Argus
Verified written intelligence from open sources.
Argus is the application of pureWRIGHT sold to clients. This is its record: how a continuous stream of open-source documents becomes assessed, provenance-linked written intelligence, why most documents are discarded on purpose, and how it connects to the rest of the system.
What a client receives
An excerpt from one brief, ranked first of five in its delivery. The fields are the ones clients read; the company is withheld.
The open web outruns any analyst
The signal an organisation needs is public: funding announcements, regulatory filings, hiring pages, procurement notices, technical reporting. The problem is volume and form. It arrives as thousands of documents in no particular order, and reading it is a full-time job nobody was hired to do.
Aggregators do not solve it. Forwarding two hundred headlines a day converts a reading problem into a bigger reading problem. What an analyst actually needs is the opposite of aggregation: assessment. Which of these documents says something new, matters, and survives being checked against its own sources?
There is a second, quieter problem: volume is not value. Argus is built to throw most of what it reads away, and the output is finished, written analysis rather than a feed.
From raw document to published analysis
One logged path. Every document carries a record of each processing step, from the feed it came from to the score that decided whether it was published.
- News reporting
- Official releases
- Sanctions lists
- Threat-intel feeds
- Stage 01Collection
- Stage 02EnrichmentEvery document also goes to the knowledge graph: entities, co-mentions, timelines.
- Stage 03Assessment councilMost documents are not published.
- Stage 04DraftingA cloud model writes the analysis for documents that pass.
Judged in parallel, published on score
No single model decides what matters. Four seats score every document independently, each with its own brief, across novelty, impact and analytical depth, and a document is published only if their average is 7 out of 10 or more and no seat scores it below 5. Most are not published. That is not a failure mode; it is the product working.
- EditorScores 1 to 10.
- AnalystScores 1 to 10.
- SpecialistSecurity, compliance or biosecurity.
- Fact-checkerScores 1 to 10.
- passPublishedDrafted by a cloud model.
- near missListedOn the wire, not published.
- failArchivedNot published.
How Argus connects to the three parts
Argus runs its own pipeline. Around it, pureMEND watches its sites and raises the alarm when they fail, pureBOT's mail tool sends the brief emails, and pureMIND indexes the engineering notes behind it. Each part keeps its own engineering record.
- watched bypureMENDSite uptime alerts
- mailed bypureBOTIts mail tool
- indexed bypureMINDArgus work notes
What a brief promises
Four commitments, each a construction fact of the system rather than a policy.
Researched, not aggregated
Each delivery is researched across funding news, filings and hiring signals, ranked by script, and picked by an editor. Five companies make a delivery; most candidates never do.
Every claim has a source
A brief quotes its sources verbatim and links them, and a validator checks every quote and link before it ships, so an analyst can check the work instead of trusting it.
A person makes the call
Scripts rank the candidates; a person makes the final editorial pick of every delivery and nothing ships without it.
In production, not in slides
Argus has delivered scheduled briefs to clients in the AI infrastructure sector since June 2026: twice a week, to a gated portal, with a short email when each delivery lands.
An application, on the record.
Argus is in production and taking a small number of new client conversations. A sample brief is the fastest way to judge it: ask, and one arrives.