See what data brokers see when they look at you. Then choose who you actually are.
🪞 Try it live → claritymirror.vercel.app
Quickstart · The Problem · How It Works · Architecture · Roadmap · Contributing
Hundreds of data brokers — Acxiom, Experian, Oracle Data Cloud, LexisNexis, LiveRamp, Epsilon, and many others — hold detailed profiles about you. These profiles map your psychological pressure points: financial anxiety, health fears, status aspirations, impulse patterns. Advertisers use these profiles to target you with messages designed to bypass your judgment.
You have a legal right to see this data (CCPA, GDPR, state privacy laws). But even when you get it, it arrives as incomprehensible category codes and opaque segment IDs.
Clarity Mirror turns that data into a tool for self-awareness instead of manipulation.
Clarity Mirror combines two core features:
Surfaces the psychological pressure points your data profile reveals — the levers advertisers pull — and reframes them as prompts for mindfulness:
What they see: Credit-monitoring behavior pattern. Classified as financially anxious, responsive to urgency-based financial products.
Your reflection: You're being shown a version of financial reality designed to keep you slightly afraid. Pause before any financial decision that arrives with a countdown timer. The urgency is manufactured.
Maps where the attention economy is allocating effort against you — which categories of content and products are competing for your attention, and what psychological mechanisms they're exploiting:
| Category | Share | Mechanism |
|---|---|---|
| Financial Products | 28% | Targeting financial anxiety signals |
| Wellness & Supplements | 22% | Leveraging health search patterns |
| Lifestyle & Aspiration | 18% | Feeding the comparison engine |
You set personal intentions ("spend more mindfully," "reduce anxiety-driven decisions"), and the system flags when your data profile is being targeted in ways that conflict with those intentions.
┌─────────────────────────────────────────────────────┐
│ YOUR DEVICE │
│ │
│ ┌─────────────┐ ┌──────────────┐ ┌──────────┐ │
│ │ Data Broker │──▸│ Normalizer & │──▸│ Clarity │ │
│ │ Retrieval │ │ Schema │ │ Engine │ │
│ │ (CCPA/GDPR) │ │ Mapping │ │ │ │
│ └─────────────┘ └──────────────┘ └────┬─────┘ │
│ │ │
│ ┌─────────────┐ ┌──────────────┐ │ │
│ │ Intention │──▸│ Conflict │◂───────┘ │
│ │ Store │ │ Detection │ │
│ └─────────────┘ └──────┬───────┘ │
│ │ │
│ ┌──────▾───────┐ │
│ │ Reflection │ │
│ │ Surface │ │
│ └──────────────┘ │
│ │
│ ⚠ Raw data NEVER leaves this boundary │
└─────────────────────────────────────────────────────┘
│ (optional, model updates only)
▾
┌───────────────┐
│ Federated │ Shared model learns which
│ Aggregation │ reflections help people most
│ Server │ — without seeing anyone's data
└───────────────┘
Everything runs locally. Your broker data, ad platform exports, intentions, and analysis never leave your device. The optional federated learning layer only transmits model weight updates — never raw data.
The current prototype demonstrates the Vulnerability Mirror and Attention Audit with simulated broker data:
git clone https://github.com/your-org/clarity-mirror.git
cd clarity-mirror
# Open the prototype directly in your browser
open app/clarity-mirror.htmlNo build step, no dependencies, no server. It's a single HTML file.
clarity-mirror/
├── app/
│ └── clarity-mirror.html # Interactive prototype (self-contained)
├── assets/
│ ├── logo.svg # Logo mark
│ ├── icon.svg # Favicon / app icon
│ └── banner.svg # Social / README banner
├── docs/
│ ├── ARCHITECTURE.md # Detailed system architecture
│ ├── BROKER-CATALOG.md # Known data brokers & access methods
│ ├── DATA-SCHEMA.md # Unified attribute schema
│ └── PRIVACY-MODEL.md # Privacy & threat model
├── src/ # (coming soon)
│ ├── retrieval/ # Broker data request automation
│ ├── normalization/ # Format parsing & schema mapping
│ ├── analysis/ # Vulnerability & attention profiling
│ ├── reflection/ # LLM-powered reflection generation
│ └── federated/ # Optional FL aggregation layer
├── LICENSE
└── README.md
Automates exercising your data rights across brokers and platforms:
| Source Type | Examples | Access Method | Response Time |
|---|---|---|---|
| Data brokers | Acxiom, Experian, LexisNexis | CCPA/GDPR request (email, API, portal) | 1–45 days |
| Ad platforms | Google, Meta, TikTok, Amazon | Platform data export tools | Minutes–hours |
| Identity resolution | LiveRamp, The Trade Desk | CCPA request | 15–45 days |
| People search | Spokeo, BeenVerified, Whitepages | Opt-out portals | 1–14 days |
Maps broker-specific attributes to a unified schema:
Acxiom "P$Income_Discretionary_Amount" ──┐
Experian "Estimated Household Income" ──┼──▸ clarity.financial.income_bracket
Oracle "HH_INCOME_RANGE" ──┘
Classifies normalized data into vulnerability categories and attention targets using a combination of rule-based mapping (broker segments → IAB taxonomy → vulnerability categories) and an optional local LLM for generating personalized reflections.
