Capabilities
Current system-level reference for the Political Strategy Intelligence Platform. What is operational, what is validated, and where the real command surface exists today.
Capabilities
22
Integrated across strategy, CRM, messaging, social media, web, memory, monitoring, strategist evaluation, recovery, governance pre-check, statistical analysis (R 4.5.3), and context-hygiene coaching
Books ingested
441
225,872 chunks fully embedded. All 441 sources verified and trusted for retrieval.
Embedded chunks
225,872
Local index built and retrieval-usable (2.5GB, qwen2.5:3b)
Eval result
12.7/13
Average score for the local+Sonnet strategy pipeline over 20 questions
  • Full production pipeline live: web form → KV queue → corpus retrieval → GDELT live context → Sonnet synthesis → master strategist evaluation (premium) → result delivery.
  • 441 books, 225,872 embedded chunks across 9 strategic schools. GDELT live-context enabled by default.
  • Master strategist tier built: 10 profiling scenarios, deliberation capture, cognitive profile synthesis, profile-conditioned evaluation, attribution + compensation.
  • Simulation engine live: sim.professionalopinions.com.au. Enrollment at join.professionalopinions.com.au. /startsim triggers SMS from Marcus Webb. Sprint / deep / slow-burn pacing.
  • AZHA evidence portals live: submit.professionalopinions.com.au (community) and evidence.professionalopinions.com.au (individual). R2 file storage, KV metadata.
  • Decision engine Stages 1–4 operational: problem spec → corpus retrieval (query expansion + reranking + geo/era/credibility weighting) → modernity translation layer → Sonnet synthesis (problem-type frames + actor constraints + adversarial stress test) → iterative refinement → tactics with 16 wired execution handlers.

Capability strengths

Corpus depth & retrieval 92%
441 books, 225k+ chunks, 9 schools. Local cosine retrieval + reranking.
Synthesis quality 99%
Claude Sonnet synthesis. Chain-of-thought reasoning, problem-type specific frames, actor constraint analysis, adversarial stress test in every output.
Live context (GDELT) 75%
Real-time geopolitical signals. Phase 1 — coverage depends on query quality.
Master strategist layer 70%
Full pipeline built: profiling → capture → profiles → evaluation → attribution. Awaiting first enrollments.
E2E pipeline & delivery 95%
Web form → KV → engine → result delivery. Enrollment, attribution, CRM sync operational. Signal CLI two-way messaging. WhatsApp send/receive. MacroDroid notification routing (5 platforms).
Observability & protection 88%
Grafana/Prometheus/Loki stack. SHA-256 manifests. Watchdog every 20 min.
Source diversity 88%
9 schools covered. 441 books. School-diversity weighting, geography boost, era weighting, query expansion, and cross-encoder reranking all active.
Preference learning 30%
Feedback captured and wired to preference weight updates. Accepted/rejected options adjust school and tactic weights per user. Model calibration active.

Where we are strong

Proven
Synthesis quality is validated at 12.7/13 across 20 blind evaluations. Every output has 3+ options, 2+ schools, recommendations, tradeoffs, and first-7-day operational plans.
Proven
Full end-to-end pipeline from web form to strategy delivery, with user attribution, CRM sync, and GDELT live context — tested and operational.
Proven
Master strategist architecture fully designed and built: 10 profiling scenarios, cognitive capture pipeline, multi-perspective evaluation, attribution tracking, and compensation model.

Known gaps

In progress
No enrolled master strategists yet. Profiling scenarios and capture pipeline are ready — first sessions will activate the premium tier.
Pending
Preference feedback loop wired: accepted/rejected/bookmarked strategy options update preference weights per school and tactic type. Polling integration live via Pollster Portal → decision engine export. Source diversity balancing (author caps, school quotas) now active.

Executive map

The system is best understood as perception, judgment, and execution linked by persistent memory and operational tooling.

Perception

  • User requests (web UI, Telegram, file drop)
  • Political corpus (441 books, 225k+ chunks)
  • GDELT live geopolitical signals
  • Master strategist cognitive profiles
  • CRM records and tester profiles
  • Memory, wiki, and dashboard state

Judgment

  • Problem structuring with strategic modes
  • Local retrieval and source-pack building (top-60 → final-12)
  • GDELT live context injection
  • Sonnet synthesis with tradeoff and recommendation logic
  • Master strategist evaluation: profile-conditioned multi-perspective analysis
  • Evaluation harnesses (12.7/13 average)

Execution

  • Reports and PDFs
  • Attribution tracking and compensation
  • Emails and broadcasts
  • CRM schema and workflows
  • Website deployment
  • Backups, dashboards, and recovery

What is built and operational

21 integrated capabilities across strategy, intelligence, social media, security, analytics, and communications. All marked LIVE are operational and validated.

Political Strategy Intelligence Assistant

LIVE

Doctrine-backed political strategy synthesis from 441 books and live geopolitical signals.

