No revision or outcome ledger; treat as an assessment or scenario.
Weighting Model
Regional Brief / 6H
NO CURRENT BRIEFING
Regional Threat Map
Stable
Elevated
Critical
Intelligence Stream
BNTI Methodology
1. Data Collection
The system monitors RSS/Atom feeds from regional news outlets, wire services, and specialized publications
for each target country. Sources include local newspapers in native languages, international news agencies,
and regional security-focused media. Data is fetched on a 2-hour workflow with bounded retries, caching,
and no-publish safeguards when feed coverage collapses.
2. LLM Attribution
Each headline is sent to OpenRouter Free for final country attribution and final category
selection. The model must decide whether the headline directly concerns one or more of Türkiye's seven
border neighbors or is IRRELEVANT.
Category weights are fixed after the LLM chooses the class, so scoring remains deterministic even though
the model owns the semantic judgment.
military_conflict: +8.0
terrorism: +7.0
political_instability: +4.0
border_security: +5.0
humanitarian_crisis: +3.0
diplomatic_tensions: +2.5
trade_agreement: -2.0
neutral: 0.0
4. Index Calculation
Per-country scores are based on avg(weight) across LLM-attributed events, then normalized to
1-10 with a saturating exponential curve. Composite BNTI = weighted average across all monitored countries.
5. Thresholds
Stable: 1.0-3.9
Elevated: 4.0-6.9
Critical: 7.0-10.0
6. Regional Briefing
A separate LLM-generated regional brief summarizes the most problematic border-neighbor events from the
last completed 6-hour window. The brief is fixed across the interface and refreshes at most four times per
day.
7. Publish Gate
The dashboard only publishes when feed coverage, LLM attribution, and summary generation all pass
validation. Failed or thin runs exit cleanly without overwriting the last good public snapshot.
