RuView: Multi-Source Room Calibration Push
Three stacked pull requests move RuView calibration from single-radio training toward per-source and joint multi-source candidates. All work stays opt-in and candidate-only, with frozen evaluation evidence and condition-specific metrics to judge readiness.
Duration: PT2M18S
Episode overview
This episode is a short developer briefing from RuView.
It explains recent repository work in plain language.
- Show: RuView
- Published: 2026-11-02T06:04:45Z
- Audio duration: PT2M18S
Transcript excerpt
This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.
It's November 2nd, 2026, and this is your RuView daily briefing.
All three pull requests push room calibration from single-radio training toward multi-source awareness, while deliberately leaving the live detector untouched.
First, stop mixing radios. Today the managed server learns its background field model from one selected radio only, so other bound radios prove they contribute but don't train independent models, and pooling them risks blending amplitude offsets. Pull request 2417 adds an opt-in, memory-only shadow bank that reuses…
Second, learn jointly but as a candidate. Pull request 2423, stacked on that work, integrates a separate managed multi-source candidate using real native radio observations with Aether representation learning and a room-presence head. The existing selected-source detector and the per-source bank remain as controls…
Finally, keep honest receipts. Pull request 2424 preserves private evaluation evidence tied to an exact frozen candidate, so results don't disappear when the runtime closes, and reports condition-specific metrics instead of hiding failures behind aggregate accuracy, without promoting the candidate to sensing…
What's next to watch is whether…
Nearby episodes from RuView
- Training Lineage Survives Resume
- Learner Publication, Zip Security Fix, and Fall Detection
- Encrypted Learning Recovery and Python CI
- Weekly Recap - Guarded Hardware, Fresher Sensing, and Joint Room Learning
- Persistent Calibration, Self-Contained Sensing, and Security Patch
- Signed Firmware and Boundary Hardening
- ESP32-C5 Guardrails and Orphan Helper Fix
- Stale Data, Repeat Triggers, and Hardware Guardrails