If a community of users opts in, Federated Averaging trains a shared model that improves over time:
- Learns which reflection framings actually help people become more mindful
- Identifies emerging broker practices across the population
- Benchmarks individual data exposure against anonymized population statistics
- No raw data ever leaves any device — only model weight deltas are transmitted
1. Awareness, not anxiety. The language is "notice" and "pause," never "danger" and "warning." The interface includes breathing exercises and grounding elements. We're building a mirror, not an alarm.
2. Local-first, always. Your data is the most sensitive thing about you. It never leaves your device. Period. The federated layer is opt-in and transmits only model updates.
3. Honest about limitations. We show you what we can infer, not what we're certain about. Broker data is incomplete, sometimes wrong, and always a simplification. The mirror is imperfect — and that's worth reflecting on too.
4. Non-judgmental. Being tagged as "impulse buyer" or "status-seeking" isn't a character flaw. These are patterns that entire industries have spent billions learning to exploit. Seeing them clearly is the goal, not fixing them.
5. Calm technology. Periodic reflections, not constant notifications. The system should feel like a journal, not a feed.
- Interactive Vulnerability Mirror with simulated data
- Attention Audit visualization
- Intention setting and conflict detection
- Brand identity and open-source setup
- Automated CCPA request generation for top 10 brokers
- Google Takeout ad profile parser
- Meta data export ad interest parser
- Unified normalization schema (v1)
- Identity verification flow
- Rule-based broker segment → vulnerability mapping
- IAB Content Taxonomy integration
- Local LLM integration for personalized reflections
- Persuasion Weather Report (periodic summary)
- Federated Averaging infrastructure
- Opt-in community intelligence
- Reflection effectiveness model
- Population-level exposure benchmarking
- Mobile app (local-first, no cloud sync)
- Browser extension for real-time ad targeting awareness
- Integration with existing privacy tools (DeleteMe, Optery)
| Layer | Technology | Rationale |
|---|---|---|
| Prototype | Vanilla HTML/CSS/JS | Zero dependencies, runs anywhere |
| Data retrieval | Python | Best library ecosystem for web scraping, email automation, PDF parsing |
| Normalization | Python + SQLite | Local-only structured storage |
| Analysis | Rule engine + local LLM (Llama/Mistral via Ollama) | No data leaves the device |
| Federated learning | PyTorch + Flower | Mature FL framework |
| Frontend (v2) | SvelteKit | Lightweight, local-first friendly |
| Mobile | React Native or Capacitor | Cross-platform with local storage |
We need people with experience in:
- Privacy law & data rights — CCPA/GDPR compliance, broker request mechanics
- Data engineering — Parsing broker response formats (PDF, CSV, XML, JSON), schema normalization
- Platform integration — Google, Meta, Amazon, TikTok data export APIs
- ML/Federated learning — FedAvg implementation, differential privacy
- UX/Product design — Translating opaque data attributes into meaningful human reflections
- Security — Identity verification flows, secure local storage, threat modeling
- Fork the repository
- Pick an issue from the Issues tab, or open a new one
- Discuss your approach in the issue before writing code
- Submit a PR with a clear description of what it does and why
Please read CONTRIBUTING.md before submitting.
This project exists to help people see clearly and choose freely. We expect all contributors to treat each other with the same respect and non-judgment we bring to the product itself. See CODE_OF_CONDUCT.md.
Data brokers build a model of you optimized for one purpose: getting you to act in ways that benefit advertisers. That model — your impulses, vulnerabilities, habits, aspirations — is a portrait painted by people who don't care about you.
But the same portrait, if you can see it clearly, becomes a tool for self-knowledge. The shift is simple:
From "here's what they know about you so they can push your buttons" to "here's what they know about you so you can notice your own patterns."
The mirror doesn't change what it reflects. But looking in it changes you.
MIT — see LICENSE for details.
This is information, not identity. Breathe. Notice. Choose.