🔵 HORIZON v3.0+

May 31, 2026 · View on GitHub

🔄 EU Parliament Monitor — Future Flowcharts

📊 From Static Intelligence Generation to AWS-Native Serverless OSINT Operations
🎯 Three Horizons: Enhanced Static (v2.0) → AWS Serverless Platform (v3.0+) → 10-Year AI Lookahead (2026-2037)

📋 Document Owner: CEO | 📄 Version: 4.1 | 📅 Last Updated: 2026-05-31 (UTC)
🔄 Review Cycle: Quarterly | ⏰ Next Review: 2026-08-31 | 🏷️ Release: v1.0.1
🏷️ Classification: Public (Open Source European Parliament Monitoring Platform)

📚 Architecture Documentation Map

Document	Focus	Description	Documentation Link
Architecture	🏛️ Architecture	C4 model showing current system structure	View Source
Future Architecture	🏛️ Architecture	C4 model showing future system structure	View Source
Mindmaps	🧠 Concept	Current system component relationships	View Source
Future Mindmaps	🧠 Concept	Future capability evolution	View Source
SWOT Analysis	💼 Business	Current strategic assessment	View Source
Future SWOT Analysis	💼 Business	Future strategic opportunities	View Source
Data Model	📊 Data	Current data structures and relationships	View Source
Future Data Model	📊 Data	Enhanced European Parliament data architecture	View Source
Flowcharts	🔄 Process	Current data processing workflows	View Source
Future Flowcharts	🔄 Process	Enhanced AI-driven workflows	This Document
State Diagrams	🔄 Behavior	Current system state transitions	View Source
Future State Diagrams	🔄 Behavior	Enhanced adaptive state transitions	View Source
Security Architecture	🛡️ Security	Current security implementation	View Source
Future Security Architecture	🛡️ Security	Security enhancement roadmap	View Source
Threat Model	🎯 Security	STRIDE threat analysis	View Source
Classification	🏷️ Governance	CIA classification & BCP	View Source
CRA Assessment	🛡️ Compliance	Cyber Resilience Act	View Source
Workflows	⚙️ DevOps	CI/CD documentation	View Source
Future Workflows	🚀 DevOps	Planned CI/CD enhancements	View Source
Business Continuity Plan	🔄 Resilience	Recovery planning	View Source
Financial Security Plan	💰 Financial	Cost & security analysis	View Source
End-of-Life Strategy	📦 Lifecycle	Technology EOL planning	View Source
Unit Test Plan	🧪 Testing	Unit testing strategy	View Source
E2E Test Plan	🔍 Testing	End-to-end testing	View Source
Performance Testing	⚡ Performance	Performance benchmarks	View Source
Security Policy	🔒 Security	Vulnerability reporting & security policy	View Source

🔐 ISMS Policy Alignment

This future flowchart is designed to implement all controls from Hack23 AB's ISMS framework as the EU Parliament Monitor platform evolves across its three horizons — v2.0 enhanced static intelligence, v3.0+ AWS-native serverless OSINT operations, and the 10-year AI lookahead. The static front door remains the cheap, cacheable, auditable substrate; dynamic AWS-native features layer behind it without weakening provenance or neutrality guarantees.

Policy Domain	Policy	Planned Implementation
🔐 Core Security	Information Security Policy	Overall security governance for enhanced monitoring & AWS workloads
🤖 AI Governance	AI Policy	AI = proposal generator; human accountability; no autonomous deploy
🛠️ Development	Secure Development Policy	Security-integrated SSDLC for Lambda/Step Functions pipelines
🌐 Network	Network Security Policy	CloudFront + AWS WAF + Shield, VPC isolation, private endpoints
🔒 Cryptography	Cryptography Policy	AWS KMS, TLS 1.3, SLSA provenance, content integrity verification
🔑 Access Control	Access Control Policy	Amazon Cognito, IAM least-privilege, API Gateway authorizers
🏷️ Data Classification	Data Classification Policy	European Parliament PUBLIC open-data classification
🔍 Vulnerability	Vulnerability Management	CodeQL, OpenSSF Scorecard, GuardDuty, Security Hub automation
🚨 Incident Response	Incident Response Plan	CloudWatch alarms, EventBridge-driven automated response
💾 Backup & Recovery	Backup Recovery Policy	S3 versioning, DynamoDB PITR, Aurora snapshots, Git provenance
🔄 Business Continuity	Business Continuity Plan	Multi-AZ serverless, CloudFront edge, static fallback
🤝 Third-Party	Third Party Management	EP MCP, World Bank MCP, IMF REST, Bedrock model provider review
🏷️ Classification	Classification Framework	Business impact analysis for platform

Compliance Framework Mapping

Framework	Version	Relevant Controls
ISO 27001	2022	A.5.1, A.5.23, A.8.25, A.8.26, A.8.27, A.8.28
NIST CSF	2.0	GV.OC, GV.RM, ID.AM, PR.AT, PR.DS, DE.CM
CIS Controls	v8.1	Control 1-5, 8, 13, 14, 16
GDPR	2016/679	Public MEP roles only; data minimization; purpose limitation

📋 Executive Summary

This document maps the evolution of EU Parliament Monitor's process and data flows across three strategic horizons. Today's platform (v1.0.x) is a pure static-site generator — gh-aw agentic workflows author Stage-B analysis markdown, the deterministic src/aggregator/** pipeline renders 14-language HTML, and the result is served from Amazon S3 + Amazon CloudFront. No servers, no runtime database, no AI-authored HTML.

