Threshold · Frontier Cascadia

A local-first entrance-monitoring prototype with a native iPhone camera demo, a synthetic radio-lab workflow, and a cinematic browser presentation.

Built for Frontier Cascadia. Research/demo software—not a certified security system or emergency service.

What is here

• iPhone camera demo: SwiftUI interface with on-device person detection in a fixed doorway region. Keep the phone stationary, well lit, and in the foreground. The camera path is implemented; independent physical acceptance remains pending.
• Rehearsal / TEST: explicitly synthetic input for demonstrating calibration, arming, a latched alert, acknowledgement, and disarming. Synthetic results are not evidence of physical sensing.
• Local radio backend: Python/FastAPI, authenticated controls, stream-health checks, recording/replay, and a browser dashboard.
• Cinematic presentation: an offline-capable browser walkthrough with clearly labelled synthetic visuals and recorded TEST screenshots.
• Experimental firmware: Arduino Nano ESP32 / ESP32-S3 CSI sender and receiver sources, with upstream attribution.

Important limits

Real received Wi-Fi CSI and physical motion detection are not verified. Transmit/ACK success does not establish reception, occupancy, exact-zone tracking, direction, or falls. The planned camera-free path requires compatible router and sensing hardware; arbitrary home-router compatibility is not claimed.

Calls are disabled/dry-run by default. No emergency delivery, background alert reliability, or production security certification is claimed. A stream fault is not an all-clear. The camera demo is separate from the shared radio alarm.

Run the local TEST dashboard

Requires Python 3.10+ and uv.

git clone https://github.com/ethaenall/frontier-cascadia.git
cd frontier-cascadia
uv sync --locked
uv run python -m threshold --mode test

Use the local address printed by the process. Pair using its locally stored control token; never put tokens in URLs, screenshots, or commits. Keep the server on localhost or a trusted private network. Do not start a second collector against an active hardware session.

Demo sequence: Enable sound → Calibrate → Arm → labelled synthetic motion → Acknowledge → Disarm. Browser audio needs a user gesture and is not guaranteed in background tabs.

uv run --locked pytest -q

Tests are included. This public packaging pass did not rerun the full test suite or physical-device checks.

iPhone

Open ios/Threshold.xcodeproj in Xcode. Choose your own signing team for a physical device. The domain package is in ios/Domain; app-host and UI tests are also included. Camera permissions and real-device behavior require your own verification.

Browser presentation

uv run python scripts/present.py

See static/cinematic/README.md for options and labels. This presentation is not a live sensor feed.

Privacy and safety

This source release excludes local .env credentials, recordings, control tokens, hardware backups, Xcode user state, and private test evidence. .env.example contains empty credential placeholders. Optional Browserbase/Daytona tools are developer utilities, not required sensing services; using cloud tools may create billable resources.

Credits and disclosures

• User-directed Prime Agent and coding assistants helped with setup, implementation, debugging, tests, and documentation. Physical testing and release decisions remain with the project lead.
• Three.js is MIT licensed; see LICENSE.three. Recorded UI images are our synthetic TEST interface captures; provenance is in assets/PROVENANCE.json.
• Espressif CSI reference material and license records are in firmware/provenance, including Apache-2.0 notices.
• Browserbase, Daytona, and Mobbin were development tools/references. The historical Browserbase screenshot shows its public marketing site, not a product feature.

Public source availability does not grant a new license to third-party material. Retained dependency licenses govern their respective code.