Chunks waiting for workers
Checked in last 12 hours
Jobs finished this cycle
Total identities issued
What This Is
The Civil Awareness Warrior is the public coordinator for the Strategic Assessment Atlas β a research project mapping zoning, governance, and environmental factors across all 3,143 U.S. counties.
Volunteers contribute idle computer time to fetch, verify, and stage public data from county GIS servers, municipal code libraries, and election portals. All work is broken into small, auditable chunks. No judgment, no scraping of private data β only public records, fetched politely.
What Workers Do
π Examination
Re-run a parser over a staged data blob and report the measurements. Deterministic β the same input always produces the same output.
π₯ Ingest
Download all pages of a county's GIS service (zoning layers, parcel data) following a frozen fetch plan, then stage the pages as compressed blobs.
π°οΈ Alt-Source Probe
Mechanical probes of candidate GIS endpoints: enumerate services, count features, test pagination, check for null codes. No judgment β just structured reconnaissance.
πΊοΈ State Walk
Walk a slice of a municipal code library (Municode, American Legal, etc.), extracting district titles and slug keys per a fixed recipe.
π Search
Run fixed query plans against search APIs (Brave, Bing) to locate election results, zoning portals, or certified PDFs. Results graded by source class.
π§ Classify (opt-in)
Grade a search-result set against a rubric with a small model running on YOUR machine. Output is proposal-grade only β it ranks hints for the operator and never enters verified data; only deterministic probes do that.
Privacy & Transparency
β What We See
- Your platform (OS/arch) and client version
- Chunk type capabilities you offer
- If you opt into local-model grading: the model family, quant and context size you declared β nothing else about your machine
- Heartbeat timestamps while a lease is active
- Structured results (manifests + blob hashes)
β What We Never See
π Your Control
- Run only the chunk types you choose (
--capabilities) - Limit concurrency (
--concurrency 1..8) - Schedule windows via your OS (cron, Task Scheduler, launchd)
- Dry-run mode (
--dry-run) stages locally, submits nothing - Local grading is opt-in: run a small open-weights model on your machine, point at a server you already run, or connect your own API key β search-result snippets leave your machine only if you choose a cloud endpoint
- Stop anytime β in-flight chunks requeue for others
π‘οΈ Supply-Chain Safety
- Native desktop app β no runtime dependencies beyond your OS
- Docker image:
FROM scratch, non-root user - ed25519-signed contracts verified before execution
- Outbound allowlist pinned per chunk (no open proxy)
- Blob uploads are Worker-mediated: warriors POST bytes to the coordinator, which stages them β volunteers never hold storage credentials
- Source code auditable on request; contracts are ed25519-signed at issuance (installer binaries currently ship unsigned)
Get Started
Download
Grab the Civil Awareness Warrior installer for your platform from the v0.1.0 release page:
Verify your download against the release's SHA256SUMS. Installers are not code-signed yet β macOS Gatekeeper and Windows SmartScreen will ask for a click-through confirmation on first launch (expected; see the release notes).
Register
Registration is currently suspended β the coordinator is not accepting new warriors while upload and verification hardening lands. Existing registered warriors keep working. Check back later.
Run
Install and launch Civil Awareness Warrior β it lives in your system tray. Once registered, open the tray menu and press Start: the app polls for leases, executes chunks, heartbeats, and uploads blobs through the coordinator (Worker-mediated β no storage credentials on your machine). Dry-run mode stages locally and submits nothing, so you can inspect exactly what a chunk would do before it runs.
Configure (Optional)
Everything is configured in the app's Settings window:
- Coordinator β endpoint URL and the ed25519 contract public key (production defaults are built in)
- Local model β none (default), auto-configure a bundled model + llama-server sidecar, or point at any OpenAI-compatible endpoint you run (llama.cpp, LM Studio, ollama) β press Check now to verify it before saving
- Concurrency β parallel work slots
- Schedule β limit work to an active window (e.g. evenings or overnight)
- System β launch at login
Requirements
β Minimum
- 64-bit CPU (x86_64 or ARM64)
- 512 MB RAM available
- 1 GB free disk for staging
- Outbound HTTPS (port 443)
- Linux (x86_64) / macOS / Windows 10+
β Recommended
- 2+ CPU cores
- 2 GB+ RAM
- 5 GB+ free disk
- Unmetered / high-cap internet
- Always-on or scheduled uptime
π Network
- No inbound ports required
- Outbound to coordinator + chunk target hosts
- Politeness enforced: per-host request budgets signed into each contract (default β€60 req/min)
- No VPN/proxy needed (residential IP preferred for some GIS)