🟢 v2.0 — Enhanced Static Intelligence (2026 H2 → 2027): keep the static architecture. Deepen the political-landscape and party/political-group dashboards and the OSINT tradecraft (51-template catalog, ICD 203 confidence, Admiralty grading, Kent/WEP bands, 5-framework political threat methodology, 2-pass AI-first quality). All dashboard data is pre-computed at build time and delivered as static assets via CloudFront. The moat is analysis quality, not real-time infrastructure.
🔵 v3.0+ — AWS-Native Serverless OSINT Platform (2028+): layer dynamic, event-driven intelligence behind the static edge using AWS Lambda, Step Functions, EventBridge, Kinesis, API Gateway, AppSync, Cognito, DynamoDB, Aurora Serverless v2, OpenSearch Serverless, Neptune Serverless, and Amazon Bedrock (foundation models, Knowledge Bases RAG, Bedrock Agents, Guardrails). Real-time EP event ingestion, natural-language query over the political knowledge graph, and an API ecosystem for journalists and researchers.
⚪ 10-Year AI Lookahead (2026 → 2037): annual major-model upgrades, competitor evaluation each release, model-agnostic routing via Bedrock, and resilience to paradigm shifts (quantum AI, neuromorphic computing) and AGI / post-AGI — all governed by the Hack23 AI Policy (AI proposes, humans are accountable, no autonomous deploy).

Workflow Transformation Across Horizons

Aspect	Current (v1.0.x)	v2.0 (Static-Enhanced)	v3.0+ (AWS Serverless)
Trigger	Scheduled `gh-aw` (cron)	Scheduled + richer build matrix	EventBridge / Kinesis (real-time)
Compute	GitHub Actions runners	GitHub Actions runners	AWS Lambda + Step Functions
Authoring	LLM (Claude) → markdown	LLM + 51-template OSINT catalog	Bedrock + Knowledge Bases RAG + Agents
Rendering	`src/aggregator/**` → HTML	Aggregator + richer dashboards	Aggregator (edge) + dynamic APIs
Dashboards	Chart.js 4 + D3 7 (in-article)	Pre-computed party/landscape datasets	Live datasets via AppSync/API Gateway
Data stores	Committed markdown + JSON	Committed artifacts + build cache	DynamoDB · Aurora · OpenSearch · Neptune
Delivery	S3 + CloudFront (static)	S3 + CloudFront (static)	CloudFront static edge + serverless APIs
Query	None (pre-rendered pages)	Faceted client-side filtering	NL query over knowledge graph (Bedrock)
Neutrality control	Style guide + 2-pass review	Style guide + ICD 203 + Admiralty	Bedrock Guardrails + human sign-off

📅 Workflow Evolution Roadmap

gantt
    title Workflow Evolution Roadmap (v2.0 to v3.0+)
    dateFormat YYYY-MM

    section v2.0 Enhanced Static Intelligence
    Party landscape dashboards            :v2a, 2026-07, 3M
    Coalition mathematics visualizations  :v2b, 2026-09, 2M
    OSINT 51-template catalog hardening   :v2c, 2026-08, 3M
    ICD 203 + Admiralty quality gates     :v2d, 2026-10, 2M
    Seat projection + election cycle views :v2e, 2026-11, 3M

    section v3.0 Serverless Foundation
    EventBridge + Kinesis ingestion       :v3a, 2028-01, 3M
    Lambda + Step Functions orchestration :v3b, 2028-02, 3M
    Neptune knowledge graph build         :v3c, 2028-04, 4M
    Cognito + API Gateway + AppSync       :v3d, 2028-06, 3M

    section v3.x Bedrock Intelligence
    Bedrock Knowledge Bases RAG           :v3e, 2028-09, 3M
    Bedrock Agents OSINT workflows        :v3f, 2028-11, 4M
    NL query over knowledge graph         :v3g, 2029-02, 3M
    Guardrails neutrality + GDPR controls :v3h, 2029-01, 3M

    section 10-Year AI Lookahead
    Model-agnostic Bedrock routing        :ai1, 2029-06, 6M
    Autonomous agentic pipelines (assisted) :ai2, 2030-01, 12M
    Multi-parliament expansion            :ai3, 2031-01, 18M

🟢 HORIZON v2.0 — Enhanced Static Intelligence Flows

Architecture unchanged: build-time generation, gh-aw + aggregator, pure static delivery on Amazon S3 + Amazon CloudFront. v2.0 adds depth — better party/political-landscape dashboards and stronger OSINT tradecraft — while every byte served remains a pre-computed static asset.

