wb:SP.DYN.LE00.IN` | | UbU | WB | `wb:SE.XPD.TOTL.GD.ZS`, `wb:SE.ADT.LITR.ZS`, `wb:SE.PRM.ENRR` | | FöU | WB | `wb:MS.MIL.XPND.GD.ZS`, `wb:MS.MIL.XPND.CD` | | MJU | WB | `wb:EN.ATM.CO2E.PC`, `wb:EG.FEC.RNEW.ZS`, `wb:AG.LND.FRST.ZS` | | KU | WB | `wb:CC.EST`, `wb:RL.EST`, `wb:VA.EST` (WGI, source=75) | | JuU | WB | `wb:VC.IHR.PSRC.P5`, `wb:CC.EST` | ### 5. (High-level reviews only) D3 coalition-flow dataset Article types: `week-ahead`, `month-ahead`, `weekly-review`, `monthly-review`. In addition to `economic-data.json`, produce a `coalition-flow.json` with `nodes[]`/`flows[]` consumed by `generateSankeySection`. Minimum 2 non-trivial flows. --- ## Coverage matrix (enforced by the validator) `scripts/validate-economic-context.ts` reads this matrix. Changes here must be mirrored in the validator's `COVERAGE_MATRIX`. | Article type | Min Chart.js canvases | Min commentary words | D3 required | May skip | |--------------------|-----------------------|----------------------|-------------|----------| | committee-reports | 2 | 60 | No | No | | propositions | 2 | 60 | No | No | | motions | 1 | 40 | No | No | | interpellations | 1 | 40 | No | No | | evening-analysis | 1 | 40 | No | No | | realtime-monitor | 1 | 30 | No | Yes | | breaking | 1 | 30 | No | Yes | | week-ahead | 2 | 80 | Optional | No | | month-ahead | 3 | 100 | Recommended | No | | weekly-review | 3 | 150 | Required | No | | monthly-review | 4 | 200 | Required | No | | deep-inspection | 1 | 40 | No | Yes | | article-generator | 1 | 40 | No | Yes | "May skip" = `economic-data.json` may be written with `skip: true` and `skipReason` when the article is a pure-process piece (e.g. "Riksdag voting schedule changed"). The validator enforces the allow-list; workflows MUST NOT use skip as a shortcut to avoid fetching data. --- ## Client-side rendering — what the agent does NOT need to write The aggregate+render pipeline (`scripts/aggregate-analysis.ts` → `scripts/render-articles.ts` → `scripts/render-lib/`) handles all client-side wiring. Whenever the rendered HTML contains a Mermaid fenced block or a Chart.js / D3 canvas, the renderer auto-injects the appropriate runtime via `scripts/render-lib/chrome.ts`: - `` — emitted whenever any `

` survives `rehype-sanitize`
- `` — Chart.js 4
- `` — scans `[data-chart-config]` canvases on DOMContentLoaded and calls `new Chart(ctx, cfg)` for each
- `` — added when any section uses `data-d3-sankey=`

Consequences for the AI agent writing analysis artifacts:

- **DO** drop ```mermaid fenced blocks straight into your `.md` artifacts under `analysis/daily/$DATE/$SUB/`. The renderer's allow-list lets `` pass through `rehype-sanitize`; `mermaid-init.mjs` handles render in the browser.
- **DO** include `