Methodology

This page explains how IranStrike collects, classifies, and presents real-time conflict data. Understanding our methodology helps you evaluate the information you see on the dashboard and its limitations.

Data Pipeline Overview

IranStrike operates a fully automated data pipeline that runs 24/7. Every three minutes, the system executes a four-stage process to transform raw Telegram messages into structured, classified events displayed on the tactical map.

Stage 1: Collection. The system scrapes the public web preview of six Telegram OSINT channels. Each channel's latest messages are fetched via HTTP, parsed from the HTML structure, and compared against a per-channel cursor (the ID of the last processed message) to identify new content. Only messages published since the previous cycle are forwarded for classification.

Stage 2: Classification. New messages are batched and sent to an AI language model (Claude Haiku 4.5 by Anthropic) with a structured prompt. The model extracts event type, severity level, geographic origin, involved entities, and a bilingual summary (English and Hebrew) from each message. Messages that do not describe military or security events are discarded. Content filters suppress commentary, opinion, and duplicate reporting.

Stage 3: Deduplication and Merge. Newly classified events are merged with the existing event database. The system checks for duplicate events — the same incident reported by multiple channels — using geographic proximity, temporal overlap, and event type matching. Deduplicated events are written to the database with timestamps, source attribution, and classification metadata.

Stage 4: Delivery. A pre-computed dashboard payload is generated containing the latest 250 events, the current spotlight headline, and metadata. This payload is cached at the edge (CDN) and served to all dashboard users with minimal latency. The frontend polls for updates every three minutes, matching the ingest cycle.

Source Selection and Vetting

Source quality is the foundation of data accuracy. IranStrike monitors six Telegram channels, each selected through a rigorous vetting process that evaluates four criteria:

Accuracy track record. The channel must have a documented history of reporting events that are later confirmed by official sources or multiple independent outlets. Channels that frequently publish unverified claims or speculative reports are excluded.
Editorial neutrality. State-controlled media, propaganda outlets, and channels with a demonstrated ideological agenda are excluded. This includes channels operated by or closely affiliated with government agencies of any party in the conflict.
Reporting density. The channel must publish military and security event reports with sufficient frequency and detail to contribute meaningfully to real-time monitoring. Channels that primarily publish commentary, analysis, or memes are excluded.
Technical accessibility. The channel must have its public web preview enabled on Telegram, allowing automated collection without requiring a Telegram API account or user authentication.

Current Sources

CIG Intel — Professional OSINT intelligence channel with neutral analysis. Covers military movements, satellite imagery analysis, and confirmed strike reports across the Middle East.
BNO News — Professional breaking news wire service. Known for speed and accuracy on confirmed events, with strict editorial standards against publishing unverified claims.
Aurora Intel — Neutral OSINT channel focused on aerospace, military aviation, and air defense events. Provides detailed technical analysis of weapons systems and intercepts.
OSINT Defender — Quantitative military OSINT with detailed event data including numbers of projectiles, damage assessments, and geographic coordinates when available.
Iran International — Independent Iranian news organization (based outside Iran) with credible conflict reporting. Provides unique coverage of events inside Iran that other English-language sources may miss.
Rocket Alert — Automated Israeli incoming threat alert system. Provides real-time notifications of incoming rockets, missiles, and drone threats detected by Israeli defense systems.

Excluded Sources

Several channels were evaluated and explicitly excluded for failing the vetting criteria. State-controlled outlets (Press TV, Sputnik International), channels with documented anti-Israel bias (Middle East Spectator, Quds News Network), and channels with poor signal-to-noise ratios (excessive commentary, memes, or spam) are not included in the monitoring pipeline. The exclusion list is reviewed periodically as channel quality changes over time.

AI Classification System

Each incoming message is classified by a structured AI prompt that extracts the following fields:

Event type: Strike, interception, drone launch, troop movement, naval incident, infrastructure damage, or other military activity.
Severity level: Low (routine military activity), medium (notable targeted operations), high (significant operations with regional impact), or critical (major escalations involving ballistic missiles or large-scale attacks).
Geographic origin: The country or region where the event occurred, mapped to one of six monitored zones (Israel, Iran, Lebanon, Syria, Iraq, Yemen).
Involved entities: The parties involved in the event, including state militaries, proxy forces (Hezbollah, Houthis, Iranian-backed militias), and defense systems.
Bilingual summary: A concise factual description in both English and Hebrew, stripped of editorial commentary.

