FLAREData Sources
Seventeen real-time magnetometer networks, three L1 solar wind monitors, and direct ACE/DSCOVR telemetry — ingested in parallel, cross-validated before any alert fires.
Multi-Source Ingestion with Cross-Validation
Every alert begins with raw magnetometer readings from 17 globally distributed ground-based networks, fused with L1 solar wind data from ACE and DSCOVR satellites. Before any processed output leaves the pipeline, a voting consensus across independent source sets must agree within ±8% on Kp magnitude. Single-source alerts — the norm at legacy providers — are rejected outright. This cross-validation step alone reduces false positives by 73% compared to single-feed systems.
Data sourced from independent benchmark studies and internal validation logs. Last updated Feb 2026.
Forecast Models
Three independently validated ensemble models — ENLIL solar wind propagation, custom Kp neural regression, and CME trajectory Monte Carlo — run in parallel and reconciled before any forecast is issued.
Ensemble Modeling with Hindcast Validation
The Kp forecast pipeline runs three models simultaneously: a physics-based ENLIL solar wind propagation model, a gradient-boosted neural network trained on 22 years of NOAA historical data, and a Monte Carlo CME trajectory ensemble with 1,000 particle simulations per event. Final forecasts are a skill-weighted average of all three outputs. Every model is validated monthly against 6-month hindcast windows. Skill scores are published in the Accuracy Report.
Data sourced from independent benchmark studies and internal validation logs. Last updated Feb 2026.
Alert Latency
From magnetometer anomaly detection to webhook delivery in under 90 seconds. Measured, published, and contractually guaranteed at the 95th percentile.
Sub-90-Second Detection-to-Delivery Pipeline
Latency is measured from the moment an upstream sensor crosses the alert threshold to the moment the first webhook fires at the client endpoint. The pipeline is fully event-driven: no polling loops, no batch jobs. Alert generation runs on dedicated compute isolated from the analytics workload, ensuring that dashboard traffic never delays operational alerts. P95 latency is contractually guaranteed at <90 seconds for Kp threshold breaches and <4 minutes for CME impact predictions.
Data sourced from independent benchmark studies and internal validation logs. Last updated Feb 2026.
Integration
REST API, webhook callbacks, MQTT for IoT environments, email digests with structured metadata, and SMS/voice escalation for on-call engineers. All delivery methods are active simultaneously.
Multi-Channel Redundant Delivery
Critical operational alerts should never depend on a single delivery channel. Flare delivers every alert simultaneously across all configured channels, with per-channel acknowledgment tracking. If a webhook endpoint fails to return HTTP 200 within 10 seconds, the system escalates automatically to the backup channel. Integration takes under 20 minutes for most operational teams: a REST API key, one webhook endpoint, and one email address is all that's required to go live.
Data sourced from independent benchmark studies and internal validation logs. Last updated Feb 2026.
Pricing & Parallel Test
You've seen the pipeline, the models, and the benchmarks. Now run both systems simultaneously for 14 days. Zero commitment. Full operational data.
Single sector, up to 5 alert recipients.
- 3 data streams included
- 72-hour Kp forecasts
- API + email delivery
- 99.5% uptime SLA
- Email support (4h SLA)
Multi-sector, unlimited recipients.
- All data streams
- 7-day probabilistic forecasts
- All delivery channels
- 99.95% uptime SLA
- Dedicated operations contact
- Custom threshold configuration
On-premise option, custom integrations, white-label.
- Everything in Mission-Critical
- On-premise deployment option
- Custom model calibration
- White-label API available
- Dedicated SRE team
Run a Parallel Test
Receive both your current provider's alerts and Flare's simultaneously for 14 days. Compare latency, accuracy, and coverage in your actual operational environment.
Download the Accuracy Report
48 pages. 24 months of hindcast validation across 3,200 geomagnetic events. Model skill scores, false positive rates, and latency distributions by event class.