🏛️ Party & Political-Landscape Dashboard Generation Flow

Build-time flow producing pre-computed party / political-group dashboard datasets from European Parliament open data. All charts render client-side from JSON baked into the static bundle — no server, no runtime query.

flowchart TD
    Trigger([🕒 gh-aw Scheduled Workflow<br/>GitHub Actions]) --> MCPFetch[🔌 EP MCP Server Fetch<br/>get_meps, get_voting_records<br/>analyze_coalition_dynamics<br/>compare_political_groups]

    MCPFetch --> Enrich[🌐 Optional Context Enrichment<br/>World Bank MCP WDI<br/>IMF REST WEO/FM]

    Enrich --> Analysis[🧠 Stage-A/B Analysis Artifacts<br/>significance-scoring, actor-mapping<br/>coalition-dynamics, voting-patterns<br/>51-template catalog]

    Analysis --> Precompute[⚙️ Pre-compute Dashboard Datasets<br/>Party cohesion + defection rates<br/>Coalition mathematics<br/>Seat projections + election cycle<br/>Cross-party alliance edges]

    Precompute --> Shape[📐 Shape Static JSON<br/>Heatmap matrices<br/>Network graph nodes/edges<br/>Time-series voting trends]

    Shape --> AggRender[🏗️ Deterministic Aggregator<br/>src/aggregator/** renders HTML<br/>Embeds Chart.js 4 + D3 7]

    AggRender --> Lang[🌍 14-Language Expansion<br/>Per-language static pages]

    Lang --> Commit[💾 Commit Artifacts + Provenance<br/>SLSA 3, signed, version-controlled]

    Commit --> Deploy[📤 Deploy Static Bundle<br/>Amazon S3 origin]

    Deploy --> CDN[⚡ Amazon CloudFront<br/>Global edge cache<br/>Lambda@Edge headers/security]

    CDN --> Client[🖥️ Browser Renders Dashboards<br/>Client-side Chart.js/D3<br/>Faceted filtering, no backend]

    Client --> Done([✅ Static Party/Landscape Dashboard Live<br/>Cacheable, cheap, auditable])

    style Trigger fill:#e8f5e9
    style MCPFetch fill:#e1f5ff
    style Precompute fill:#fff4e1
    style AggRender fill:#e1f5ff
    style CDN fill:#fff4e1
    style Done fill:#d4edda

Dashboard Datasets Produced (v2.0)

Dashboard	Source MCP Tools	Visualization	Delivery
Party / group landscape	`generate_political_landscape`, `get_meps`	Seat-share treemap, group cards	Static JSON + Chart.js
Coalition mathematics	`analyze_coalition_dynamics`, `compare_political_groups`	Stacked-bar majority calculator	Static JSON + D3
Cohesion & defection	`sentiment_tracker`, `detect_voting_anomalies`	Cohesion heatmap, z-score flags	Static JSON + Chart.js
Cross-party alliances	`network_analysis`, `analyze_voting_patterns`	Force-directed network graph	Static JSON + D3
Seat projection / election cycle	electoral-domain artifacts	Trend lines + projection bands	Static JSON + Chart.js
MEP / party scorecards	`assess_mep_influence`, `analyze_legislative_effectiveness`	Radar + ranked tables	Static JSON + Chart.js

🔬 Enhanced OSINT Quality Pipeline (v2.0)

The 2.0 moat. Every analysis artifact passes structured OSINT tradecraft — ICD 203 confidence verbalization, Admiralty source grading, Kent/WEP probability bands, structured analytic techniques (ACH) — and the mandatory 2-pass AI-first quality loop. STRIDE is explicitly rejected for political analysis in favour of the 5-framework political threat methodology (Political Threat Landscape 6D + Attack Trees + Kill Chain + Diamond Model + ICO Profiling).

flowchart TD
    Raw([📥 Raw EP Open Data<br/>MCP feeds, documents, votes]) --> Collect[🗂️ Collection & Provenance<br/>Source capture<br/>PUBLIC-only classification]

    Collect --> Grade[🏷️ Admiralty Source Grading<br/>Reliability A-F<br/>Credibility 1-6]

    Grade --> Template[📚 51-Template Catalog Routing<br/>artifact-catalog.md maps<br/>artifact to methodology]

    Template --> Pass1[✍️ Pass 1 - Initial Analysis<br/>SWOT, actor-mapping<br/>coalition-dynamics, risk-matrix<br/>significance-scoring]

    Pass1 --> Confidence[📊 ICD 203 Confidence Verbalization<br/>High / Moderate / Low<br/>Kent/WEP probability bands]

    Confidence --> Threat[🎯 5-Framework Threat Methodology<br/>Threat Landscape 6D, Attack Trees<br/>Kill Chain, Diamond, ICO<br/>STRIDE rejected]

    Threat --> ACH[🔀 Structured Analytic Techniques<br/>Competing hypotheses ACH<br/>Key assumptions check<br/>Devil's advocacy]

    ACH --> Pass2[🔁 Pass 2 - Full Read-Back<br/>Word-by-word improvement<br/>Expand shallow sections<br/>Add evidence citations]

    Pass2 --> Gate{✅ Quality Gate<br/>Reference thresholds met?}

    Gate -->|❌ Below floor| Refine[🔧 Refine & Re-deepen<br/>Add cross-references<br/>Add confidence levels]
    Refine --> Pass2