The classification model (Claude Haiku 4.5) was selected after evaluating four candidate models on 179 live Telegram messages across 8 classification batches. It was chosen for the highest recall (event detection rate), best severity accuracy, and most reliable geographic attribution. Alternative models tested included GPT-4.1 Mini (58% recall), Gemini 2.5 Flash (14% recall), and GPT-4.1 Nano (13% recall) — none met the minimum quality threshold for deployment.

Severity Scale

The four-tier severity scale provides a consistent framework for evaluating the significance of each event:

Low — Routine military positioning, minor exchanges of fire, reconnaissance activity, or unconfirmed reports. Limited immediate impact on civilian populations.
Medium — Confirmed targeted strikes, successful drone interceptions, localized rocket barrages, or verified troop deployments. May affect specific areas but contained in scope.
High — Significant military operations involving multiple weapons systems, large-scale air defense activations, or coordinated attacks across multiple fronts. Potential for regional escalation.
Critical — Major escalations involving ballistic missiles, mass drone swarms, strategic infrastructure strikes, or events that cross established red lines. Significant risk of wider conflict.

Infrastructure Monitoring

In addition to event tracking, IranStrike monitors internet and cellular network availability across conflict zones. Infrastructure disruptions are frequently correlated with military activity — either as direct targets of strikes, as a result of power grid damage, or as deliberate shutdowns by governments during operations.

Internet connectivity data is sourced from the IODA project (Internet Outage Detection and Analysis), operated by the Georgia Institute of Technology. IODA aggregates data from BGP routing tables, active probing, and network telescopes to detect country-level and regional internet disruptions in near real-time.

Network performance data is supplemented by Cloudflare Radar, which provides traffic volume and anomaly detection from Cloudflare's global network of data centers. Both sources are polled every 30 minutes and displayed in the infrastructure vitals panel on the dashboard.

Data Freshness and Staleness

IranStrike is designed to present information as close to real-time as technically feasible. The typical end-to-end latency from an event being reported on Telegram to appearing on the dashboard is 3 to 6 minutes, depending on when the report falls within the ingest cycle.

The system monitors its own pipeline health. If the most recent ingest cycle is more than 9 minutes old (three times the normal interval), the dashboard displays a "DELAYED" badge in the header instead of "LIVE." This indicates that data may be stale due to a pipeline interruption, upstream source outage, or infrastructure issue. The badge automatically returns to "LIVE" once fresh data is available.

AI-Generated Headlines and Summaries

Two additional AI-generated features provide context beyond raw event tracking:

Spotlight headline: Every 15 minutes, the system reviews recent events and selects the most significant one for prominent display at the top of the dashboard. The headline includes a brief contextual explanation of why the event matters.
Key developments: A narrative summary of the most important events from the past 24 hours, generated during each ingest cycle. This provides a quick overview for users who want to understand the current situation without scrolling through individual events.

Both features use AI to synthesize information from classified events. They are clearly labeled as AI-generated and should be cross-referenced with the underlying event data in the live feed.

Limitations and Caveats

IranStrike is an automated monitoring tool with inherent limitations that users should understand:

Source dependency. All event data originates from third-party OSINT channels. If a channel publishes an inaccurate report, it may appear on the dashboard before being corrected or removed by the source.
Classification errors. AI classification is not perfect. The system may occasionally misclassify event types, assign incorrect severity levels, or attribute events to the wrong geographic region. Classification accuracy is continuously monitored and the prompt is refined over time.
Coverage gaps. Events that are not reported by any of the six monitored channels will not appear on the dashboard. Covert operations, events in areas with limited media access, and rapidly evolving situations may be underrepresented.
Latency. The 3-minute ingest cycle means events are never truly real-time. Critical events may take 3 to 6 minutes to appear. For immediate safety decisions, always rely on official civil defense systems such as the Israeli Home Front Command (Pikud HaOref).
No prediction. IranStrike monitors and reports events that have already occurred. It does not predict future events, assess the probability of escalation, or provide strategic forecasting.

Important: IranStrike is an informational tool for civilian awareness. It is not a substitute for official government warnings, military intelligence, or emergency services. In an active security situation, always follow instructions from your local civil defense authorities. Do not make safety decisions based solely on information from this dashboard.