FUTURE_DATA_MODEL.md

May 31, 2026 · View on GitHub

📊 Riksdagsmonitor — Future Data Architecture Model

🔮 Three-Horizon Evolution: Static JSON/CSV → Richer Static Pre-Compute → AWS Serverless Intelligence
🎯 Neptune Graph · Aurora Serverless v2 · OpenSearch Vector · Bedrock Knowledge Bases · API Gateway · Cognito

🏆 Evidence & Compliance Badges

📋 Document Owner: CEO | 📄 Version: 3.0 | 📅 Last Updated: 2026-05-31 (UTC)
🔄 Review Cycle: Annual | ⏰ Next Review: 2027-05-31
🏢 Owner: Hack23 AB (Org.nr 5595347807) | 🏷️ Classification: Public

📚 Architecture Documentation Map

Document	Type	Description
Architecture	🏛️ Current	C4 model showing system structure
Data Model	📊 Current	Data entities and relationships
Flowcharts	🔄 Current	Process flows and pipelines
State Diagrams	🔄 Current	System state transitions
Mindmap	🗺️ Current	System conceptual map
SWOT	💼 Current	Strategic analysis
Future Architecture	🏗️ Future	System evolution roadmap
Future Data Model	📊 Future	Enhanced data architecture (this doc)
Future Flowcharts	🔄 Future	Advanced process flows
Future State Diagrams	🔄 Future	Advanced state management
Future Mindmap	🗺️ Future	Future capability map
Future SWOT	💼 Future	Strategic outlook
Security Architecture	🛡️ Security	Defense-in-depth controls
Future Security Architecture	🛡️ Future	Security roadmap
Threat Model	🎯 Security	STRIDE analysis

🎯 Executive Summary

Riksdagsmonitor today is a static, evidence-first political intelligence platform: pre-computed JSON/CSV products derived from Swedish parliamentary open data, served as static HTML/CSS in 14 languages with no client-side framework. The current data model (DATA_MODEL.md v1.3) already encodes 2,494 politicians (349 active MPs), 3,529,786 voting records, 109,259 documents, 8 active parties (+ 32 historical = 40 total), 15 committees, 20 governments, and 15 CIA analytical subsystems that compile into 19 user-facing intelligence products.

This Future Data Model defines three explicit horizons that preserve the platform's static-first, public-data-only, neutral mission while progressively deepening analytical richness:

Horizon	Window	Data Architecture	Mission Continuity
Horizon 1 — Baseline	v1.x (now)	Static pre-computed JSON/CSV; CIA subsystems; npm typed surface; IMF/SCB/World Bank caches	Evidence-first, neutral, public-source-only
Horizon 2 — Richer Static	v2.0 (2026–2027)	Still 100% static: build-time party-cohesion matrices, coalition/bloc graphs, party-vs-party datasets, OSINT structures (network edges, temporal series, anomaly scores, source-graded INTOP metadata)	Same hosting, deeper pre-computation
Horizon 3 — Serverless	v3.0+ (2028–2037)	AWS serverless data tier (Neptune Serverless, Aurora Serverless v2, DynamoDB, OpenSearch Serverless, Timestream, Bedrock Knowledge Bases) exposed via API Gateway (GraphQL/REST) + Amazon Cognito	Interactive queries layered atop the same primary-source ground truth

All future metrics in this document are TARGETS, not achieved measurements. Horizons 2 and 3 are forward-looking design intent. The platform processes only public data, treats political opinions as GDPR Article 9 special-category data (lawful bases 9(2)(e) manifestly made public; 9(2)(g) substantial public interest), maintains strict party neutrality, and contains no surveillance capability of private individuals.

graph LR
    subgraph H1["Horizon 1 — v1.x Baseline (now)"]
        A1["Static JSON/CSV<br/>CIA subsystems"]
        A2["npm typed surface<br/>riksdagsmonitor"]
        A3["IMF / SCB / World Bank caches"]
    end
    subgraph H2["Horizon 2 — v2.0 Richer Static (2026-2027)"]
        B1["Party cohesion matrices<br/>Coalition / bloc graphs"]
        B2["OSINT structures<br/>edges · series · anomaly · INTOP"]
    end
    subgraph H3["Horizon 3 — v3.0+ Serverless (2028-2037)"]
        C1["Neptune · Aurora v2 · DynamoDB"]
        C2["OpenSearch · Timestream · Bedrock KB"]
        C3["API Gateway + Cognito"]
    end
    H1 --> H2 --> H3
    style A1 fill:#bbdefb,stroke:#1565c0,color:#000000
    style A2 fill:#bbdefb,stroke:#1565c0,color:#000000
    style A3 fill:#bbdefb,stroke:#1565c0,color:#000000
    style B1 fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style B2 fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style C1 fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style C2 fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style C3 fill:#e1bee7,stroke:#6a1b9a,color:#000000

📋 Table of Contents

Three-Horizon Data Evolution Overview
Horizon 1 — v1.x Baseline Data Model
Horizon 2 — v2.0 Static Intelligence Data Models (2026–2027)
Horizon 3 — v3.0+ AWS Serverless Data Tier (2028–2037)
Source Ingestion & Integration
GraphQL API Schema
Data Model Diagrams
Implementation Roadmap
Technology Stack Evolution & Cost Projections
ISMS Compliance & Data Governance
IMF Data Domain — Filesystem Cache → Aurora Schema
AI/LLM Data Architecture Evolution (2026–2037)
Related Documentation

1. Three-Horizon Data Evolution Overview

The platform evolves along three horizons without ever abandoning its static-first, evidence-first foundation. Each horizon is additive: the serverless tier (Horizon 3) is layered on top of — never instead of — the static pre-computed ground truth that Horizons 1 and 2 establish.

1.1 Horizon Comparison

Dimension	H1 — Baseline (v1.x)	H2 — Richer Static (v2.0, 2026–2027)	H3 — Serverless (v3.0+, 2028–2037)
Data form	Pre-computed JSON/CSV	Pre-computed JSON (party matrices, graphs, OSINT)	Live queryable stores + static fallback
Compute	GitHub Actions build-time	GitHub Actions build-time (heavier)	AWS Lambda + Step Functions on demand
Storage	S3 + CloudFront, GitHub Pages DR	Same	Neptune, Aurora v2, DynamoDB, OpenSearch, Timestream
Access	Static HTTP fetch	Static HTTP fetch	API Gateway (GraphQL/REST) + Cognito
Query model	File-addressed	File-addressed	Graph traversal, SQL, vector, time-series
AI	Newsroom (Opus 4.x) authoring	+ RAG-ready embeddings (build-time)	Bedrock Knowledge Bases RAG
Cost	~CDN + Actions minutes	Marginally higher build cost	Pay-per-use serverless (target $X/mo)
Neutrality / GDPR	Art. 9 public-data only	Same	Same, with Cognito audit trail

1.2 Data Volume Projections (Targets)

Metric	H1 (now)	H2 target (2027)	H3 target (2030)
Politicians (all-time)	2,494	2,494+	3,000+
Active MPs tracked	349	349	349
Voting records	3,529,786	~3.7M	~4.5M
Documents indexed	109,259	~130,000	~200,000
Chamber speeches (anföranden)	indexed	+ entity-linked	+ vector-embedded
Pre-computed party datasets	CIA subsystems	+ cohesion/coalition/bloc	served via API
News corpus (HTML)	~3,953	~6,000	~10,000
Languages	14	14	14

Projections are planning targets to size infrastructure, not commitments or forecasts of political outcomes.

