Argos MonitorWhat Argos Monitor is
Argos Monitor is an automated open-source news and crisis monitor covering country media ecosystems across domestic press, local-language diaspora outlets abroad, and international reporting. Instead of asking a language model to summarize the news—which risks hallucinated claims, flattened nuance, or unattributed spin—Argos shows only original headlines and verbatim pull-quotes verified character-for-character against the source article.
Seven non-negotiable rules
- No source, no statement: Every headline and quote on the page links directly to the article it came from. Nothing is paraphrased or synthesized.
- Zero LLM in the core pipeline: Clustering, topic classification, ranking, and quote extraction are 100% deterministic (multilingual embeddings + rules). Any post-processing summary layer must strictly verify quotes via substring match against stored article text.
- Independent outlet counting: State wire reprints and syndicated copies are detected by text similarity (≥ 0.95 cosine similarity) and counted once, so 20 reprints of one wire dispatch never masquerade as 20 independent confirmations.
- Tri-scope perspective balance: Every story displays its distribution across domestic licensed media, local-language diaspora/exile press, and international reporting, alongside editorial alignment, political leaning, and factuality.
- Official vs. reported (Official / Reported): Government and ruling-party outlets are explicitly tagged so official statements are immediately distinguishable from independent reporting.
- Live methodology synchronization: The About page and system specifications are updated in the same change whenever country desks, analytical layers, or pipeline stages are modified.
- Zero emojis across the broadsheet: Built for diplomatic and institutional press review. No emojis are permitted anywhere in the UI, alerts, badges, logs, or documentation. External RSS emojis are stripped upon ingestion.
- Priority election coverage: National, presidential, parliamentary, and local elections, ballot results, voter turnout, referendums, and official campaign outcomes receive top policy weighting (1.0), deterministic ranking priority boosts, and prominent dispatch placement in diplomatic intelligence briefs.
How the 11-stage pipeline works
A scheduled GitHub Actions runner executes the pipeline every 3 hours and publishes static JSON/JS bundles to Cloudflare Pages:
01 · IngestPolls 760+ RSS/Atom and Google News feeds across 87 country desks via curl.
02 · ExtractFetches full article text with Trafilatura for domestic and accessible sources.
03 · NormaliseApplies Unicode NFC normalisation (crucial for Vietnamese diacritics and Khmer script).
04 · RelevanceFilters out sports/lifestyle sections and requires country mentions on world feeds.
05 · EmbedEncodes headlines and leads locally using ONNX Runtime (
intfloat/multilingual-e5-small).06 · ClusterGroups articles into stories via agglomerative cosine clustering + second-pass centroid merge, and flags syndicated reprints.
07 · ClassifyAssigns each story to one of 6 policy topics by cosine similarity against bilingual topic anchors.
08 · Policy & RankScores stories by independent outlet count, scope diversity, recency, and policy themes (
themes.yaml).09 · Excerpts & VerifySelects informative sentences and asserts each quote is an exact substring of the stored article text.
10 · Offline TranslationTranslates non-English representative headlines into English using CPU-friendly MarianMT sequence-to-sequence models with entity glossaries, caching results in SQLite while keeping original text authoritative.
11 · Signals & ExportMerges GDACS/USGS/NASA EONET/UN Peace & Security/FCDO/US State Dept alerts, auxiliary social signals, filtered Google Trends, and multi-subreddit discussions into static bundles.
Verbatim Evidence & Quote Verification Methodology
To adhere strictly to Rule 1 (No source, no statement), Argos strictly rejects generative LLM synthesis. Synthetic language models frequently invent diplomatic claims, confuse names, or smooth over critical contradictions between rival governments. Every fact displayed on Argos traces to an exact quote:
01 · Substring Assertion Engine
Pull-quotes are extracted directly from full text obtained via Trafilatura. Every candidate sentence is checked with a strict assertion:
assert quote in stored_article_text. If extraction, diacritic processing, or formatting altered a single character or punctuation mark, the quote is discarded.
02 · Extractive Multi-Source Summarisation
Story summaries are purely extractive. They consist of the representative headline plus 2–3 verbatim excerpts selected from different reporting outlets (e.g. domestic, diaspora, international). Each quote displays its publisher, publication time, and links directly to the source URL.