    Gate -->|✅ Meets floor| Neutral{⚖️ Neutrality & GDPR Check<br/>Politically neutral?<br/>Public roles only?}

    Neutral -->|Biased / PII risk| Adjust[🎨 Neutralize & Minimize<br/>Remove opinion<br/>Public MEP roles only]
    Adjust --> Neutral

    Neutral -->|✅ Neutral| Audit[🧾 Self-Audit Artifacts<br/>workflow-audit<br/>methodology-reflection]

    Audit --> Publish([✅ Committed Analysis Artifacts<br/>Feed aggregator + dashboards])

    style Raw fill:#e8f5e9
    style Gate fill:#fff4e1
    style Neutral fill:#fff4e1
    style Refine fill:#fff9c4
    style Adjust fill:#fff9c4
    style Publish fill:#d4edda
``$

### \text{OSINT} \text{Tradecraft} \text{Standards} \text{Applied}

| \text{Standard} | \text{Purpose} | \text{Where} \text{Enforced} |
| -------- | ------- | -------------- |
| **\text{ICD} 203** | \text{Analytic} \text{confidence} \text{verbalization} | \text{Every} \text{assessment} \text{carries} \text{High}/\text{Moderate}/\text{Low} |
| **\text{Admiralty} \text{System}** | \text{Source} \text{reliability}  \times  \text{credibility} \text{grading} | \text{Collection} \text{stage}, \text{before} \text{analysis} |
| **\text{Kent} / \text{WEP} \text{bands}** | \text{Calibrated} \text{probability} \text{language} | \text{Forecast} & \text{scenario} \text{artifacts} |
| **\text{ACH}** | \text{Competing}-\text{hypothesis} \text{discipline} | \text{Significance} & \text{actor}-\text{mapping} \text{artifacts} |
| **5-\text{framework} \text{threat}** | \text{Political} \text{threat} \text{analysis} (\text{not} \text{STRIDE}) | $political-threat-landscape` artifact |
| **2-pass AI-first** | Depth & evidence quality floor | All artifacts, Pass 1 + Pass 2 |

---

# 🔵 HORIZON v3.0+ — AWS-Native Serverless OSINT Operations

> **All-in on AWS, fully serverless.** The static CloudFront edge stays the public,
> cacheable front door; dynamic intelligence is layered behind it. No Kubernetes,
> no self-managed message bus, no self-hosted databases — every component is a
> managed AWS service with IAM least-privilege and zero-ops scaling.