2. Horizon 1 — v1.x Baseline Data Model

Horizon 1 is the live, shipping data model defined authoritatively in DATA_MODEL.md v1.3 (2026-05-06). This section summarizes it as the anchor that Horizons 2 and 3 extend.

2.1 Core Entities

erDiagram
    POLITICIAN ||--o{ ASSIGNMENT : holds
    POLITICIAN ||--o{ VOTE : casts
    POLITICIAN ||--o{ SPEECH : delivers
    PARTY ||--o{ POLITICIAN : includes
    PARTY ||--o{ VOTE : aggregates
    COMMITTEE ||--o{ ASSIGNMENT : staffs
    COMMITTEE ||--o{ DOCUMENT : produces
    DOCUMENT ||--o{ VOTE : triggers
    MINISTRY ||--o{ ROLE : defines
    GOVERNMENT ||--o{ MINISTRY : organizes
    GOVERNMENT ||--o{ ROLE : appoints

    POLITICIAN {
        string intressent_id PK
        string namn
        string parti FK
        string valkrets
        string status
        date fodd
    }
    PARTY {
        string kod PK
        string namn
        boolean active
        int seats
    }
    COMMITTEE {
        string kod PK
        string namn
        string organ
    }
    DOCUMENT {
        string dok_id PK
        string doktyp
        string titel
        string rm
        date publicerad
    }
    VOTE {
        string votering_id PK
        string dok_id FK
        string intressent_id FK
        string rost
        string punkt
    }
    MINISTRY {
        string kod PK
        string namn
        string government FK
    }
    GOVERNMENT {
        string id PK
        string namn
        date from
        date tom
    }
    ROLE {
        string id PK
        string ministry FK
        string intressent_id FK
        string titel
    }
    SPEECH {
        string anforande_id PK
        string intressent_id FK
        string dok_id FK
        string rm
    }

Baseline counts (from DATA_MODEL.md): 2,494 politicians; 349 active MPs; 8 active parties (S, M, SD, C, V, MP, KD, L) + 32 historical = 40 total; 15 committees; 109,259 documents; 3,529,786 voting records; 20 governments (76 roles, 500 role members); chamber speeches (anföranden) indexed at /anforande/. Historical coverage: 1971–2026.

2.2 CIA Subsystems → User-Facing Products

The 15 CIA data subsystems (anomaly, coalition, committee, distribution, election, election-cycle, ministry, parties, party, percentile, politician, pre-election, risk, seasonal, voting) compile into 19 user-facing intelligence products: 4 dashboards + 10 Top-10 rankings + 5 advanced analytics.

graph TD
    subgraph SUB["15 CIA Subsystems (build-time)"]
        S1["anomaly · coalition · committee"]
        S2["distribution · election · election-cycle"]
        S3["ministry · parties · party · percentile"]
        S4["politician · pre-election · risk"]
        S5["seasonal · voting"]
    end
    subgraph PROD["19 User-Facing Products"]
        P1["4 Dashboards"]
        P2["10 Top-10 Rankings"]
        P3["5 Advanced Analytics"]
    end
    SUB --> PROD
    style S1 fill:#bbdefb,stroke:#1565c0,color:#000000
    style S2 fill:#bbdefb,stroke:#1565c0,color:#000000
    style S3 fill:#bbdefb,stroke:#1565c0,color:#000000
    style S4 fill:#bbdefb,stroke:#1565c0,color:#000000
    style S5 fill:#bbdefb,stroke:#1565c0,color:#000000
    style P1 fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style P2 fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style P3 fill:#c8e6c9,stroke:#2e7d32,color:#000000

2.3 npm Typed Surface

Package riksdagsmonitor v0.9.40 ("type":"module", SLSA provenance attested) exposes typed subpaths generated by scripts/generate-types-from-cia-schemas.ts:

Subpath	Contents
`./`	Root types
`./shared`, `./shared/*`	Shared domain types
`./cia/*`	CIA subsystem types
`./dashboards/*`	Dashboard data shapes
`./ui/*`	UI component data contracts

2.4 Economic & Statistical Caches

Source	Role	Cache location
IMF (primary economic)	WEO/FM/IFS/BOP/DOTS/GFS_COFOG/PCPS/ER/MFS_IR/MFS_PR; T+5 projections	`analysis/data/imf/{indicator}/{country}.json` + `.meta.json`
SCB (Swedish ground truth)	PxWeb v2 national statistics	build-time fetch
World Bank (non-economic only)	Governance (WGI), environment, social residue	build-time fetch

Per ADR 0001, the IMF client is a pure-TypeScript client (scripts/imf-client.ts), not an MCP server. Economic World Bank codes are deprecated in favour of IMF.

2.5 Analysis Artifacts & News Corpus

Analysis artifact families: Family A core synthesis = 9 artifacts; Family B (2), Family C (5), Family D (7), Family E (per-document). Gate: scripts/agentic/analysis-gate.ts.
Newsroom: 14 gh-aw agentic workflows, Claude Opus 4.8 (Sonnet 4.6 for translation), 14 languages, zero human editors. News corpus ~3,953 HTML files under news/.
Political-intelligence catalog: pre-computed cross-entity intelligence products surfaced in political-intelligence.html (+ 13 localized variants).

3. Horizon 2 — v2.0 Static Intelligence Data Models (2026–2027)

Horizon 2 remains 100% static. No servers, no databases, no auth tier. It deepens pre-computation: the GitHub Actions build emits richer party-focused and OSINT-structured JSON datasets that static HTML pages consume directly. This horizon proves analytical value before any serverless investment.

3.1 Design Principles

Static-first invariant — every H2 dataset is a build-time artifact addressable by URL; no runtime compute.
Party-centric depth — cohesion, coalition, and bloc analytics become first-class pre-computed products.
OSINT rigor — network/temporal/anomaly structures carry source-grading and INTOP metadata so every edge and score is traceable to a primary source.
Forward-compatible shapes — H2 JSON schemas are designed to map cleanly onto H3 graph/SQL/vector stores.

3.2 Party Cohesion Matrices

A pre-computed matrix scoring intra-party voting discipline per voting period, per committee, and per policy domain.

{
  "schema": "party-cohesion-matrix@2.0",
  "generated_at": "2027-01-15T00:00:00Z",
  "rm": "2026/27",
  "source_grading": { "system": "Admiralty", "reliability": "A", "credibility": "1" },
  "parties": ["S", "M", "SD", "C", "V", "MP", "KD", "L"],
  "matrix": [
    {
      "party": "S",
      "cohesion_index": 0.97,
      "rebel_votes": 14,
      "total_votes": 5120,
      "by_committee": { "FiU": 0.99, "SoU": 0.95, "UU": 0.98 },
      "evidence_dok_ids": ["H801FiU1", "H801SoU12"]
    }
  ]
}

cohesion_index ∈ [0,1] = share of party MPs voting with the party majority.
Every party row carries evidence_dok_ids so the static page can deep-link to primary documents.
source_grading applies the Admiralty/NATO reliability–credibility scale at dataset level.

3.3 Coalition & Bloc Graphs

A pre-computed graph of inter-party alignment derived from co-voting frequency, expressed as nodes (parties) and weighted edges (alignment strength).