03 · Syndication & Wire De-duplication
State wire agencies (such as VNA in Vietnam, MTI in Hungary, ČTK in Czechia, or Yonhap in South Korea) frequently distribute identical press copy across dozens of domestic commercial news sites. Argos computes pairwise cosine similarity across lead sentences; copies with similarity ≥ 0.95 are collapsed as syndicated reprints.
04 · Independent Outlets vs. Gross Article Volume
Story prominence is measured by the count of distinct, independent publishers covering the event, rather than gross article count. Twenty reprints of a single state wire dispatch count as 1 independent outlet, preventing coordinated media blitzes from artificially exaggerating diversity scores.
Offline Machine Translation Methodology
Argos covers 87 country desks published in over 20 languages (including Vietnamese, Lao, Hungarian, Ukrainian, Arabic, and Russian). To enable fast scanning without relying on external cloud APIs or violating data sovereignty, Argos uses dedicated offline translation models:
01 · MarianMT Sequence-to-Sequence Models
Representative headlines are translated using lightweight Marian neural machine translation models (
Helsinki-NLP/opus-mt-*) running locally on CPU. These models require zero external API keys, zero token fees, and operate entirely air-gapped within the ingestion runner.
02 · Geopolitical Entity Alignment Glossaries
To prevent historical translation artifacts in low-resource language pairs—such as older models translating Sino-Vietnamese political names into Chinese pinyin (e.g. rendering General Secretary Tô Lâm as Song Lin) or literal translations of state institutions—Argos enforces curated post-processing glossaries.
03 · Persistent SQLite Translation Cache
Translated headlines are stored in the
translation database table keyed by article_id. Once translated, a headline is resolved in O(1) time, keeping incremental pipeline refreshes light on CPU memory and completing within seconds.
04 · Original Language Authoritative
In accordance with Rule 4, translation is purely an aid for machine reading and speed. The original foreign-language headline remains authoritative and can be inspected at any time via the
[Show original] toggle. Extracted quotes are never translated server-side, preventing translation drift from contaminating primary evidence.
Source Analysis Methodology
Argos evaluates every monitored publication across four structural dimensions to deliver objective media-intelligence analysis across democratic, hybrid, and authoritarian media systems:
01 · Alignment (Power & Ownership Structure)
Measures structural relationship to political authority: State-controlled (statutory news agencies and ministry organs), Pro-government (private commercial media editorially allied with the ruling coalition), Public broadcaster (chartered public-service media with statutory editorial safeguards), Independent (commercial, cooperative, or investigative press free of government ownership), and Opposition (dissident, diaspora, or exile press).
02 · Editorial Leaning (Political Spectrum)
Calibrated using international registries (Media Bias/Fact Check, AllSides, Reuters Institute): Left, Center-Left, Center, Center-Right, and Right. In state-directed media ecosystems (e.g. Vietnam, China), outlets are classified as State-aligned to reflect official diplomatic messaging rather than domestic electoral competition.
03 · Factuality (Verification & Correction Standards)
Assesses sourcing transparent attribution, primary corroboration, and correction practices: Very High (stringent primary-source corroboration, e.g. major wires), High (credible mainstream reporting with professional editorial oversight), Mostly Factual (factual core but prone to partisan framing or sensationalism), and Mixed (vulnerable to state censorship, selective omission, or unverified claims).
04 · State-linked (Official Corroboration Flag)
A binary indicator (Yes / No) designating statutory organs and state-funded broadcasters. When any state-linked source covers a story cluster, it is tagged Official to distinguish state messaging from independent reporting. When over 70% of corroborating coverage originates from state media, the story is stamped State-Dominated.
Crisis, Conflict & Security Methodology
To adhere strictly to Rule 1 (No source, no statement), Argos rejects speculative heuristic keyword tagging of general news headlines for violent events. Instead, the Disasters and Conflict & Security monitors rely on an authoritative dual-track model:
Track 1 · Verified External Sensor & Advisory Layer
Integrates structured feeds from international sensor networks and statutory foreign ministry alerts: Natural Hazards & Disasters via Global Disaster Alert and Coordination System (GDACS), US Geological Survey (USGS M4.5+ seismic events), and NASA EONET; Conflict, Terrorism & Security Advisories via United Nations Peace & Security incident updates, US Department of State (elevated Levels 2–4), UK Foreign, Commonwealth & Development Office (FCDO), and the International Crisis Group.