## ⚡ Real-Time EP Event Ingestion Flow

**Event-driven pipeline** replacing scheduled polling with near-real-time capture
of plenary votes, committee activity, and document publication.

```mermaid
flowchart TD
    EP([🌍 European Parliament Events<br/>Plenary votes, committees, docs]) --> Ingest[🔌 Ingestion Lambda<br/>EP MCP + Open Data Portal<br/>Poll/subscribe adapters]

    Ingest --> Stream[🌊 Amazon Kinesis Data Streams<br/>Ordered event shards<br/>Durable buffer]

    Stream --> Bus[📡 Amazon EventBridge<br/>Event bus + rules<br/>Schema registry]

    Bus --> Route{🔀 EventBridge Rule Routing<br/>detail-type}

    Route -->|Plenary Vote| VoteSF[🗳️ Vote Step Functions<br/>Enrich + score significance]
    Route -->|Committee Activity| CommSF[📋 Committee Step Functions<br/>Activity + document linkage]
    Route -->|Document Published| DocSF[📄 Document Step Functions<br/>Parse + classify + index]
    Route -->|Anomaly Signal| AnomSF[🚨 Anomaly Step Functions<br/>detect_voting_anomalies]

    VoteSF --> Persist[💾 Persist & Index Lambda]
    CommSF --> Persist
    DocSF --> Persist
    AnomSF --> Persist

    Persist --> DDB[(🟦 DynamoDB<br/>Hot single-table state)]
    Persist --> Aurora[(🟦 Aurora Serverless v2<br/>Voting history)]
    Persist --> OS[(🟦 OpenSearch Serverless<br/>Full-text + vector)]
    Persist --> Neptune[(🟦 Neptune Serverless<br/>Knowledge graph)]

    Persist --> Notify[📣 Amazon SNS / SQS<br/>Downstream fan-out]

    Notify --> DLQ{Delivery OK?}
    DLQ -->|❌ Fail| DeadLetter[(🪦 SQS Dead-Letter Queue<br/>Replay + alarm)]
    DLQ -->|✅ OK| Trigger([⚙️ Triggers Generation Pipeline<br/>See Bedrock flow])

    style EP fill:#e8f5e9
    style Stream fill:#e1f5ff
    style Bus fill:#e1f5ff
    style Route fill:#fff4e1
    style DLQ fill:#fff4e1
    style DeadLetter fill:#ffcdd2
    style Trigger fill:#d4edda

🧠 Bedrock-Backed Article & Intelligence Generation Flow

Managed generative pipeline using Amazon Bedrock foundation models, Knowledge Bases (RAG over the EP corpus + committed analysis artifacts), Bedrock Agents for agentic OSINT tool use, and Guardrails for neutrality, GDPR, and hallucination control. Output still flows through the deterministic aggregator — AI proposes content; humans remain accountable for what publishes.

sequenceDiagram
    autonumber
    participant EB as EventBridge
    participant SF as Step Functions
    participant Agent as Bedrock Agent
    participant KB as Bedrock Knowledge Base (RAG)
    participant FM as Bedrock Foundation Model
    participant GR as Bedrock Guardrails
    participant Agg as Deterministic Aggregator
    participant Rev as Human Editor
    participant S3 as S3 + CloudFront

    EB->>SF: Significant EP event detected
    SF->>Agent: Invoke OSINT generation workflow
    Agent->>KB: Retrieve EP corpus + analysis artifacts
    KB-->>Agent: Grounded context + citations
    Agent->>FM: Generate analysis proposal (model-agnostic)
    FM-->>Agent: Draft intelligence artifact
    Agent->>GR: Apply neutrality / PII / hallucination checks
    GR-->>Agent: Pass or block with reasons
    Agent->>SF: Return graded artifact + confidence
    SF->>Rev: Route to human review queue
    Rev-->>SF: Approve / request changes (accountable sign-off)
    SF->>Agg: Approved markdown artifact
    Agg->>S3: Render 14-language static HTML
    S3-->>EB: Publish event (article.published)

Bedrock Layer Responsibilities

Component	Role	Governance
Bedrock Foundation Models	Draft analysis (Claude, Nova; model-agnostic)	Routed by cost/quality/latency
Bedrock Knowledge Bases	Managed RAG grounding over EP corpus	Citations required; no ungrounded claims
Bedrock Agents	Agentic OSINT tool use / orchestration	Tool allow-list, least-privilege IAM
Bedrock Guardrails	Neutrality, PII/GDPR, hallucination control	Blocks biased or private-life content
Human Editor	Accountable approval before publish	Per AI Policy — no autonomous deploy

🔐 API Request Flow (Journalists / Researchers / API Consumers)

Authenticated, serverless API layered behind the static edge. Amazon Cognito federates identity; API Gateway (REST/WebSocket) and AppSync (GraphQL) front Lambda resolvers that read the polyglot data layer.

flowchart TD
    User([👤 Journalist / Researcher / API Client]) --> Edge[⚡ Amazon CloudFront<br/>Static edge + API routing]

    Edge --> WAF[🛡️ AWS WAF + Shield<br/>Rate limiting, OWASP rules]

    WAF --> Auth[🔑 Amazon Cognito<br/>User pools, federated auth<br/>JWT / OAuth2 tokens]

    Auth --> AuthZ{✅ Token Valid + Scoped?}
    AuthZ -->|❌ Denied| Reject[🚫 401/403 Response]

    AuthZ -->|✅ REST| APIGW[🌐 Amazon API Gateway<br/>REST + WebSocket APIs<br/>Authorizer + throttling]
    AuthZ -->|✅ GraphQL| AppSync[🔗 AWS AppSync<br/>GraphQL + subscriptions]

    APIGW --> Resolver[⚙️ Lambda Resolvers<br/>Business logic, least-privilege IAM]
    AppSync --> Resolver