{
  "schema": "coalition-bloc-graph@2.0",
  "rm": "2026/27",
  "nodes": [
    { "id": "S", "bloc": "left", "seats": 107 },
    { "id": "M", "bloc": "right", "seats": 68 }
  ],
  "edges": [
    {
      "source": "S",
      "target": "V",
      "alignment": 0.88,
      "shared_yes_votes": 4210,
      "divergent_votes": 560,
      "intop_class": "OPEN-SOURCE",
      "evidence_dok_ids": ["H801AU3"]
    }
  ]
}

graph LR
    S((S)) ---|0.88| V((V))
    S ---|0.79| MP((MP))
    M((M)) ---|0.91| KD((KD))
    M ---|0.84| L((L))
    M ---|0.62| SD((SD))
    C((C)) ---|0.55| M
    style S fill:#ef9a9a,stroke:#b71c1c,color:#000000
    style V fill:#ef9a9a,stroke:#b71c1c,color:#000000
    style MP fill:#a5d6a7,stroke:#1b5e20,color:#000000
    style M fill:#90caf9,stroke:#0d47a1,color:#000000
    style KD fill:#90caf9,stroke:#0d47a1,color:#000000
    style L fill:#90caf9,stroke:#0d47a1,color:#000000
    style SD fill:#fff59d,stroke:#f57f17,color:#000000
    style C fill:#c8e6c9,stroke:#2e7d32,color:#000000

Bloc labels reflect published, self-declared parliamentary alignments and co-voting evidence — never editorial judgement. Edge weights are reproducible from the public voting record.

3.4 Party-vs-Party Comparison Datasets

Symmetric pairwise comparison datasets enabling static comparison pages (e.g., "S vs M on welfare").

{
  "schema": "party-vs-party@2.0",
  "pair": ["S", "M"],
  "domains": {
    "welfare": { "agreement": 0.41, "votes": 612, "evidence_dok_ids": ["H801SoU5"] },
    "defence": { "agreement": 0.83, "votes": 188, "evidence_dok_ids": ["H801FoU2"] },
    "economy": { "agreement": 0.37, "votes": 540, "evidence_dok_ids": ["H801FiU1"] }
  }
}

3.5 OSINT Data Structures

Horizon 2 formalizes four OSINT structure families as build-time JSON, each carrying provenance metadata so the static UI can show how we know.

3.5.1 Network Edges

{
  "schema": "osint-network-edges@2.0",
  "edge_type": "co-sponsorship",
  "edges": [
    {
      "from": "intressent_0123",
      "to": "intressent_0456",
      "weight": 23,
      "rm": "2026/27",
      "evidence_dok_ids": ["H802Mot123"],
      "source_grading": { "reliability": "A", "credibility": "1" }
    }
  ]
}

3.5.2 Temporal Series

{
  "schema": "osint-temporal-series@2.0",
  "metric": "speech_activity",
  "entity": "intressent_0123",
  "interval": "monthly",
  "points": [
    { "t": "2026-09", "v": 12 },
    { "t": "2026-10", "v": 19 }
  ]
}

3.5.3 Anomaly Scores

{
  "schema": "osint-anomaly-scores@2.0",
  "method": "seasonal-decomposition + z-score",
  "scores": [
    {
      "entity": "intressent_0123",
      "metric": "vote_attendance",
      "z": -3.1,
      "flag": "low-attendance-outlier",
      "evidence_dok_ids": ["H802Vot44"],
      "explanation": "Attendance 3.1σ below seasonal baseline; documented leave of absence."
    }
  ]
}

Anomaly flags are descriptive statistical signals on public records, always paired with a neutral, evidence-linked explanation. They are never accusatory and never applied to private individuals.

3.5.4 Source-Grading & INTOP Metadata

Every H2 dataset embeds a source_grading$ \text{block} (\text{Admiralty} \text{reliability} \text{A}–\text{F} \times \text{credibility} 1–6) \text{and} \text{an} $intop_class field (e.g., OPEN-SOURCE, OFFICIAL-PUBLIC) so downstream consumers — and H3's RAG pipeline — inherit provenance.

3.5.5 Evidence & Provenance Invariant

Every Horizon 2 dataset is rejected by scripts/agentic/analysis-gate.ts unless it satisfies the evidence-first invariant:

Requirement	Enforcement
`evidence_dok_ids` non-empty on every scored row/edge	Gate hard-fail
`source_grading$ (\text{Admiralty} \text{A}–\text{F} \times 1–6) \text{present} \text{at} \text{dataset} \text{level}	\text{Gate} \text{hard}-\text{fail}
$intop_class` present on every relational edge	Gate hard-fail
Reproducible from public voting/document record	CI re-computation diff
Neutral, non-accusatory language on anomaly explanations	Editorial lint

This guarantees that the H3 RAG pipeline (§4.7) inherits provenance: an embedding can always be traced back to a dok_id.

3.6 Government & Minister Scorecard Datasets

Horizon 2 adds pre-computed accountability datasets covering the executive branch — the 20 governments (76 roles, 500 role members) in the baseline — without any editorial scoring of "good" or "bad". Metrics are purely descriptive counts on the public record.

{
  "schema": "minister-scorecard@2.0",
  "rm": "2026/27",
  "ministry": "Finansdepartementet",
  "role_holder": "intressent_0789",
  "interpellations_received": 41,
  "written_questions_answered": 118,
  "propositions_introduced": 9,
  "evidence_dok_ids": ["H801FiU1", "H801Prop44"],
  "source_grading": { "reliability": "A", "credibility": "1" }
}

3.7 Rebellion & Defection Timelines

A temporal OSINT dataset tracking individual MPs who voted against their party majority, with neutral context and primary-source links.

{
  "schema": "rebellion-timeline@2.0",
  "entity": "intressent_0123",
  "party": "S",
  "events": [
    {
      "t": "2026-11-18",
      "votering_id": "H801Vot88",
      "party_majority": "Ja",
      "individual_vote": "Nej",
      "dok_id": "H801SoU12",
      "context": "Voted against party line on welfare amendment; public record only."
    }
  ]
}

3.8 Build-Time Generation Pipeline

sequenceDiagram
    participant API as Riksdag/Regering APIs
    participant GH as GitHub Actions
    participant GEN as Dataset Generators
    participant GATE as analysis-gate.ts
    participant S3 as S3 + CloudFront
    API->>GH: Fetch voteringar / dokument / anföranden
    GH->>GEN: Compute cohesion / coalition / OSINT
    GEN->>GEN: Attach source_grading + INTOP + dok_ids
    GEN->>GATE: Submit artifacts
    GATE-->>GEN: Pass (provenance complete) / Reject
    GEN->>S3: Publish static JSON datasets
    S3->>S3: GitHub Pages DR mirror

3.9 Horizon 2 ERD

erDiagram
    PARTY ||--o{ COHESION_ROW : scored_in
    PARTY ||--o{ BLOC_EDGE : participates
    PARTY ||--o{ PAIR_COMPARE : compared
    POLITICIAN ||--o{ NETWORK_EDGE : connects
    POLITICIAN ||--o{ TEMPORAL_POINT : measured
    POLITICIAN ||--o{ ANOMALY_SCORE : flagged
    SOURCE_GRADING ||--o{ COHESION_ROW : grades
    SOURCE_GRADING ||--o{ BLOC_EDGE : grades
    SOURCE_GRADING ||--o{ NETWORK_EDGE : grades

    COHESION_ROW {
        string party FK
        float cohesion_index
        int rebel_votes
        json evidence_dok_ids
    }
    BLOC_EDGE {
        string source FK
        string target FK
        float alignment
        string intop_class
    }
    PAIR_COMPARE {
        string party_a FK
        string party_b FK
        json domains
    }
    NETWORK_EDGE {
        string from FK
        string to FK
        int weight
        string edge_type
    }
    TEMPORAL_POINT {
        string entity FK
        string t
        float v
    }
    ANOMALY_SCORE {
        string entity FK
        float z
        string flag
    }
    SOURCE_GRADING {
        string reliability
        string credibility
        string intop_class
    }

4. Horizon 3 — v3.0+ AWS Serverless Data Tier (2028–2037)

Horizon 3 layers an interactive, queryable serverless tier atop the static ground truth. The static H1/H2 artifacts remain the system of record and disaster-recovery fallback; the serverless stores are derived, query-optimized projections hydrated from those artifacts. Access is mediated by Amazon API Gateway (GraphQL & REST) with Amazon Cognito for authentication and rate-limiting.

