About
About Phronesis
Forensic document analysis for journalists, researchers, legal professionals, and anyone holding institutions to account.
Mission
- Democratize forensic intelligence. The same analytical rigour available to large firms should be available to anyone with documents and a question.
- Make institutional accountability accessible to individuals. One person with the right methodology should be able to hold any institution to account.
- Prove systematic failure with statistical rigour, not anecdote. Patterns need numbers. Bias needs p-values. Conclusions need source citations. Opinion isn’t enough.
Built For
Journalists
Investigating institutional failure — surface contradictions, trace claim origins, quantify bias in source materials across multi-year investigations.
Legal Professionals
Handling complex multi-document cases — construct timelines, map entity relationships, and identify evidential gaps across hundreds of documents.
Self-Represented Litigants
Facing institutional opponents — level the analytical playing field with systematic methodology that produces court-ready, fully cited findings.
Researchers
Studying institutional behaviour — analyse how claims propagate across agencies, how authority accumulates through repetition, and how selective citation distorts conclusions.
Regulatory Complainants
Building evidence packages — generate structured complaints for Ofcom, ICO, LGO, HCPC, GMC, SRA with proper citation formatting and evidence packaging.
Architecture
Local-First Architecture
Your documents never leave your machine. Full SQLite database runs entirely on-device. No cloud dependencies, no telemetry, no phone-home. Your disk, your data. The application runs offline after installation — network access is only needed for optional AI features, and those use your own API keys.
Privacy by Design
Sensitive case materials stay under your control. AI features are optional and use your own API keys with no data retention on provider servers. No usage analytics. No document fingerprinting. No metadata collection. The platform knows nothing about you or your cases beyond what’s on your local machine.
Multi-Provider AI
Claude for deep reasoning and document analysis. Groq for speed-critical operations. Gemini for multimodal processing. Replicate for specialist models. Choose the right provider for each task, or run entirely without AI using the rule-based analysis engines.
Regulatory Ready
Pre-configured complaint templates for UK regulatory bodies: Ofcom (Broadcasting Code), ICO (UK GDPR), LGO (maladministration), HCPC (health and care professions), GMC (medical practitioners), SRA (solicitors). Each template structures evidence to the body’s requirements with proper citation formatting.
Design Philosophy
Evidence-Grade Output
Every finding cites [Doc:Page:Para] references. No claim exists without a traceable source. If the engine can’t cite it, it doesn’t output it.
No Hallucination Tolerance
AI-generated content is validated against source documents. Claims without source citations are rejected. The system distinguishes FACT (source-cited), INFERENCE (logically derived), and SPECULATION (hypothesis) explicitly.
Full Audit Trail
Every analytical step is logged and reproducible. From raw document to final finding, the entire chain of reasoning is preserved. Any finding can be traced backward to its source in seconds.
Human-in-the-Loop
Automation surfaces findings. Humans make decisions. The platform never generates conclusions — it generates evidence for humans to evaluate. Ambiguous results are flagged for review, not silently resolved.
Core Principles
- 1Open Source — MIT licensed. Full source available on GitHub. Every methodology is published. Every algorithm is inspectable. No black boxes, no proprietary methods, no “trust our process.” If you want to know how a finding was generated, read the code.
- 2Privacy First — Documents never leave your machine. No cloud processing. No telemetry. No data collection. No analytics. Your analysis is yours — not training data, not a product metric, not monetizable. Privacy isn’t a feature. It’s the architecture.
- 3Built by Practitioners — Every engine tested on real investigations with 1,000+ page document bundles. Methodology refined through actual case analysis, not theoretical exercises. S.A.M. and CASCADE were developed because existing approaches failed on real problems — and validated because they succeeded where those approaches didn’t.
Open Source & Transparency
- License: MIT — use it, fork it, modify it, contribute to it
- Source: Available on GitHub
- Methodology: Published and peer-reviewable — see Methodology
- Research: 47 articles across 13 categories — see Research Hub
- Standards: Every analytical method documented with academic foundations and validation studies