RuView: Critical Hardware Compatibility and Accuracy Fixes
A major firmware fix resolves zero packet yield on display-less ESP32 boards that was causing widespread deployment failures. Parallel improvements tackle model accuracy issues and complete documentation retractions from inflated performance claims.
Duration: PT2M5S
Transcript
Good morning, this is RuView for June 3rd, 2026.
The team just shipped critical fixes for what appears to be a widespread hardware compatibility crisis that was undermining the project's credibility.
The biggest story is firmware compatibility. Pull request 906 identified why users with display-less ESP32 boards were getting zero packet yield - the CSI collector was filtering for management frames only, but these boards needed data frame capture. This wasn't just a minor bug. Users were flashing pre-built binaries, seeing no presence detection, and filing fraud accusations. The team immediately rebuilt all firmware variants in PR 908, bumping to version 0.6.7 to get working binaries into users' hands.
Parallel to the hardware crisis, there's been a systematic accuracy cleanup. PR 907 fixed eigenvalue-based person counting that was reporting "10 persons when 1 present" on noisy CSI data - the count wasn't bounded to match other estimators on the same single-link data. PR 919 improved error handling when users passed the wrong model format, replacing cryptic magic number errors with actionable diagnostics. And PR 920 caught a silent bug where the export model flag was writing placeholder sine-wave weights instead of actually training models.
The documentation work tells an important story about project honesty. PR 916 completed the retraction of "100% presence accuracy" claims across all remaining docs. The team already explained this was measured on single-class overnight data where a constant "yes" predictor would score 99.98%, and they've replaced it with an 82.3% held-out temporal metric throughout the codebase.
Looking ahead, the firmware refresh should resolve the immediate deployment issues, while the accuracy bounds and model diagnostics should reduce false fraud reports. The team appears to be prioritizing deployment reliability over feature expansion right now.
That's RuView for today.