## Radiant Glow — Platform Update (March–April 2026)
Over the past month we've made substantial progress across Radiant Glow's intelligence pipeline, public-facing product, and platform integration. Here's a summary of what's been rolled out.
Intelligence Pipeline
**Training Event Detection & Suppression** — Radiant Glow now actively detects and suppresses training exercises. When training language is identified on a talkgroup ("this is a drill", "training training", etc.), event creation on that talkgroup is blocked for one hour. Admin corrections for training incidents trigger the same suppression. We're also tracking training event frequency per talkgroup to help identify channels that are primarily used for drills and may need to be excluded from monitoring.
**Cover Assignment Recognition** — The system now recognizes mutual-aid standby dispatches — where units are sent to stage at a firehouse while the home department handles a working fire. Previously these created false structure fire events at each firehouse address. The pipeline now correctly identifies this dispatch pattern and filters it out.
**Batched Dispatch Handling** — Some dispatch centers transmit multiple unrelated incidents in a single radio transmission. Escalation phrases like "people pinned" in one incident were incorrectly escalating all incidents in the batch. Escalation matching is now scoped strictly to each individual incident within a transmission.
**Simulcast Dedup** — The same dispatch often airs on multiple talkgroups simultaneously, and speech-to-text can produce different spellings of the same street on each channel. Radiant Glow now geocodes event locations and uses proximity matching to detect and merge duplicate events that resolve to the same physical location.
**Geocode Address Correction** — When the geocoder resolves a garbled street name to a real address, the event is updated with the corrected address rather than retaining the transcription error.
**Unit Normalization** — Speech-to-text frequently spells out unit numbers ("Engine Twenty Seven") or drops spaces ("Engine27"). The pipeline now converts these automatically, producing cleaner unit lists and better correlation.
**Firefighter Down Guard** — Refined mayday detection filtering to reduce false positives from noisy transcription while preserving detection of genuine firefighter distress signals.
IncidentSource Site
**Production Launch** — The Incident Source landing page and Radiant Glow authenticated portal are now live in production with a public homepage, authentication gate, and terms/privacy pages.
**Activity Sparklines** — The homepage now displays real-time sparkline charts showing signal, event, and major event volume trends.
**15-Minute Delay** — Public major event data is delayed by 15 minutes clearly labeled on the interface.
**Mobile Experience** — Significant improvements to the mobile layout including responsive navigation, tighter hero sections, and improved major events table rendering.
**Coverage Page** — Added coverage information and call-to-action links on the homepage.
Platform & Integration
**Shared Data Layer** — Created a shared data layer providing canonical functions for data access, major event queries, context resolution, and display formatting. This is now used by external consumers including the Broadcastify Listen alert page.
**Multi-Group Display** — Events now surface all talkgroups involved (not just the primary group) through the `groupDisplaysSeen` data path, giving a more complete picture of multi-channel incidents.
**Broadcastify Integration** — Major events from Radiant Glow are now displayed on Broadcastify Listen county pages and the alert page, bringing incident awareness directly into the existing listener experience.
**Transcription Statistics** — Added a transcription stats dashboard for monitoring ASR provider performance, queue health, and routing decisions.
**Real-Time Stats Collection** — Implemented backed statistics collection for signals, events, major events, and training detections with per-context and per-type breakdowns.