GTI Danville PA — Longitudinal Fleet Health GTI Danville PA — Salud Longitudinal

Clusters of silent radios in one room usually indicate a local coverage or shielding problem (bad hop path, obstruction, dead repeater). Clusters spanning multiple rooms indicate a shared gateway or firmware issue. Silenciosos agrupados en una sala = problema local; en múltiples salas = gateway o firmware compartido.

P1 — 4 silent radios

• Flower 6: 4 radios — 4×node

P2 — 6 silent radios

• Flower 11: 3 radios — 2×node, 1×repeater
• Flower 9: 2 radios — 2×node
• Flower 8: 1 radios — 1×repeater

P3 — 1 silent radios

• Flower 18: 1 radios — 1×node

P4 — 13 silent radios

• Flower 25: 3 radios — 3×node
• Flower 30: 3 radios — 2×node, 1×repeater
• Flower 35: 3 radios — 3×node
• Flower 31: 2 radios — 2×node
• Flower 28: 1 radios — 1×node
• Flower 32: 1 radios — 1×repeater

Other — 5 silent radios

• Hallways: 4 radios — 4×repeater
• (no room): 1 radios — 1×node

P1 — Immediate (this week)P1 — Inmediato

1. Field-check the 29 silent radios (see ranked table above). Attend to backbone repeaters first — one silent repeater takes its full downstream tree offline. Verificar en sitio los 29 radios silenciosos.
2. Reposition / add a repeater near Flower 20 to fix H1110005900 hop (LQI 13% vs UI 44%). Reubicar/añadir repetidor en Flower 20.
3. Re-audit repeater density in P4 (F22–F40) — the 2024-05 6-repeater install predates the P4 expansion. Auditar densidad de repetidores en P4.

P2 — Preventive (4–8 weeks)P2 — Preventivo

1. Battery-replacement wave for Nodes at batt_p50 < 3.4 V (see Worst Battery table). Reemplazo de baterías <3.4V.
2. Firmware sweep — confirm all repeaters on latest Wirepas firmware. Firmware de repetidores al día.

P3 — MonitoringP3 — Monitoreo

• Weekly re-run of this pipeline (both current-2d snapshot and 2-day room stitch for the trailing bimonthly window). Reejecutar semanalmente.
• Alert when any room has ≥3 silent radios simultaneously. Alerta si sala ≥3 radios silenciosos.

Inventory: GET /public_api/devices/?facility=3003&pagination=0 — 1515 device records, classified 734 radios + 689 sensors after excluding retired room 11368. Radios = mesh participants (gateway·sink·repeater·node); sensors = probes (TEROS-12·TEROS-One·ATMOS-14). Only Node + TEROS-12 are battery-powered — everything else is PoE. Inventario: 1515 registros → 734 radios + 689 sensores.

Current snapshot: /public_api/devices/<id>/chart/ with start/end covering the trailing 2-day window. Called for all 734 radios, 3 workers with 0.12s inter-call sleep to stay under the ~2–3 worker 429 threshold documented in aroya-public-api-device-telemetry. Gateways & sinks return HTTP 200 with empty series (architectural artifact — same behaviour observed across every AROYA facility). Sensors are NOT called on this endpoint — their metrics live in the room chart. Instantánea actual: devices/<id>/chart/ para los 734 radios, 3 workers, sleep 0.12s.

Longitudinal (117 weeks): /public_api/rooms/<id>/chart/ for each active room × 14 bimonthly 2-day windows (2024-04-14 → 2026-06-14). This endpoint returns zone- and room-level metric series with a hard 2-day cap per call — matches the stitched pattern in aroya_14d_stitch_pull.py (memory aroya-stitch-pull-no-browser). 43 rooms × 14 windows = 602 stitched calls, ~ 41 min total runtime. Longitudinal (117 semanas): rooms/<id>/chart/ × 14 ventanas bimestrales. 602 llamadas stitcheadas.

Historical per-radio telemetry: The /devices/<id>/chart/ endpoint is empirically confirmed to return only the rolling recent ~2 days regardless of the requested start — historical windows return HTTP 200 with empty series. This is an architectural limit of the public API; per-radio 117-week battery/RSSI/LQI trending requires either the SPA /dataReadings/ endpoint (localStorage-scoped token, not accessible unattended) or a batch/history endpoint added to public_api. This report substitutes zone-level data delivery density from stitched room chart as the longitudinal proxy — since every zone is downstream of exactly one Node radio, zone-delivery density tracks mesh health at the Node granularity. Telemetría histórica por radio: El endpoint devices/<id>/chart/ solo devuelve ~2 días recientes; usamos densidad longitudinal por zona (proxy directo de salud de la radio parenta).

Thresholds: Signal p05 < −85 = CRIT, −75 to −85 = WARN; LQI p05 < 100 = CRIT; battery p50 < 3.0 V = CRIT, 3.0–3.4 = WARN. Umbrales: señal p05 < −85 CRIT; LQI p05 < 100 CRIT; bat < 3.0V CRIT.

Data caveats: Gateways & sinks show as "unclassifiable" (not silent) because they emit no device-chart series by design. TEROS-One (ter15) shares its base H343 serial with its parent Node — same physical device, listed twice in the API (one radio side, one sensor side). Total device-record count (1515) is the raw API count; dash-2 replacements were checked (none present in current inventory). Advertencias: Gateways/sinks emiten cero series por diseño; TEROS-One comparte serial con Node H343; total 1515 records raw sin duplicados dash-2.