Track 2 · Monitored Defence & Security Press Coverage
Presents actual story clusters deterministically classified into the
defence_security domain by section feeds and multilingual embeddings. Stories report their own headlines and verified quotes without synthetic combat badges, ensuring administrative military handovers and diplomatic cooperation are never misclassified as kinetic violence.
Track 3 · Religious & Sectarian Sensitivity Analysis
Enveloped strictly within the
defence_security domain to prevent sectarian blind spots or denominational advocacy. Monitored security stories and advisories are cross-referenced across four confessional traditions (Judaism & Jewish communities, Christianity & church bodies, Islam & Muslim minority enclaves, and regional historic faiths / sacred sites) across four structural vectors: targeted asymmetric extremism, compound disaster vulnerability, legal/state coercion, and resource-communal armed clashes. Stories are tagged with clear typographical stamps (REL SENSITIVITY: TRADITION) linked to verifiable verbatim quotations.
Delta Analysis Methodology ("What Changed?")
To evaluate news cycle momentum rather than static cumulative volume, Argos compares the active monitoring window (e.g. last 24h or last 7d) against the immediately preceding contiguous period of equal length without LLMs or synthetic claims:
01 · Story Velocity & Emergence
Deterministic movement states classify stories by publication timestamp: fresh emergence (NEW, zero articles in the prior period), accelerating reporting (SURGE +ΔN, volume gain ≥ 3 articles and velocity multiplier ≥ 1.75x), and fading cycles (COOLING, volume decline ≥ 3 and velocity ≤ 0.40x).
02 · Scope Escalation ("Going Global")
Detects when a strictly domestic controversy is picked up by international wires (INTL PICKUP) or foreign-language diaspora outlets (DIASPORA PICKUP), flagging emerging diplomatic escalation.
03 · Macro Agenda Shifts
Calculates percentage-point attention shifts (ΔS) across the 6 core policy categories (e.g.
Economics ↑ +23.8%, Politics ↓ -19.4%) displayed in the country Agenda Movement Bar with quick one-click filtering.
04 · State Media Narrative Drift
Calculates shifts in state-affiliated media share (Δstate_share ≥ +30%) to flag official state narrative campaigns (STATE SURGE) or sudden state withdrawal from an active controversy.
05 · Bilateral & Multilateral Delta Cross-Comparisons
In the Joint & Multilateral Affairs monitor, cross-references coverage across partner countries to calculate bilateral momentum multipliers, net article growth (ΔN), new breakout stories, and primary diplomatic policy domains.
Policy Relevance & Foreign-Office Ranking Model
Standard news aggregators optimize for social clicks and sensationalism, elevating celebrity controversies, local crimes, and traffic accidents. Argos Monitor is engineered specifically for foreign-ministry desk officers, diplomatic missions, and macroeconomic policy analysts:
01 · Six Core Policy Categories
Every ingested story is mapped into one of six core policy domains: Politics & Governance (internal state personnel, constitutional affairs, party directives), Diplomacy & Foreign Policy (bilateral summits, treaties, multilateral forums), Economy & Trade (macroeconomic data, trade corridors, sovereign debt, foreign direct investment), Defence & Security (armed forces procurement, joint military drills, border security, deterrence), Society & Rights (judiciary, civil rights, labor, religious affairs), or Environment & Energy (critical minerals, grid infrastructure, climate disasters).
02 · Hybrid Semantic Classification
Categorisation uses a hierarchical dual-stage classifier: structured URL paths and RSS section feeds (e.g.
/thoi-su/, /the-gioi/, /wirtschaft/) provide authoritative native categorization, cross-referenced with multilingual dense embeddings (intfloat/multilingual-e5-small) evaluated against bilingual policy description anchors.
