How the desk actually files intelligence.

1. Source tiering

The desk treats sources unequally on purpose. Not every feed is the same. Headlines are tagged with a tier when they are ingested, and downstream logic (alerts, verification, sentiment weighting) uses the tier to decide what to do.

The goal is simple: every item has to survive collection, verification, translation where needed and scoring. If a research tool cannot show its workings, you should not rely on it.

2. Verification chain

A single Telegram post is not news. A single tier-1 headline is a signal, not a confirmed story. The desk explicitly tracks the verification chain for major events: the first source to file, the first independent confirmation, and any contradiction. The chain is shown on the item so the reader can make their own call about reliability.

Every event carries a reliability letter, A, B, or C. An item marked A has been corroborated across at least two independent tiers. B is a single credible source not yet confirmed. C is a single OSINT source that is worth watching but is not yet standing on anything. The letters travel with the headline into any alert that is fired.

3. Refresh cadence

• Wire and squawk feeds. Polled on a 120-second cycle. New items surface within that window.
• Telegram OSINT channels. Polled on a one-minute cycle, with a four-hour archive window for items that carry forward through the day.
• ISW, Critical Threats, long-form think-tank feeds. Polled on a six-hour cycle; these are slower sources by design.
• Market data. Provider-dependent quotes for Brent, WTI, Gold, DXY and selected equities are shown with their observation time and freshness. The desk does not promise tick-level commodity data or execution-grade pricing.
• AIS vessel context. Provider-dependent via AISStream.io, with bounding boxes for the Persian Gulf, the Strait of Hormuz, the Gulf of Oman, the Bab el-Mandeb, Suez and the Bosphorus when transponders broadcast.

4. Translation workflow

A large fraction of the most useful conflict intelligence surfaces in Russian, Farsi and Arabic hours before it touches the English wires. The desk may machine-translate non-English sources into clean English when the upstream translation request succeeds before scoring. Both the original text and the translation are preserved on the item so the reader can verify the translation when it matters; unavailable translations stay labelled as such.

Translation is intentionally literal, not interpretive. The desk does not try to editorialise a Tasnim denial into something it is not. If the original is vague, the translation is vague; if the original is a claim rather than a fact, it is labelled as a claim.

5. Sentiment and alert logic

Every headline is scored in real time for sentiment and risk direction, risk-on, risk-off, or neutral. The current production scorer is deterministic: it uses keyword pattern matching tuned for energy and defence language plus source and alert rules for tripwires (for example: "strike", "ballistic", "closure of the strait"). It is a classification aid, not a predictive model.

Alerts escalate through three levels:

1. Informational. Tagged as relevant to your watchlist or keyword list. Shows up on the desk, no notification.
2. High. A notable event for a tracked region or ticker. Browser notification, optional email.
3. Critical. A market-moving tripwire: strikes, sanctions releases, central bank surprises, named chokepoint events. It requires at least two independent evidence buckets before it can route as critical, sound the audio squawk, push a notification, or flag portfolio impact.

6. Action-queue score

The queue score ranks what to inspect first; it is not probability, conviction, or investment advice. For a cross-checked claim it combines action state (45%), proof/trust score (42%), market confirmation (up to 12 points), and confirmation speed (up to 8 points). The result is clamped to 0–100: 88+ is critical, 70–87 high, 52–69 watch, and below 52 monitor.

Hard review gate: raw evidence volume is not corroboration. A claim with fewer than two independent buckets is labelled VERIFY, cannot be critical, and its action-queue score is capped at 35. Severity remains visible separately from confidence and market confirmation so an urgent claim cannot borrow authority from a keyword match, repost count, or market move.

7. Market-impact scoring

An intelligence event is only interesting to most Intel Desk users if it also moves markets. The desk keeps a rolling correlation window between tagged events (by tag, e.g. iran_strike, opec_surprise, fed_meeting) and the price action in linked instruments, Brent, WTI, LNG equities, defence primes, oil-majors, DXY, gold. When an event fires with a tag that historically correlates to a real price move, the alert carries that expectation so you can react with context rather than guess.

8. How the desk is designed to be used

• Intelligence, not advice. The desk surfaces verified information and scores relevance so you can make faster decisions. Your edge is speed plus judgement.
• Tripwires, not trade signals. Critical alerts fire when a threshold is crossed. What you do with that information is where your expertise comes in.
• Aggregation with provenance. Every item links to its original source. The desk gets you there first; you verify the detail.
• OSINT with clear labelling. Telegram posts surface with a source label and reliability letter. Single-source items are flagged. Multi-source items earn their way to confirmed status.
• Translation with originals preserved. Farsi, Russian, and Arabic sources are auto-translated with the original text retained. Provisional translations are flagged.
• Built for analysts, not to replace them. The desk handles the monitoring, scoring, and correlation. You handle the synthesis and the call.

If you want to see the sources themselves, which feeds, which categories, how they map to tiers, read the Sources page. Public case files are published only after a complete canonical reconstruction and timestamp audit.

For detailed breakdowns of the verification process, lead time performance, and how the desk handles contradictions, see the proof pages: How We Verify, How Fast We Catch It, When Sources Disagree.