How DoomRadar works

1. Collection and grouping

The worker collects headlines, source URLs and timestamps. Multilingual semantic similarity groups items that may concern the same event. This is a candidate-generation step, not confirmation that all linked stories belong together.

For events already selected for analysis, the worker attempts to read up to three public source pages from different domains. Usable article text is cached; a limited selection of passages is supplied to the existing analysis pass. Unavailable pages fall back to their collected headlines. Page extraction does not execute scripts or bypass access restrictions, and opening a published brief does not trigger this process.

2. Relevance and severity

An automated processing stage screens candidates for meaningful wider disruption and stores a severity estimate. Routine entertainment, sports, lifestyle and minor local stories are normally excluded. Classification can be wrong and is not itself permission to publish a summary.

3. Source-based publication review

A single main AI pass handles classification, source-linked details and practical consequences together. It uses up to eight source links distributed across hostnames. Mixed events and insufficient coverage are held rather than padded. Ordinary code checks the returned structure and references; another AI rewriting pass is not run.

Programmatic checks reject fabricated source IDs, incomplete results, obvious boilerplate and unsupported numeric literals. These checks do not establish the truth of every assertion. Passing them is not human review or independent fact-checking. See the publication standards for the complete distinction.

4. The main feed and archives

The main page considers up to 200 recent publication records from the last 48 hours, validates them, and displays up to 40 briefs after your chosen sort and filter. Hot narrows this to 24 hours and balances severity with freshness: an event's ranking weight halves over 12 hours from its first-report or latest material-update timestamp. This keeps serious developments visible as they age, while giving comparable new developments an advantage. More source links do not raise Hot rank, and the displayed Doom Score does not change. Latest orders by the first-report timestamp; Doom orders by stored severity. This is a selected feed, not an exhaustive count of world events.

Old source records remain reachable. Unpublished and held records show their links without generated summaries or severity scores, are excluded from the sitemap, and are marked not to be indexed. Published briefs can remain useful after leaving the current feed.

5. What the Doom Score means

The stored 0–100 score estimates potential severity and wider impact. It does not decay with age; freshness affects Hot ordering separately. It is not a probability, a verification percentage, a casualty prediction or a forecast of financial losses. The publication reviewer does not invent a numerical breakdown that was never recorded.

6. Comparing sources

Each reported detail can be expanded to show the references used for it. Numbered links open the source list, with a short stored excerpt when one is available. The publication timeline sorts those references by their recorded publication times, not by when the incident happened. Missing publication times are shown separately without substituting collection times.

Source-link counts, domains and repeated headlines describe coverage, not agreement, accuracy or independent reporting. Sources can repeat one another even when their headlines differ. Practical interpretations link back to the reports used as their factual starting point. New briefs can include up to six distinct details and two additional context points, only when supported by the supplied material. Historical context is omitted when no supporting reference is available; it is not generated from an assumed analogy.

7. Optional interpretations

Country and market interpretations are available on demand for published briefs only. Their input comes from the source-based brief rather than an unreviewed event summary. Cached interpretations are tied to the content version, not merely to the time it was checked. Changing a review timestamp does not require regeneration. They describe possible practical effects, not trading instructions, live safety advice or guaranteed outcomes.

8. Practical analysis and historical context

The analysis pages combine reusable source-based background with up to six relevant published briefs from the latest 48 hours within the same 200-record candidate feed. Matching uses reported headlines and details, not speculative consequences. Topic membership does not establish a common cause or independently verify a claim.

Historical examples link to their sources and state where the comparison stops. Lessons and scenarios are interpretations, not predictions or measured probabilities. Background has a fixed revision date; news entries show their own timestamps. Calculators use your supplied figures or booking conditions, do not fetch live prices or flight status, and do not call AI. Reloading an analysis page refreshes the selection of published briefs without regenerating the background.

Limits and corrections

The system does not independently verify the complete text of every original article. It can misunderstand wording, select a wrong source, miss a contradiction or omit an important event. For a consequential decision, inspect the linked original material. Report an error with the event URL and the exact assertion or source involved.