4.1 Serverless Store Selection

Store	Purpose	Data shape
Amazon Neptune Serverless	Political relationship graph (co-voting, co-sponsorship, coalition)	Property graph (Gremlin)
Amazon Aurora Serverless v2 (PostgreSQL)	Relational facts (politicians, parties, documents, votes, committees, IMF cache)	SQL tables
Amazon DynamoDB	High-velocity key lookups (dashboard state, rankings, session)	Key-value / document
Amazon OpenSearch Serverless	Full-text + vector search over documents & speeches	Inverted index + k-NN vectors
Amazon Timestream	Time-series metrics (attendance, activity, anomaly z-scores)	Time-series
Amazon Bedrock Knowledge Bases	RAG over corpus for the newsroom & citizen Q&A	Managed vector KB

4.2 Neptune Graph (Gremlin)

// Vertices
g.addV('politician').property('intressent_id','intressent_0123')
                    .property('namn','Example MP')
                    .property('parti','S')
                    .property('valkrets','Stockholm')

g.addV('party').property('kod','S').property('namn','Socialdemokraterna')

// Edges (derived from public voting record)
g.V().has('politician','intressent_id','intressent_0123').as('a')
 .V().has('politician','intressent_id','intressent_0456').as('b')
 .addE('co_voted').from('a').to('b')
   .property('weight',4210)
   .property('rm','2026/27')
   .property('evidence_dok_id','H801AU3')

4.3 Aurora Serverless v2 (PostgreSQL) Schema

CREATE TABLE politicians (
    intressent_id   TEXT PRIMARY KEY,
    namn            TEXT NOT NULL,
    parti           TEXT REFERENCES parties(kod),
    valkrets        TEXT,
    status          TEXT,
    fodd            DATE
);

CREATE TABLE parties (
    kod     TEXT PRIMARY KEY,
    namn    TEXT NOT NULL,
    active  BOOLEAN NOT NULL DEFAULT TRUE,
    seats   INTEGER
);

CREATE TABLE documents (
    dok_id      TEXT PRIMARY KEY,
    doktyp      TEXT,
    titel       TEXT,
    rm          TEXT,
    publicerad  DATE
);

CREATE TABLE votes (
    votering_id     TEXT PRIMARY KEY,
    dok_id          TEXT REFERENCES documents(dok_id),
    intressent_id   TEXT REFERENCES politicians(intressent_id),
    rost            TEXT CHECK (rost IN ('Ja','Nej','Avstår','Frånvarande')),
    punkt           TEXT
);

CREATE TABLE committees (
    kod     TEXT PRIMARY KEY,
    namn    TEXT NOT NULL,
    organ   TEXT
);

CREATE INDEX idx_votes_intressent ON votes(intressent_id);
CREATE INDEX idx_votes_dok ON votes(dok_id);
CREATE INDEX idx_docs_rm ON documents(rm);

4.4 DynamoDB Tables

{
  "TableName": "rm_dashboard_state",
  "KeySchema": [
    { "AttributeName": "pk", "KeyType": "HASH" },
    { "AttributeName": "sk", "KeyType": "RANGE" }
  ],
  "AttributeDefinitions": [
    { "AttributeName": "pk", "AttributeType": "S" },
    { "AttributeName": "sk", "AttributeType": "S" }
  ],
  "BillingMode": "PAY_PER_REQUEST"
}

Access patterns: pk=RANKING#top10-attendance, sk=RM#2026/27; pk=PARTY#S, sk=COHESION#2026/27.

4.5 OpenSearch Serverless Index

{
  "mappings": {
    "properties": {
      "dok_id": { "type": "keyword" },
      "titel": { "type": "text" },
      "body": { "type": "text" },
      "rm": { "type": "keyword" },
      "doktyp": { "type": "keyword" },
      "embedding": { "type": "knn_vector", "dimension": 1024 },
      "source_grading": { "type": "object" }
    }
  }
}

4.6 Timestream Schema

Dimension	Measure	Example
`intressent_id`, `metric`	`value` (double)	attendance per session
`party`, `metric`	`value` (double)	cohesion index over time
`entity`, `metric`	`z_score` (double)	anomaly signal over time

4.7 Bedrock Knowledge Bases RAG

// Build-time: embed corpus → Bedrock KB; runtime: retrieve + ground
import { BedrockAgentRuntimeClient, RetrieveAndGenerateCommand }
  from "@aws-sdk/client-bedrock-agent-runtime";

const client = new BedrockAgentRuntimeClient({ region: "eu-west-1" });

const response = await client.send(new RetrieveAndGenerateCommand({
  input: { text: "How did party S vote on the 2027 defence bill?" },
  retrieveAndGenerateConfiguration: {
    type: "KNOWLEDGE_BASE",
    knowledgeBaseConfiguration: {
      knowledgeBaseId: "rm-corpus-kb",
      modelArn: "arn:aws:bedrock:eu-west-1::foundation-model/anthropic.claude-opus",
      retrievalConfiguration: {
        vectorSearchConfiguration: { numberOfResults: 8 }
      }
    }
  }
}));
// Every generated answer MUST cite retrieved dok_ids — no ungrounded claims.

The RAG pipeline is retrieval-grounded only: generated answers must cite retrieved dok_id evidence. This enforces the evidence-first invariant at the AI layer and prevents hallucinated political claims.

4.8 Access Tier — API Gateway + Cognito

graph TD
    U["Citizen / Journalist / Researcher"] --> CF["CloudFront"]
    CF --> APIGW["API Gateway (GraphQL + REST)"]
    APIGW --> COG["Amazon Cognito<br/>auth · rate-limit · audit"]
    COG --> L["AWS Lambda Resolvers"]
    L --> NEP["Neptune Serverless"]
    L --> AUR["Aurora Serverless v2"]
    L --> DDB["DynamoDB"]
    L --> OS["OpenSearch Serverless"]
    L --> TS["Timestream"]
    L --> KB["Bedrock Knowledge Bases"]
    L -. fallback .-> S3["Static H1/H2 JSON (system of record)"]
    style APIGW fill:#ffe0b2,stroke:#e65100,color:#000000
    style COG fill:#ffcc80,stroke:#e65100,color:#000000
    style L fill:#fff9c4,stroke:#f9a825,color:#000000
    style NEP fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style AUR fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style DDB fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style OS fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style TS fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style KB fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style S3 fill:#bbdefb,stroke:#1565c0,color:#000000

API Gateway + Cognito provide authentication, throttling, and an audit trail — supporting GDPR accountability — while the public read corpus stays open. AppSync may serve as a managed-GraphQL resolver option behind API Gateway, but API Gateway + Lambda is the headline contract.

4.9 Store Sizing & Access Patterns (Targets)

Store	Target volume (2030)	Primary access pattern	Why this store
Neptune Serverless	~2.5M edges (co-voting, co-sponsorship)	Multi-hop traversal ("who co-votes with whom")	Graph queries are O(edges) not O(joins)
Aurora Serverless v2	~200K docs, ~4.5M votes	Relational filters, aggregates	ACID facts, SQL analysts
DynamoDB	~19 product partitions × RM	Single-digit-ms key lookup	Dashboard/ranking hot path
OpenSearch Serverless	~200K docs + speeches	Text + k-NN vector	Semantic + keyword search
Timestream	~10M points	Range scans, anomaly windows	Native time-series rollups
Bedrock KB	corpus embeddings	RAG retrieve-and-generate	Managed grounding

Each store is derived from the static system-of-record; loss of any serverless store degrades gracefully to static H1/H2 JSON, never to data loss.