    Resolver --> DDB[(🟦 DynamoDB<br/>Hot key-value + DAX cache)]
    Resolver --> Aurora[(🟦 Aurora Serverless v2<br/>Relational voting history)]
    Resolver --> OS[(🟦 OpenSearch Serverless<br/>Full-text + vector search)]
    Resolver --> Neptune[(🟦 Neptune Serverless<br/>Political knowledge graph)]

    DDB --> Compose[🧩 Compose Response<br/>Shape JSON / GraphQL payload]
    Aurora --> Compose
    OS --> Compose
    Neptune --> Compose

    Compose --> Trace[📊 CloudWatch + X-Ray<br/>Metrics, traces, CloudTrail audit]

    Trace --> Return([✅ Response to Client<br/>Cached at edge where safe])

    style User fill:#e8f5e9
    style Auth fill:#e1f5ff
    style AuthZ fill:#fff4e1
    style Reject fill:#ffcdd2
    style Return fill:#d4edda

💬 Natural-Language Query Over the Knowledge Graph

Conversational OSINT access. A user asks a question in natural language; Bedrock translates intent into graph + search queries against Neptune and OpenSearch, grounds the answer via Knowledge Bases, and returns a cited, neutral response — never an ungrounded opinion.

flowchart TD
    Q([💬 NL Question<br/>Which groups co-voted most<br/>on energy dossiers in 2028?]) --> Front[⚡ CloudFront + Cognito Auth]

    Front --> NLU[🧠 Bedrock Agent<br/>Intent + entity extraction<br/>Amazon Comprehend NLP]

    NLU --> Plan[🗺️ Query Planning<br/>Decompose to graph + search ops]

    Plan --> GraphQ[🔷 Neptune Serverless<br/>Gremlin/openCypher traversal<br/>MEP-group-dossier-vote edges]
    Plan --> SearchQ[🔎 OpenSearch Serverless<br/>Vector + full-text retrieval]
    Plan --> KBQ[📚 Bedrock Knowledge Base<br/>RAG over analysis artifacts]

    GraphQ --> Fuse[🧩 Evidence Fusion<br/>Join graph + search + RAG<br/>Attach source citations]
    SearchQ --> Fuse
    KBQ --> Fuse

    Fuse --> Generate[✨ Bedrock Foundation Model<br/>Compose grounded answer]

    Generate --> Guard{⚖️ Bedrock Guardrails<br/>Neutral? Grounded? PII-safe?}

    Guard -->|❌ Block| Safe[🚧 Safe Fallback<br/>Decline + show raw evidence]
    Guard -->|✅ Pass| Answer([📋 Cited Neutral Answer<br/>Links to dashboards + sources])

    style Q fill:#e8f5e9
    style NLU fill:#e1f5ff
    style Guard fill:#fff4e1
    style Safe fill:#ffcdd2
    style Answer fill:#d4edda

🔄 Event-Driven Continuous-Improvement Loop

Self-monitoring, human-governed. Quality and engagement signals feed model routing and prompt/template refinement. Per the AI Policy, model or prompt changes are proposals that require human approval and canary validation before promotion — there is no autonomous production deploy.

flowchart TD
    Observe[📊 Observe<br/>CloudWatch metrics<br/>Quality scores, citations<br/>User engagement] --> Lake[🪣 S3 Data Lake<br/>Glue catalog + Athena<br/>QuickSight BI]

    Lake --> Detect[🔍 Pattern Detection<br/>SageMaker anomaly/quality models<br/>What drives quality?]

    Detect --> Propose[💡 Improvement Proposals<br/>Model routing tweaks<br/>Prompt + template updates<br/>Guardrail adjustments]

    Propose --> Human{👤 Human Review<br/>AI Policy gate}

    Human -->|❌ Reject| Archive[🗄️ Archive + Document<br/>Rationale captured]
    Human -->|✅ Approve| Canary[🐤 Canary via Lambda Alias<br/>Weighted traffic shift<br/>Monitor X-Ray + alarms]

    Canary --> Result{Canary Healthy?}
    Result -->|❌ Regress| Rollback[↩️ Rollback Alias<br/>Revert weights, post-mortem]
    Result -->|✅ Improve| Promote[🚀 Promote<br/>Update routing + templates]

    Promote --> Observe
    Rollback --> Observe
    Archive --> Observe

    style Observe fill:#e8f5e9
    style Human fill:#fff4e1
    style Result fill:#fff4e1
    style Rollback fill:#ffcdd2
    style Promote fill:#d4edda

📊 Performance Targets

v2.0 — Static-Enhanced Targets

Metric	Target	Monitoring
Build-to-publish (full site)	<30 min per scheduled run	GitHub Actions timing
Dashboard dataset freshness	≤ daily (per EP feed cadence)	Provenance timestamps
Page load (LCP)	<2.0 s on CloudFront edge	RUM / Lighthouse CI
Analysis quality floor	All artifacts meet reference thresholds	Stage-C quality gate
Accessibility	WCAG 2.1 AA	axe-core CI
Languages	14 complete	Aggregator coverage check

v3.0+ — AWS Serverless Targets

Metric	Target	Monitoring
Event ingestion latency	<30 s (event → indexed)	CloudWatch + X-Ray
Breaking analysis (assisted)	<5 min event → reviewed draft	Step Functions metrics
API response (P95)	<200 ms	API Gateway / AppSync metrics
NL query (P95)	<3 s grounded answer	Bedrock + X-Ray traces
Static edge availability	99.99% (CloudFront + S3)	CloudWatch Synthetics
Guardrail block accuracy	>95% neutrality/PII enforcement	Bedrock Guardrails reports
Cost per 1k API calls	Within serverless budget envelope	Cost Explorer / Budgets

📚 Workflow Comparison: Current vs v2.0 vs v3.0

Process Step	Current (v1.0.x)	v2.0 (Static-Enhanced)	v3.0+ (AWS Serverless)
Event detection	Scheduled `gh-aw` cron	Scheduled cron (richer matrix)	EventBridge + Kinesis (<30 s)