What it does
  • Corpus retrieval with query expansion and reranking
  • GDELT live-context injection per case
  • Sonnet synthesis with problem-type frames
  • Iterative refinement loop
  • 16 wired tactic execution handlers
Current state
Strength: Full E2E pipeline live, 12.7/13 eval score

Limit: Preference feedback active, not yet recalibrating model

Master Strategist Evaluation

LIVE

Named human expert profiles condition every premium recommendation before delivery.

What it does
  • 10 profiling scenarios per strategist
  • Cognitive profile synthesis
  • Profile-conditioned evaluation
  • Attribution and compensation tracking
  • Governance compliance scoring
Current state
Strength: Architecture complete, weighting system live

Limit: Awaiting first enrollee

Polling Intelligence Platform

LIVE

Transforms serial private polling waves into a live queryable asset connected to the strategy engine.

What it does
  • Multi-wave canonical variable registry
  • Longitudinal trend analysis with confidence bands
  • AXIOM coherence mapping by topic
  • Auto-export to decision engine on ingest
  • Stripe Connect invoice lifecycle
Current state
Strength: Full portal live — upload, client view, payments

Limit: No real polling data ingested yet

AXIOM — Autonomous Political Engagement

LIVE

Two pseudonymous social accounts running independent progressive and conservative campaigns.

What it does
  • Content pipeline from curated AU news feeds
  • Thinker engagement — likes, follows, replies
  • Quote-retweet layer (8/day cap)
  • Topic scanner for topical entry points
  • Panel weighting for audience quality scoring
Current state
Strength: Both campaigns operational, posting daily

Limit: Twitter API restricts cold replies on non-participant threads

AXIOM Panel Weighting

LIVE

Audience quality scoring correcting for bots, volume bias, and account maturity in all AXIOM analytics.

What it does
  • Authenticity scoring (bot probability)
  • Volume and account maturity weighting
  • Network tier classification (nano→mega)
  • Engagement depth tracking
  • Composite panel_weight per interactor
Current state
Strength: All fields operational, applied to aggregates

Limit: Coherence mapping to polling pending first real dataset

Social Messaging — SMS, Signal & WhatsApp

LIVE

Two-way messaging via SMS, Signal, and WhatsApp from the Pixel 10 Pro.

What it does
  • SMS send/receive via Android SMS Gateway
  • Signal send/receive via signal-cli
  • WhatsApp auto-reply via MacroDroid routing
  • Contact lookup table wired
  • Inbound flagging for Doron attention
Current state
Strength: All three channels operational

Limit: WhatsApp reply requires MacroDroid macro active on phone

Attio CRM

LIVE

Contact and relationship management for all platform clients and leads.

What it does
  • Contact and company records
  • Lead capture from intake forms
  • Note and activity tracking
  • API integration for automated record creation
Current state
Strength: Operational, intake pipeline wired

Limit: No automated lead scoring or enrichment yet

Canary Protocol

LIVE

Steganographic fingerprinting for leak attribution — every exported briefing carries an invisible recipient mark.

What it does
  • Unicode variation selector encoding
  • 2M+ unique fingerprint IDs
  • Survives print, scan, copy-paste, OCR
  • Maps to delivery record in local ledger
  • Proprietary — no cloud storage
Current state
Strength: Encoding system built and tested

Limit: Not yet integrated into automated export pipeline

Website & Intake Infrastructure

LIVE

Cloudflare Pages hosting with full autonomous deployment, contact forms, and client intake.

What it does
  • Cloudflare Pages deploy via Wrangler CLI
  • Contact form Worker with KV lead storage
  • Client intake form with prefill endpoint
  • Custom domains with DNS management
  • Privacy-first web analytics
Current state
Strength: Full autonomous control, all sites live

Limit: Domain registration requires one manual CAPTCHA step

Memory & Knowledge System

LIVE

Persistent memory across sessions via daily notes, curated long-term memory, and semantic vector search.

What it does
  • Daily notes (memory/YYYY-MM-DD.md)
  • Curated long-term MEMORY.md
  • LanceDB vector search (qwen2.5:3b)
  • Wiki vault with backlinks and dashboards
  • Auto-capture and auto-recall
Current state
Strength: Multi-layer system operational

Limit: Wiki compile occasionally lags on large session dumps

Observability & Recovery

LIVE

Prometheus, Grafana, and Loki monitoring with OneDrive backup and documented disaster recovery.

What it does
  • Prometheus metrics collection
  • Grafana dashboards (v12.4.2)
  • Loki log aggregation
  • OneDrive automated backup sync
  • Disaster recovery protocol documented
Current state
Strength: Stack live, metrics flowing

Limit: Rollback is manual — no automated snapshot-restore

AZHA Antisemitism Classifier

LIVE

Local ML classifier identifying antisemitic content across the IHRA definitional framework.