4.10 Multi-Source Consistency Model

The serverless tier is eventually consistent with the static SoR. The hydration contract (§5) guarantees: (a) the static artifact version is stamped on every hydrated record; (b) API responses expose as_of vintage; (c) parity diffs run continuously in CI. No write path exists that bypasses the static SoR — the database cannot drift from the public record.

4.11 Political-Intelligence Capability Data Structures (OSINT/INTOP, to 2037)

Master catalog: these schemas back the capabilities in FUTURE_MINDMAP.md §Political-Intelligence Capability Catalog and the architecture in FUTURE_ARCHITECTURE.md §4A. They extend the H2 OSINT structures (§3.5) with the fusion, warning, forecasting and provenance entities the current model does not yet carry. Every record is provenance-stamped and reproducible from public sources; every analytic record carries Admiralty grading and documented uncertainty.

4.11.1 Resolved-entity (cross-source entity resolution — C1)

{
  "schema": "intel-resolved-entity@3.0",
  "canonical_id": "person:0123",
  "labels": ["Förnamn Efternamn"],
  "links": [
    { "source": "riksdag", "ref": "intressent_0123", "confidence": 0.99 },
    { "source": "lobby-register", "ref": "org-2231", "confidence": 0.82, "relation": "former-employee" },
    { "source": "company-register", "ref": "orgnr-556xxx", "confidence": 0.71, "relation": "board-member" }
  ],
  "resolution_method": "embedding-match + deterministic-keys",
  "source_grading": { "reliability": "B", "credibility": "2" },
  "as_of": "2028-09-01",
  "ethics": "public-records-only; no private-life data"
}

4.11.2 Multi-INT fusion edge (C6)

{
  "schema": "intel-fusion-edge@3.0",
  "from": "person:0123",
  "to": "org:2231",
  "int_families": ["OSINT", "FININT"],
  "relation": "received-funding-while-voting-on-related-bill",
  "evidence": [
    { "type": "vote", "dok_id": "H802Vot44", "grade": "A1" },
    { "type": "funding-disclosure", "ref": "party-fin-2027", "grade": "B2" }
  ],
  "salience": 0.74,
  "neutrality_note": "Descriptive correlation on public records; not an allegation of wrongdoing.",
  "as_of": "2028-10-01"
}

4.11.3 Indications & Warning indicator (C14)

{
  "schema": "intel-iw-indicator@3.0",
  "tripwire": "coalition-rupture",
  "indicators": [
    { "name": "govt-bloc-cohesion-delta", "value": -0.18, "threshold": -0.15, "breached": true },
    { "name": "confidence-motion-filed", "value": 1, "threshold": 1, "breached": true }
  ],
  "warning_level": "elevated",
  "probability": { "point": 0.42, "band": "roughly even", "wep_lexicon": "ICD-203" },
  "evidence_dok_ids": ["H802Vot44", "H802Mot201"],
  "recommended_retasking": ["intel-multi-int-fusion", "intel-forecast-calibrate"],
  "human_review": "required",
  "as_of": "2029-02-14T09:00:00Z"
}

4.11.4 Forecast + calibration record (C13 / C29)

{
  "schema": "intel-forecast@3.0",
  "question_id": "PIR-2029-coalition-after-budget",
  "forecast": { "p": 0.38, "band": "unlikely", "horizon_days": 90 },
  "method": "ensemble (gradient-boost + LLM-scenario)",
  "assumptions_checked": ["KAC-passed"],
  "calibration": { "rolling_brier": 0.14, "n_resolved": 47, "trend": "improving" },
  "resolved": null,
  "as_of": "2029-03-01"
}

4.11.5 FIMI / coordinated-inauthentic-behaviour signal (C20)

{
  "schema": "intel-fimi-signal@3.0",
  "narrative_id": "frame:2029-energy-cost",
  "disarm_ttps": ["T0049.003", "T0017"],
  "amplification": { "coordinated_accounts_est": 0, "method": "aggregate-network-only", "individual_profiling": false },
  "attribution_confidence": { "band": "low", "wep_lexicon": "ICD-203" },
  "evidence": [{ "type": "public-discourse-aggregate", "grade": "C3" }],
  "ethics": "aggregate public discourse only; no citizen profiling; advisory, not accusatory",
  "as_of": "2029-04-10"
}

4.11.6 Estimative product (NIE-style key judgment — C22)

{
  "schema": "intel-estimative@3.0",
  "title": "Government durability through 2026 budget cycle",
  "key_judgments": [
    { "kj": "Government likely survives the budget vote.", "confidence": "moderate", "p": 0.66, "dissent": "minority view: snap election if SD defects" }
  ],
  "icd203_compliance": { "sources_characterized": true, "uncertainty_expressed": true, "assumptions_distinguished": true },
  "neutrality_audit": "party-symmetry-passed",
  "evidence_dok_ids": ["H802Vot44", "H802Bet12"],
  "human_signoff": "analyst-id",
  "as_of": "2029-05-01"
}

4.11.7 Provenance / content-credential record (C8 / C9)

{
  "schema": "intel-provenance@3.0",
  "asset_id": "evidence-9f2a",
  "origin": { "source": "riksdag", "url": "https://data.riksdagen.se/...", "fetched_at": "2029-01-02T08:00:00Z" },
  "c2pa": { "signed": true, "kms_key": "alias/intel-provenance", "manifest_hash": "sha256:..." },
  "synthetic_media_check": { "ran": true, "verdict": "authentic", "model": "df-detector-v3" },
  "chain_of_custody": ["fetch", "extract", "grade", "embed"],
  "refuse_to_cite": false
}

4.11.8 Capability-entity ERD (Horizon 3)

erDiagram
    RESOLVED_ENTITY ||--o{ FUSION_EDGE : participates
    FUSION_EDGE }o--|| PROVENANCE : "anchored by"
    IW_INDICATOR ||--o{ FORECAST : "re-tasks"
    FORECAST ||--o{ ESTIMATIVE : "feeds"
    ESTIMATIVE }o--|| PROVENANCE : "cites"
    FIMI_SIGNAL }o--|| PROVENANCE : "evidenced by"
    RESOLVED_ENTITY {
        string canonical_id PK
        string source_grading
        date as_of
    }
    FUSION_EDGE {
        string from FK
        string to FK
        float salience
    }
    IW_INDICATOR {
        string tripwire
        string warning_level
        float probability
    }
    FORECAST {
        string question_id PK
        float p
        float rolling_brier
    }
    ESTIMATIVE {
        string title
        string neutrality_audit
        string human_signoff
    }
    FIMI_SIGNAL {
        string narrative_id PK
        string attribution_confidence
    }
    PROVENANCE {
        string asset_id PK
        bool refuse_to_cite
    }

Data-governance rails. All capability entities inherit the §10 GDPR Article 9 posture (lawful bases 9(2)(e)/9(2)(g)), the §3.5.5 evidence-and-provenance invariant (no analytic record without a dok_id/primary-source anchor + Admiralty grade), and the §4.10 consistency contract (serverless stores never drift from the public static SoR). FININT/SOCMINT entities carry an explicit ethics field asserting public-records-only, aggregate, non-accusatory processing with no citizen profiling.