Compute substrate	GitHub Actions runners	GitHub Actions runners	Lambda + Step Functions
Data fetching	EP MCP batch	EP MCP + WB/IMF enrichment	Streaming ingestion adapters
Analysis authoring	LLM → markdown	LLM + 51-template OSINT	Bedrock + KB RAG + Agents
Quality control	2-pass review	ICD 203 + Admiralty + 2-pass	Guardrails + human sign-off
Dashboards	In-article Chart.js/D3	Pre-computed party/landscape	Live AppSync/API Gateway data
Rendering	Aggregator → HTML	Aggregator → HTML	Aggregator (edge) + dynamic APIs
Data stores	Committed markdown/JSON	Committed artifacts + cache	DynamoDB · Aurora · OpenSearch · Neptune
Query	None (pre-rendered)	Client-side faceted filter	NL query over knowledge graph
Identity	None (public static)	None (public static)	Amazon Cognito federated auth
Delivery	S3 + CloudFront	S3 + CloudFront	CloudFront edge + serverless APIs
Observability	Actions logs	Actions logs + RUM	CloudWatch · X-Ray · CloudTrail

🕵️ Intelligence-Cycle Capability Flows (2026 → 2037)

The flows above cover analysis and delivery. These flows add the missing intelligence-cycle stages — direction (PIR), indications and warning, adversarial review, integrity analytics, and counter-FIMI — that turn the platform from an analysis pipeline into a full OSINT observatory. They are the process view of the capability roadmap in FUTURE_MINDMAP.md. Every flow ends at a human-accountability gate; none publishes autonomously.

Collection Management — Priority Intelligence Requirements Flow

flowchart TD
    REQ([📋 Intelligence Requirements<br/>standing + event-driven]) --> NAI[🎯 Define Named Areas of Interest<br/>contested dossier, fracturing group, election]
    NAI --> EEI[🧩 Derive Essential Elements of Information<br/>what must be known to answer]
    EEI --> TASK[🛰️ Task Agentic Collectors<br/>EP MCP, DOCEO, external, ASR]
    TASK --> COV{📊 Coverage vs Gap?}
    COV -->|Gap found| RETASK[🔁 Re-task Collection<br/>add source, widen window]
    RETASK --> TASK
    COV -->|Sufficient| MANI[🧾 Provenance Manifest<br/>collection-plan.json]
    MANI --> HAND([➡️ Hand to Processing + Analysis])

    style REQ fill:#e7f0ff
    style COV fill:#fff4e1
    style RETASK fill:#fff9c4
    style HAND fill:#d4edda

Indications and Warning — Tripwire Flow

flowchart TD
    WATCH([👁️ Watchlist Indicators<br/>cohesion, attendance, abstention, rhetoric]) --> BASE[📈 Compare to Rolling Baseline]
    BASE --> TRIP{🚨 Tripwire breached?}
    TRIP -->|No| LOG[🗒️ Log Baseline Drift<br/>suppress false alarms]
    TRIP -->|Yes| DRAFT[✍️ Draft Warning<br/>WEP band + evidence chain]
    DRAFT --> HUMAN{🧑‍⚖️ Human confirms?}
    HUMAN -->|No| LOG
    HUMAN -->|Yes| EMIT([📣 Graded Warning<br/>brief + alert + dashboard])
    EMIT --> CAL[🎯 Record for Calibration]
    LOG --> CAL
    CAL -.feedback.-> BASE

    style WATCH fill:#e7f0ff
    style TRIP fill:#fff4e1
    style HUMAN fill:#fff4e1
    style DRAFT fill:#fff9c4
    style EMIT fill:#d4edda
    style LOG fill:#eceff1

Multi-Agent ACH and Red-Team Flow

flowchart TD
    Q([❓ Estimative Question<br/>will dossier X pass? will coalition hold?]) --> HYP[🧠 Generate Competing Hypotheses<br/>minimum two, mutually exclusive]
    HYP --> EVID[🔗 Map Evidence to Hypotheses<br/>cited PUBLIC sources only]
    EVID --> ACH[🔀 ACH Matrix<br/>diagnostic evidence weighting]
    ACH --> RED[😈 Devil's Advocate Agent<br/>attack the leading hypothesis]
    RED --> KAC[🧾 Key Assumptions Check<br/>what would falsify this?]
    KAC --> SPLIT{🤝 Analyst agreement?}
    SPLIT -->|Dissent| RECORD[📝 Record Dissent<br/>preserve minority view]
    RECORD --> ADJ[🧑‍⚖️ Human Adjudication]
    SPLIT -->|Converged| ADJ
    ADJ --> EST([📊 WEP-Banded Estimate<br/>confidence + evidence chain])

    style Q fill:#e7f0ff
    style RED fill:#ffe0e0
    style SPLIT fill:#fff4e1
    style RECORD fill:#fff9c4
    style EST fill:#d4edda

Integrity and Counter-FIMI Detection Flow

flowchart TD
    PUB([🗂️ PUBLIC Sources<br/>declarations, register, roll-calls, media]) --> NORM[🧹 Normalize + Entity Resolve]
    NORM --> INT[🔍 Integrity Analytics<br/>lobby-to-vote, revolving-door overlap]
    NORM --> FIMI[🛰️ Counter-FIMI Detection<br/>DISARM TTP tagging, coordination signals]
    INT --> QUEST[❓ Sourced Question<br/>NOT an accusation]
    FIMI --> CTX[🧭 Neutral Context<br/>sourced fact vs spin]
    QUEST --> GATE{🧑‍⚖️ Human Review<br/>neutral, sourced, public-role?}
    CTX --> GATE
    GATE -->|Needs work| REFINE[🔧 Re-source / Re-frame]
    REFINE --> GATE
    GATE -->|Approved| OUT([✅ Published with Evidence Chain])

    style PUB fill:#e7f0ff
    style FIMI fill:#ffe0e0
    style INT fill:#ede7f6
    style GATE fill:#fff4e1
    style REFINE fill:#fff9c4
    style OUT fill:#d4edda

Boundary reminder. Integrity findings are sourced questions for journalistic review, never adjudicated accusations. Counter-FIMI is detection and context only — the platform never conducts influence operations and never targets individuals' private lives.