03 · Policy Weighting Formulation
Story ranking combines coverage volume with policy domain importance defined in
themes.yaml: Score = Volume × (0.4 + 0.8 × Weight) + 3 × Weight. Diplomacy, trade, and strategic security carry top weight (0.9–1.0), elevating pivotal institutional developments over viral tabloid trivia.
04 · Local Incident Suppression
Individual traffic accidents, local crime trials, weather reports, and celebrity gossip are assigned low baseline weights (0.05–0.2). An individual murder trial with 15 domestic articles will consistently rank below a bilateral trade corridor agreement with only 4 articles, ensuring high signal-to-noise ratio in daily executive briefs.
05 · Standing Priority for Elections & Referendums
National, presidential, parliamentary, and municipal elections, ballot results, voter turnout, referendums, and official campaign outcomes receive top policy weighting (1.0), ranking priority boosts, and prominent ELECTION PRIORITY placement in diplomatic intelligence briefs across all countries.
Auxiliary Signals & Real-Time Context
High-interest crisis desks require real-time visibility into fast-breaking events before formal newspaper editorial cycles publish articles. Argos integrates auxiliary real-time feeds while maintaining a strict evidentiary cordon:
01 · Public Telegram Channel Web Previews
Ingests publicly accessible channel previews (
t.me/s/channel) without API credentials, phone numbers, or bot tokens. Monitors official regional emergency channels and trusted conflict observers across Eastern Europe, the Middle East, and the Caucasus.
02 · Open Bluesky Feeds (AT Protocol)
Polls public AT Protocol endpoints for verified regional correspondents, security analysts, and diplomatic observers without authentication. Incoming texts undergo full Unicode NFC normalization and emoji stripping.
03 · Multi-Subreddit & Search Surges
Monitors public Reddit discussion clusters across regional diaspora communities and tracks Google Trends regional search momentum surges to identify emerging public interest shifts.
04 · Strict Segregation & Unverified Status
Auxiliary signals provide valuable context but do not meet the corroboration threshold of verified journalism. They are displayed in a dedicated sidebar drawer, stamped as unverified context, and are never mixed into verified story clusters or used to calculate official coverage metrics.
Infrastructure, Privacy & Zero-Cost Architecture
Argos Monitor is designed to be completely free, privacy-preserving, transparent, and durable over decades of operation without commercial API dependencies or hosting costs:
01 · Local-First & Cloud Automation
The pipeline runs entirely on standard CPU hardware without requiring GPUs or cloud server infrastructure. An automated GitHub Actions workflow executes every 3 hours, updates the SQLite database cache, and finishes in approximately 10 minutes.
02 · Static Edge Delivery
The entire site is static (plain HTML, modern CSS, vanilla JavaScript, zero frameworks). The pipeline outputs pre-calculated JSON bundles deployed to Cloudflare Pages global edge network, delivering instant sub-100ms response times worldwide.
03 · Zero Telemetry & Absolute Privacy
Argos does not set cookies, does not load third-party analytics trackers, does not require user accounts, and makes zero tracking calls. Reader search queries, combobox selections, and drawer toggles remain strictly within the browser's local memory. Outbound links enforce strict no-referrer isolation (
rel="noopener noreferrer" and Referrer-Policy: no-referrer) alongside HTTP Content-Security-Policy and HSTS enforcement so monitored foreign media outlets cannot observe or correlate reader traffic.
04 · European Broadsheet Typography
Designed with a pure white canvas, serif masthead, classic double rule, and vertical-bar story blocks inspired by historic European broadsheets (Le Monde, Neue Zürcher Zeitung, Financial Times). In accordance with Rule 7, zero emojis are allowed across the interface.
Monitored Outlets & Source Directory
Every publication Argos monitors across all 87 country desks, classified by reporting scope (domestic, diaspora abroad, international), editorial alignment, political leaning, factuality, and state affiliation.
Auxiliary Signals: In addition to authoritative press feeds, high-interest crisis desks ingest open, unauthenticated public Telegram channel previews (t.me/s/) and open Bluesky feeds (AT Protocol) as unverified real-time context.
Loading sources…
Author & open source
Argos Monitor is designed and maintained by Károly Gergely. Source code and methodology documentation are published under the MIT License at github.com/gergkar-gif/argos-monitor.