5. Source Ingestion & Integration

sequenceDiagram
    participant SRC as Riksdag / Regering / IMF / SCB / WB
    participant EB as EventBridge (scheduled)
    participant ING as Lambda Ingestors
    participant SF as Step Functions
    participant ART as Static Artifacts (S3)
    participant HYD as Hydrators
    participant STORE as Serverless Stores
    EB->>ING: Trigger scheduled fetch
    ING->>SRC: Pull public data
    ING->>ART: Write versioned JSON (system of record)
    ART->>SF: Emit "artifact updated"
    SF->>HYD: Orchestrate hydration
    HYD->>STORE: Upsert Neptune / Aurora / DynamoDB / OpenSearch / Timestream
    HYD->>STORE: Embed corpus → Bedrock KB

5.1 Migration Strategy

Dual-write era — static artifacts remain canonical; serverless stores are hydrated copies. UI can fall back to static at any time.
Read-shadow — API Gateway serves reads from serverless while continuously diffed against static for parity.
Promote — once parity is proven, interactive features (graph traversal, vector search) are exposed; static remains DR.

5.2 Integration Components

Component	Role
EventBridge	Scheduled ingestion triggers (replaces some cron in GitHub Actions)
Step Functions	Orchestrates multi-store hydration with retries
Kinesis (optional)	Streaming ingestion for high-frequency updates
Lambda	Stateless ingestors, hydrators, GraphQL resolvers

5.3 Data Quality & Cross-Source Reconciliation

Economic context blends three providers with a strict precedence rule (IMF primary, SCB ground truth, World Bank non-economic residue). Reconciliation guards against silent divergence:

graph TD
    IMF["IMF SWE (WEO/FM)"] --> CMP{">0.3pp delta?"}
    SCB["SCB national accounts"] --> CMP
    CMP -->|No| OK["Accept; tag dual-source provenance"]
    CMP -->|Yes| REV["Editorial review flag"]
    REV --> NOTE["Footnote both values + vintages"]
    WB["World Bank (non-economic only)"] --> GOV["Governance / environment"]
    style IMF fill:#bbdefb,stroke:#1565c0,color:#000000
    style SCB fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style WB fill:#fff9c4,stroke:#f9a825,color:#000000
    style CMP fill:#ffcc80,stroke:#e65100,color:#000000
    style REV fill:#ef9a9a,stroke:#b71c1c,color:#000000
    style OK fill:#a5d6a7,stroke:#1b5e20,color:#000000
    style NOTE fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style GOV fill:#bbdefb,stroke:#1565c0,color:#000000

Check	Rule	Action on failure
IMF ↔ SCB GDP/inflation delta	> 0.3pp triggers review	Footnote both values + vintages
Voting record completeness	every `votering_id` resolves to a `dok_id`	Reject artifact
Party roster integrity	active parties ∈ {S,M,SD,C,V,MP,KD,L}	Flag schema drift
Vintage freshness	IMF `vintage` within one WEO cycle	Re-fetch

6. GraphQL API Schema

type Politician {
  intressentId: ID!
  namn: String!
  parti: Party!
  valkrets: String
  status: String
  votes(rm: String): [Vote!]!
  speeches(rm: String): [Speech!]!
  anomalyScores: [AnomalyScore!]!
}

type Party {
  kod: ID!
  namn: String!
  active: Boolean!
  seats: Int
  cohesion(rm: String!): CohesionRow!
  alignments(rm: String!): [BlocEdge!]!
}

type Vote {
  voteringId: ID!
  document: Document!
  politician: Politician!
  rost: String!
  punkt: String
}

type Document {
  dokId: ID!
  doktyp: String
  titel: String
  rm: String
  publicerad: String
}

type Speech {
  anforandeId: ID!
  politician: Politician!
  document: Document
  rm: String
}

type CohesionRow {
  party: String!
  cohesionIndex: Float!
  rebelVotes: Int!
  evidenceDokIds: [String!]!
}

type BlocEdge {
  source: String!
  target: String!
  alignment: Float!
  intopClass: String!
  evidenceDokIds: [String!]!
}

type AnomalyScore {
  entity: String!
  metric: String!
  z: Float!
  flag: String!
  explanation: String!
}

type Query {
  politician(intressentId: ID!): Politician
  party(kod: ID!): Party
  searchDocuments(query: String!, rm: String): [Document!]!
  ragAnswer(question: String!): GroundedAnswer!
}

type GroundedAnswer {
  answer: String!
  citations: [String!]!   # dok_ids — never empty
}

type Mutation {
  refreshHydration(source: String!): HydrationResult!
}

type HydrationResult {
  source: String!
  updatedAt: String!
  recordsUpserted: Int!
}

type Subscription {
  newAnomaly(party: String): AnomalyScore!
  voteAdded(rm: String!): Vote!
}

GroundedAnswer.citations is non-nullable and must be non-empty — the schema itself enforces evidence-first AI output.

7. Data Model Diagrams

7.1 Cross-Horizon ERD

erDiagram
    POLITICIAN ||--o{ VOTE : casts
    POLITICIAN ||--o{ SPEECH : delivers
    POLITICIAN ||--o{ NETWORK_EDGE : connects
    POLITICIAN ||--o{ ANOMALY_SCORE : flagged
    PARTY ||--o{ POLITICIAN : includes
    PARTY ||--o{ COHESION_ROW : scored
    PARTY ||--o{ BLOC_EDGE : aligns
    DOCUMENT ||--o{ VOTE : triggers
    DOCUMENT ||--o{ EMBEDDING : indexed
    COMMITTEE ||--o{ DOCUMENT : produces
    POLITICIAN {
        string intressent_id PK
        string parti FK
    }
    PARTY {
        string kod PK
        boolean active
    }
    DOCUMENT {
        string dok_id PK
        string rm
    }
    VOTE {
        string votering_id PK
        string rost
    }
    SPEECH {
        string anforande_id PK
    }
    COHESION_ROW {
        string party FK
        float cohesion_index
    }
    BLOC_EDGE {
        string source FK
        string target FK
        float alignment
    }
    NETWORK_EDGE {
        string from FK
        string to FK
        int weight
    }
    ANOMALY_SCORE {
        string entity FK
        float z
    }
    EMBEDDING {
        string dok_id FK
        int dimension
    }
    COMMITTEE {
        string kod PK
    }

7.2 AWS Service Integration

graph TB
    subgraph EDGE["Edge"]
        CF["CloudFront"]
        WAF["AWS WAF"]
    end
    subgraph ACCESS["Access Tier"]
        APIGW["API Gateway"]
        COG["Cognito"]
    end
    subgraph COMPUTE["Compute"]
        L["Lambda"]
        SF["Step Functions"]
        EB["EventBridge"]
    end
    subgraph DATA["Data Tier"]
        NEP["Neptune"]
        AUR["Aurora v2"]
        DDB["DynamoDB"]
        OS["OpenSearch"]
        TS["Timestream"]
        KB["Bedrock KB"]
        S3["S3 (static SoR)"]
    end
    CF --> WAF --> APIGW --> COG --> L
    EB --> SF --> L
    L --> NEP & AUR & DDB & OS & TS & KB
    L -. fallback .-> S3
    style CF fill:#ffe0b2,stroke:#e65100,color:#000000
    style WAF fill:#ffe0b2,stroke:#e65100,color:#000000
    style APIGW fill:#ffcc80,stroke:#e65100,color:#000000
    style COG fill:#ffcc80,stroke:#e65100,color:#000000
    style L fill:#fff9c4,stroke:#f9a825,color:#000000
    style SF fill:#fff9c4,stroke:#f9a825,color:#000000
    style EB fill:#fff9c4,stroke:#f9a825,color:#000000
    style NEP fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style AUR fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style DDB fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style OS fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style TS fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style KB fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style S3 fill:#bbdefb,stroke:#1565c0,color:#000000