🔮 Visionary Workflow Roadmap: 2027-2037

The platform's workflows evolve from human-orchestrated gh-aw pipelines, through AWS-native serverless OSINT operations, toward increasingly autonomous — but always human-accountable — agentic intelligence. Three principles hold across the decade: model-agnostic routing (no single-vendor lock-in via Amazon Bedrock), neutrality and provenance by construction, and AI as a proposal generator with humans accountable for every publish (per the AI Policy).

AI Model Evolution — DevSecOps & Development Perspective

Year	AI Model	DevSecOps Capability Evolution
2026	Opus 4.6–4.9	🟢 AI-assisted code review, automated test generation, agentic CI/CD workflows
2027	Opus 5.x	🔵 Predictive vulnerability detection, intelligent dependency management
2028	Opus 6.x	🟣 Multi-modal security analysis (code + architecture + runtime), automated threat modeling
2029	Opus 7.x	🟠 Autonomous security pipeline orchestration, self-healing build systems
2030	Opus 8.x	🔴 Near-expert automated security review, AI-driven architecture validation
2031–2033	Opus 9–10.x / Pre-AGI	⚪ Autonomous secure development lifecycle management
2034–2037	AGI / Post-AGI	⭐ Transformative software engineering with built-in security assurance

Assumptions: major AI model upgrades occur annually; competitors (OpenAI, Google, Meta, EU sovereign AI) are evaluated at each release; the architecture accommodates potential paradigm shifts (quantum AI, neuromorphic computing). Full cross-perspective analysis lives in the Hack23 Information Security Strategy § AI Model Evolution Strategy; governance per AI Policy.

Phase 5: Autonomous Agentic Pipelines — Assisted (2027-2029)

Self-Directing OSINT Workflows: Bedrock Agents propose which analyses to generate based on parliamentary-activity signals and significance scoring; humans approve scope. AI decides what to draft, not what to publish.
Model-Agnostic Bedrock Routing: every generation task routed to the optimal foundation model (Opus 5.x, Amazon Nova, open-source) by cost, quality, and latency — swappable without rewriting pipelines.
Zero-Touch Drafting, Human Publish: routine summaries flow end-to-end to a reviewed draft automatically; a human editor remains the accountable gate before CloudFront publication.

Phase 6: Predictive Intelligence Workflows (2029-2032)

Anticipatory Frameworks: Step Functions pre-stage analytical scaffolds for upcoming votes and committee sessions from the EP calendar in Neptune.
Dynamic Workflow Composition: Bedrock Agents assemble optimal Step Functions graphs on demand by content type, urgency, and quality requirement.
Cross-Parliament Synthesis: serverless fan-out detects related legislative activity across multiple parliaments for comparative, neutral analysis.

Phase 7: Cognitive Workflow Intelligence (2032-2035)

Causal Analysis Pipelines: trace legislative impacts through economic (IMF/World Bank), social, and environmental indicators with cited evidence chains.
Adaptive Guardrails: Bedrock Guardrails dynamically tune to content sensitivity and regulatory context while holding the neutrality floor.
Global Event Correlation: Kinesis + Neptune correlate parliamentary events worldwide to surface emerging legislative trends.

Phase 8: AGI-Native Operations — Governed (2035-2037)

Continuous Democratic Monitoring: AGI-class systems monitor covered parliaments with real-time, evidence-cited analysis — under standing human oversight and audit (CloudTrail, Security Hub).
Natural-Language Workflow Definition: stakeholders describe desired analyses in plain language; systems design and execute optimal serverless workflows, subject to AI-Policy approval gates.
Self-Evolving, Human-Accountable Pipelines: workflows propose their own efficiency and quality improvements; promotion always requires human sign-off and canary validation — never autonomous production deploy.

AWS Lambda — https://aws.amazon.com/lambda/
AWS Step Functions — https://aws.amazon.com/step-functions/
Amazon EventBridge — https://aws.amazon.com/eventbridge/
Amazon Kinesis — https://aws.amazon.com/kinesis/
Amazon API Gateway — https://aws.amazon.com/api-gateway/
AWS AppSync — https://aws.amazon.com/appsync/
Amazon Cognito — https://aws.amazon.com/cognito/
Amazon DynamoDB — https://aws.amazon.com/dynamodb/
Amazon Aurora Serverless v2 — https://aws.amazon.com/rds/aurora/serverless/
Amazon OpenSearch Serverless — https://aws.amazon.com/opensearch-service/
Amazon Neptune Serverless — https://aws.amazon.com/neptune/
Amazon Bedrock — https://aws.amazon.com/bedrock/
Amazon SageMaker — https://aws.amazon.com/sagemaker/
Amazon CloudFront — https://aws.amazon.com/cloudfront/

Governance & Methodology

Document Status: ✅ APPROVED FOR PLANNING
Last Updated: 2026-05-31 (UTC) | Release: v1.0.1
Next Review: 2026-08-31 (Quarterly)
Classification: Public