What it does
  • Stage 1 binary detection
  • Stage 2 theory-of-harm classification (7 categories)
  • Community evidence portal (submit.professionalopinions.com.au)
  • Individual portal (evidence.professionalopinions.com.au)
  • R2 file storage with KV metadata
Current state
Strength: Classifier operational, both portals live

Limit: 4 gold-standard examples — needs ~100 for paper-quality accuracy

Stripe Payment Integration

LIVE

Full payment lifecycle for platform services including invoicing, collection, and disbursement.

What it does
  • Stripe Connect for multi-party payments
  • Invoice creation and client payment links
  • Automatic fund disbursement to pollsters
  • GST flag per client
  • Full audit trail per transaction
Current state
Strength: Operational via Pollster Portal

Limit: No customer-facing refund or dispute dashboard

GDELT Live Context

LIVE

Live geopolitical signals injected into every strategy synthesis from the GDELT global event database.

What it does
  • Real-time event feed by geography and theme
  • Automatic injection into synthesis prompts
  • Phase 1: media and conflict signals
  • Configurable by case geography
Current state
Strength: Live and injected by default on all cases

Limit: Coverage depends on query quality and GDELT source availability

Governance & Risk Pre-Check

LIVE

Mandatory four-domain risk gate on every AI recommendation — BLOCK/FLAG/CAUTION/CLEAR before delivery.

What it does
  • Reputational, legal, financial, political domains
  • Seven Australian statutory frameworks
  • Configured hard red lines per client
  • BLOCK outputs forensically logged, never delivered
  • Adviser compliance tracking over time
Current state
Strength: Gate operational on all AI outputs

Limit: Statutory annotation engagement not yet tracked per adviser

Political Strategy Simulation Engine

LIVE

Scenario-based political crisis simulations with SMS delivery to participants.

What it does
  • Scenario schema and facilitator engine
  • SMS delivery via Pixel 10 Pro
  • 36-Hour Window crisis scenario built
  • Sprint/deep/slow-burn pacing modes
  • Enrollment at join.professionalopinions.com.au
Current state
Strength: Engine live, SMS delivery operational

Limit: One scenario built — content library needs expansion

Statistical Analysis Engine (R 4.5.3)

LIVE

Full R analytics stack integrated via Python bridge — effect sizes, survival, MRP, and causal inference.

What it does
  • Effect sizes (Cohen's d, OR, RR, NNT)
  • Kaplan-Meier and Cox PH survival analysis
  • Linear mixed models (lme4)
  • MRP stack (multilevel + poststratification)
  • Causal inference (MatchIt, estimatr, ri2)
Current state
Strength: 35 packages installed, ARM64 native

Limit: No automated report generation yet

Security & Autonomous Monitoring

LIVE

14-point nightly scanner with auto-remediation, aligned to NIST CSF 2.0 and ASD Essential Eight.

What it does
  • Nightly scan at 03:00 AEST
  • Auto-corrects permission drift
  • Kills rogue services on detection
  • Checks credentials, ports, auth, SSH
  • Alerts Doron only on actionable findings
Current state
Strength: 0 open alerts, scanner active nightly

Limit: ISO 27001 certification targeted 2027

Voice Cloning

LIVE

Local XTTS-v2 voice synthesis using Doron's voice — private, on-premise, no external API.

What it does
  • XTTS-v2 local model
  • 5 training samples, ~107 minutes of voice data
  • WAV output at 22050Hz
  • Fully local — no external API calls
Current state
Strength: Synthesis operational

Limit: Batch generation only — no real-time streaming

Google Cloud Integration

LIVE

Google Calendar, Drive, Gmail, Search Console, PageSpeed, and Meet via authenticated APIs.

What it does
  • Calendar event creation with Meet links
  • Drive file access and sharing
  • Gmail send/receive
  • Search Console data pull
  • PageSpeed and Custom Search APIs
Current state
Strength: All APIs authenticated and operational

Limit: OAuth tokens require periodic refresh

Print & Physical Output

BETA

Print-ready PDF and document generation for real-world briefing delivery.

What it does
  • PDF rendering pipeline
  • Report and deck generation
  • Physical briefing output paths
Current state
Strength: Rendering paths operational

Limit: Printer device control not explicitly integrated
Security & Compliance

Built secure. Verified daily.

Aligned to NIST CSF 2.0 and the ASD Essential Eight. Autonomous 14-point security scan runs nightly at 03:00 AEST. Every credential vaulted and permission-enforced. Every AI output carries a forensic audit trail. ISO/IEC 27001 certification targeted for 2027.

0 Open Alerts Daily Scan Active Last Audit: 22 Apr 2026
Full security posture →
🔑
Credential Vault
All secrets vaulted, permission-enforced, git-clean
🤖
Auto Monitoring
14-check scanner, nightly, auto-remediates
🏠
Local-First AI
Sensitive data never leaves the server
📋
Audit Trail
Forensic log on every AI output, BLOCK enforced