7.3 Query Data-Flow Sequence

sequenceDiagram
    participant U as User
    participant CF as CloudFront
    participant GW as API Gateway
    participant C as Cognito
    participant L as Lambda
    participant N as Neptune
    participant O as OpenSearch
    participant K as Bedrock KB
    U->>CF: GraphQL query
    CF->>GW: Forward
    GW->>C: Authorize + throttle
    C-->>GW: Token OK
    GW->>L: Resolve
    L->>N: Graph traversal (co-voting)
    L->>O: Vector + text search
    L->>K: RAG retrieve (grounded)
    K-->>L: Answer + dok_id citations
    L-->>U: Response (evidence-linked)

7.4 Neptune Graph Visualization

graph LR
    P1((MP A · S)) ---|co_voted 4210| P2((MP B · V))
    P1 ---|co_sponsored 23| P3((MP C · S))
    P2 ---|committee FiU| P4((MP D · MP))
    P3 ---|co_voted 1880| P4
    style P1 fill:#ef9a9a,stroke:#b71c1c,color:#000000
    style P3 fill:#ef9a9a,stroke:#b71c1c,color:#000000
    style P2 fill:#ef9a9a,stroke:#b71c1c,color:#000000
    style P4 fill:#a5d6a7,stroke:#1b5e20,color:#000000

7.5 Time-Series Flow

graph LR
    SRC["Vote / speech events"] --> ING["Lambda ingestor"]
    ING --> TS["Timestream"]
    TS --> AGG["Scheduled aggregation"]
    AGG --> ANOM["Anomaly z-scores"]
    ANOM --> API["API Gateway"]
    style SRC fill:#bbdefb,stroke:#1565c0,color:#000000
    style ING fill:#fff9c4,stroke:#f9a825,color:#000000
    style TS fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style AGG fill:#fff9c4,stroke:#f9a825,color:#000000
    style ANOM fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style API fill:#ffcc80,stroke:#e65100,color:#000000

7.6 Bedrock KB RAG Pipeline

graph TD
    DOCS["Documents + speeches (public)"] --> CHUNK["Chunk + grade source"]
    CHUNK --> EMB["Embed (Titan / Bedrock)"]
    EMB --> KB["Knowledge Base vector store"]
    Q["Citizen question"] --> RET["Retrieve top-k"]
    KB --> RET
    RET --> GEN["Generate grounded answer"]
    GEN --> CITE["Attach dok_id citations"]
    CITE --> OUT["Answer (never ungrounded)"]
    style DOCS fill:#bbdefb,stroke:#1565c0,color:#000000
    style CHUNK fill:#fff9c4,stroke:#f9a825,color:#000000
    style EMB fill:#fff9c4,stroke:#f9a825,color:#000000
    style KB fill:#e1bee7,stroke:#6a1b9a,color:#000000
    style RET fill:#fff9c4,stroke:#f9a825,color:#000000
    style GEN fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style CITE fill:#c8e6c9,stroke:#2e7d32,color:#000000
    style OUT fill:#a5d6a7,stroke:#1b5e20,color:#000000

8. Implementation Roadmap

gantt
    title Riksdagsmonitor Data Architecture Roadmap (2026-2037)
    dateFormat YYYY-MM-DD
    section Horizon 2 (Static)
    Party cohesion matrices        :h2a, 2026-06-01, 180d
    Coalition / bloc graphs         :h2b, after h2a, 150d
    OSINT structures + INTOP        :h2c, 2026-09-01, 240d
    Build pipeline hardening        :h2d, after h2c, 120d
    section Horizon 3 Phase 1 (Foundation)
    Aurora v2 + DynamoDB            :h3a, 2028-01-01, 200d
    Ingestion (EventBridge/Step)    :h3b, after h3a, 150d
    section Horizon 3 Phase 2 (Graph + Search)
    Neptune Serverless              :h3c, 2029-01-01, 220d
    OpenSearch + vectors            :h3d, after h3c, 180d
    section Horizon 3 Phase 3 (AI)
    Bedrock Knowledge Bases RAG     :h3e, 2030-01-01, 240d
    Timestream anomaly pipeline     :h3f, after h3e, 180d
    section Horizon 3 Phase 4 (Access)
    API Gateway + Cognito GA        :h3g, 2031-01-01, 200d
    Interactive features GA         :h3h, after h3g, 365d
    section Long Horizon
    Pre-AGI scaling                 :lh1, 2032-01-01, 730d
    AGI/Post-AGI data ops           :lh2, 2034-01-01, 1095d

Phase	Window	Outcome
H2 Static	2026–2027	Richer pre-computed party + OSINT datasets, no servers
H3 Phase 1	2028	Relational + KV foundation, hydration from static SoR
H3 Phase 2	2029	Graph traversal + vector search
H3 Phase 3	2030	RAG + anomaly time-series
H3 Phase 4	2031+	API Gateway + Cognito GA, interactive UX
Long Horizon	2032–2037	Pre-AGI → AGI data operations

9. Technology Stack Evolution & Cost Projections

9.1 Stack by Horizon

Layer	H1	H2	H3
Hosting	S3 + CloudFront, GH Pages DR	Same	+ serverless tier
Compute	GitHub Actions	GitHub Actions (heavier)	Lambda + Step Functions
Storage	JSON/CSV files	+ party/OSINT JSON	Neptune/Aurora/DynamoDB/OpenSearch/Timestream
AI	Newsroom Opus 4.x	+ build-time embeddings	Bedrock KB RAG
Access	Static fetch	Static fetch	API Gateway + Cognito

9.2 Cost Posture (Targets)

Horizon	Cost model	Notes
H1	Near-zero marginal (CDN + Actions)	Static economics
H2	Slightly higher Actions minutes	No new runtime cost
H3	Pay-per-use serverless	Scales to zero; static fallback caps risk

All serverless stores chosen for scale-to-zero / on-demand billing to preserve the platform's low-cost, sustainable, open-source posture.

10. ISMS Compliance & Data Governance

10.1 Framework Mapping

Control area	ISO 27001:2022	NIST CSF 2.0	CIS Controls v8.1
Access control (Cognito)	A.5.15, A.5.18	PR.AA	CIS 6
Data classification	A.5.12	ID.AM	CIS 3
Logging & audit (API GW)	A.8.15	DE.CM	CIS 8
Cryptography (in transit/at rest)	A.8.24	PR.DS	CIS 3
Supply chain (SLSA, npm)	A.5.19–A.5.21	ID.SC	CIS 16
Secure development	A.8.25–A.8.28	PR.PS	CIS 16

Political opinions = special-category data; lawful bases 9(2)(e) (manifestly made public) and 9(2)(g) (substantial public interest).
Public data only — exclusively official primary sources; no private individuals, no leaked/hacked data.
Data minimisation, purpose limitation, storage limitation, integrity & confidentiality applied across all horizons.
DPIA required before activating any H3 interactive feature processing personal data at new scale.

