Every component uses free tiers: Groq (100K tokens/day), Supabase (500MB), Vercel (100GB bandwidth), GitHub Actions (unlimited public repo minutes). Zero vendor lock-in. Zero credit card required. Zero maintenance cost.
Once deployed, GNI runs 2x daily without any human action. The frequency controller adjusts run intervals based on world escalation score. The adaptive pipeline responds to breaking threats automatically without any trigger from a human.
Mission Control monitors 6 health dimensions every 30 minutes. Telegram alerts fire before failures cascade. The heartbeat detects anomalies and triggers corrective pipelines autonomously. The system fixes itself.
Core intelligence pipeline. RSS collection from 25 sources — injection detection across 66 patterns — MD5 deduplication — geopolitical funnel scoring — 3-temperature AI analysis with confidence intervals — Three Pillar domain reports — Supabase persistence — Telegram notification. The sacred run that never fails.
Quadratic MAD Protocol. Four AI agents (Bull, Bear, Black Swan, Ostrich) debate the latest report across 3 rounds with Arbitrator coaching after each round. Produces BULLISH/BEARISH/NEUTRAL verdict with confidence score, action recommendation, blind spot warning, and GPVS prediction.
Escalation monitoring and NYSE alert system. Reads latest escalation score, compares to previous pipeline run, detects signal divergence between Pipeline and MAD, fires NYSE open/close alerts, triggers adaptive pipeline when escalation delta exceeds threshold.
Emergency fresh analysis when world escalation spikes. CRITICAL level = 0 Groq calls (cached data only). HIGH = 4 calls. LOW = 19 calls. Frequency controller adjusts run interval autonomously: CRITICAL=30min, HIGH=2h, ELEVATED=4h, MODERATE=6h, LOW=12h.
25 global news sources, up to 20 articles each, 400-500 raw articles per run. MD5 deduplication prevents duplicate event reporting within 6-hour rolling windows.
Every article scanned across 7 security layers: Unicode normalization, source credibility scoring, context boundary detection, Named Entity Recognition, Groq hardened JSON output, output sanitization, SHA-256 audit chain for every intelligence item.
Geopolitical significance scoring (0-20 points), top-N selection with source diversity enforcement, 3 independent AI runs at temperatures 0.1/0.3/0.7, t-distribution confidence interval (t=4.303, n=3, alpha=0.05) for every sentiment score.
After verify_date passes, actual SPY movement measured against prediction direction. Correct predictions: source weight x1.1 via EMA. Wrong predictions: weight x0.9. Weights bounded 0.5 to 2.0. Creates compounding accuracy advantage.
Heartbeat monitors escalation delta every 30 minutes. Mission Control checks Supabase connection, report freshness, quota ceiling, source health, pipeline recency. Telegram CRITICAL/WARNING alerts with specific action recommendations.
The Dev Console perspective provides the technical evidence for GNI's most significant engineering claim: a production-grade autonomous AI intelligence system running at .00/month. This is not a prototype or a demo — it is a live system processing real news, making real predictions, and verifying them against real market outcomes.
The 4-pipeline architecture, the self-healing heartbeat, and the Mission Control monitoring layer collectively demonstrate L7 autonomy: the system manages itself, responds to world events, and maintains quality — all without human intervention after initial deployment.