10.3 Data Classification

Class	Examples	Handling
Public	Votes, documents, speeches, party metadata	Open read; integrity-protected
Derived-Public	Cohesion/coalition/OSINT datasets	Provenance-tagged; reproducible
Operational	Hydration state, audit logs	Cognito-gated; retention-limited

10.4 Data Lifecycle

stateDiagram-v2
    [*] --> Ingested: Fetch public source
    Ingested --> Validated: analysis-gate.ts
    Validated --> Published: Static artifact (SoR)
    Published --> Hydrated: Serverless projection
    Hydrated --> Served: API Gateway + Cognito
    Served --> Archived: Retention policy
    Archived --> [*]
    Validated --> Rejected: Provenance incomplete
    Rejected --> [*]

10.5 Stakeholder & Risk Analysis

Stakeholders (power × interest, neutral framing):

Stakeholder	Interest	Data-tier implication
Citizens	Accessible, neutral accountability	Static-first, 14 languages, WCAG 2.1 AA
Journalists / researchers	Queryable evidence with citations	H3 GraphQL + grounded RAG
Parliamentary parties (all 8)	Fair, equal, reproducible treatment	Symmetric datasets; no editorial scoring
Hack23 maintainers	Sustainable, low-cost ops	Scale-to-zero serverless; static DR
Regulators (GDPR/NIS2)	Lawful, auditable processing	Cognito audit trail; DPIA gate

Top data-architecture risks (target mitigations):

Risk	Likelihood	Impact	Mitigation
Serverless drift from public record	Low	High	Static SoR canonical; continuous parity diff
Source API change (Riksdag/IMF/SCB)	Medium	Medium	Versioned ingestors; vintage stamping; allowlisted hosts
AI ungrounded/biased output	Medium	High	Non-empty `dok_id` citations enforced by schema; neutrality lint; human-in-loop
Cost overrun (H3)	Low	Medium	Pay-per-use; static fallback caps blast radius
Provenance loss	Low	High	`source_grading` + INTOP mandatory at gate
Perceived partisanship	Low	High	Symmetric party datasets; published methodology; reproducibility

The single most important control is the static system-of-record: because every serverless and AI output is derived from immutable, citation-tagged public artifacts, the platform cannot silently fabricate or skew the political record.

11. IMF Data Domain — Filesystem Cache → Aurora Schema

IMF is the primary economic data domain (ADR 0001). Today it is a filesystem cache; in Horizon 3 it is projected into Aurora while the cache remains the system of record.

11.1 Current Cache (H1/H2)

Client: scripts/imf-client.ts (pure TypeScript, not an MCP server).
Cache: analysis/data/imf/{indicator}/{country}.json + .meta.json sidecar.
Vintage labels: e.g. WEO-2026-04.
Dataflows: WEO, FM, IFS, BOP, DOTS, GFS_COFOG, PCPS, ER, MFS_IR, MFS_PR; T+5 projections.
Allowlisted hosts: data.imf.org, api.imf.org, www.imf.org.

11.2 Aurora Projection (H3)

CREATE TABLE imf_cache (
    indicator      TEXT NOT NULL,
    country        TEXT NOT NULL,
    period         TEXT NOT NULL,
    value          DOUBLE PRECISION,
    is_projection  BOOLEAN NOT NULL DEFAULT FALSE,
    vintage        TEXT NOT NULL,         -- e.g. 'WEO-2026-04'
    dataflow       TEXT NOT NULL,         -- WEO/FM/IFS/...
    fetched_at     TIMESTAMPTZ NOT NULL,
    PRIMARY KEY (indicator, country, period, vintage)
);

CREATE TABLE article_economic_provenance (
    article_id     TEXT NOT NULL,
    indicator      TEXT NOT NULL,
    country        TEXT NOT NULL,
    period         TEXT NOT NULL,
    vintage        TEXT NOT NULL,
    used_at        TIMESTAMPTZ NOT NULL,
    PRIMARY KEY (article_id, indicator, country, period, vintage)
);

CREATE INDEX idx_imf_dataflow ON imf_cache(dataflow);
CREATE INDEX idx_imf_country ON imf_cache(country);

11.3 EconomicDataSource Discriminated Union

type EconomicDataSource =
  | { provider: "IMF"; dataflow: "WEO" | "FM" | "IFS" | "BOP" | "DOTS"
      | "GFS_COFOG" | "PCPS" | "ER" | "MFS_IR" | "MFS_PR"; vintage: string }
  | { provider: "SCB"; table: string }                 // Swedish ground truth
  | { provider: "WorldBank"; series: string; economic: false }; // non-economic only

11.4 Provider Decision Matrix

Need	Provider	Rationale
GDP, inflation, fiscal, BoP	IMF	Primary economic; T+5 projections
Swedish national statistics	SCB	Authoritative ground truth
Governance / environment / social	World Bank	Non-economic residue only
Economic series	❌ World Bank	Deprecated — use IMF

11.5 IMF Data Classification

IMF cache entries are Derived-Public: openly published source values, provenance-tagged with vintage + dataflow, fully reproducible, and linked to articles via article_economic_provenance for editorial accountability.

12. AI/LLM Data Architecture Evolution (2026–2037)

The newsroom and analytical layer evolve with frontier-model capability. The table below translates the AI model roadmap into data-architecture implications — what each capability tier demands from the data tier.

Year	Model Tier	Status	Data-Architecture Implication
2026	Opus 4.6–4.9	🟢 Current	Static build-time authoring; corpus as files; RAG-ready embeddings begin
2027	Opus 5.x	🔵 Near	Richer H2 OSINT datasets feed model context; embeddings standardized
2028	Opus 6.x	🟣 Planned	H3 Phase 1: relational + KV stores hydrate model retrieval
2029	Opus 7.x	🟠 Projected	Graph + vector retrieval (Neptune + OpenSearch) for grounded synthesis
2030	Opus 8.x	🔴 Horizon	Bedrock KB RAG GA; anomaly time-series inform model prompts
2031–2033	Opus 9–10.x / Pre-AGI	⚪ Speculative	Real-time grounded Q&A via API Gateway; strict citation enforcement
2034–2037	AGI / Post-AGI	⭐ Visionary	Autonomous evidence-bound analysis; human-in-the-loop governance retained

timeline
    title AI ↔ Data Tier Co-Evolution
    2026 : Opus 4.x : Static authoring + embeddings
    2027 : Opus 5.x : H2 OSINT context
    2028 : Opus 6.x : Aurora + DynamoDB retrieval
    2029 : Opus 7.x : Neptune + OpenSearch grounding
    2030 : Opus 8.x : Bedrock KB RAG GA
    2031 : Pre-AGI : Real-time grounded Q&A
    2034 : AGI/Post-AGI : Autonomous evidence-bound analysis

Governance invariant across all tiers: every AI-generated claim is retrieval-grounded and dok_id-cited, neutrality is enforced, and a human-in-the-loop governance gate is retained even at AGI tiers. Capability never overrides the evidence-first and neutrality principles.

12.1 RAG Data Contract

Field	Requirement
`answer`	Generated text
`citations`	Non-empty array of `dok_id`
`source_grading`	Admiralty reliability/credibility per retrieved chunk
`neutrality_check`	Passed before publication

Document	Relationship
DATA_MODEL.md	Current-state baseline (Horizon 1) this doc extends
FUTURE_ARCHITECTURE.md	Target system architecture
FUTURE_SECURITY_ARCHITECTURE.md	Target security controls
FUTURE_THREAT_MODEL.md	Target threat model
ARCHITECTURE.md	Current architecture
analysis/imf/README.md	IMF data domain reference

📄 Document Control

Field	Value
Document Owner	CEO
Version	3.0
Last Updated	2026-05-31 (UTC)
Review Cycle	Annual
Next Review	2027-05-31
Classification	Public
Data Posture	Public sources only; GDPR Art. 9 (9(2)(e), 9(2)(g)); strict neutrality

🔗 Hack23 Ecosystem

Riksdagsmonitor is part of the Hack23 open-source transparency ecosystem, operated under the Hack23 ISMS-PUBLIC governance framework (ISO 27001:2022, NIST CSF 2.0, CIS Controls v8.1, GDPR, NIS2).

Mission: empower citizens, strengthen democratic accountability, and illuminate the political process with rigorous, neutral, evidence-based intelligence drawn exclusively from public sources.