This newsroom is a research and editorial operating system. It continuously watches configured public sources, preserves evidence, links claims to provenance, asks agents to look for contradictory or independent evidence, and then hands the result to people for editorial decisions. It is designed to make research faster without hiding where a claim came from or automating the final editorial judgement.
+ +The system moves from detection to evidence to a human editorial decision.
The source layer is intentionally mixed: primary records, official datasets, notices, analysis, market data and corporate first-party material can all be monitored, but they are not treated as equivalent evidence.
| Class | What it means | Typical use |
|---|---|---|
| official_primary | Primary institutional record or publication. | What an institution formally published, filed or stated. |
| official_data | First-party statistical or operational data. | Metrics, series, releases and machine-readable evidence. |
| official_registry | Authoritative registry records. | Companies, registrations, removals and formal status. |
| official_notice | Legally or formally published notices. | Gazette notices, market announcements and similar events. |
| official_analysis | Analysis published by an authoritative institution. | Forecasts, commentary and analytical reports. |
| official_market_notice / official_market_data | Exchange announcements or direct market observations. | Company disclosures, price/volume context and market events. |
| corporate_primary | A company speaking for itself. | Strong evidence of what the company announced; not independent validation of its claims. |
| independent_research / newsroom / specialist | External analysis and reporting. | Independent corroboration, criticism, interpretation and subject expertise. |
A trust weight is an editorial prior about a source, not a probability that the source is telling the truth.
Each configured domain can carry a 0–1 editorial weight. The weight represents how useful and authoritative that source generally is for the type of evidence it publishes. Editors can change these weights in Source Trust.
Multiple URLs owned by the same publisher belong to one independence group. Ten articles from one organisation therefore do not become ten independent confirmations.
A company newsroom is excellent evidence that the company made an announcement. It is not independent evidence that the company's performance, safety or impact claims are correct.
Agents evaluate the strongest source in each independence group. Additional genuinely independent groups can strengthen a source set, while duplicates inside one group do not inflate confidence.
The newsroom distinguishes discovery metadata from substantive evidence so an attractive headline cannot masquerade as a verified source.
A title, sitemap entry or listing tells the newsroom that something exists. It is useful for discovery, but usually does not satisfy the sourcing gate.
First-party RSS, APIs and structured payloads can be preserved directly when they contain enough substantive content and provenance.
The strongest class in the current system: the actual page or document is fetched, text is extracted, hashed, timestamped and frozen as evidence.
Why freeze evidence? Web pages change. A Story Thread should be able to show what the newsroom actually saw when a claim was researched, including the source URL, retrieval time, content hash, excerpt and snapshot identity.
+The original URL remains available for readers, but the frozen snapshot is the provenance record used by the research graph.
+A Story Thread is not an article draft. It is the research object from which an editorial decision can later be made.
A thread can contain evidence, claims, counterclaims, trends, forecasts, actuals, questions, context, organisations, people, places and metrics. Relationships state why one node supports, contradicts, qualifies, updates or investigates another.
For every apparent story, the newsroom asks: what does this appear to say, and what is the strongest evidence that this interpretation is incomplete or wrong? Counter-evidence is a required research pass, not an optional decoration.
Agents accelerate research and maintenance. They do not own the editorial decision.
Monitor sources, gather evidence, freeze documents, identify entities, compare statistics, find contradictory evidence, propose relationships, audit sourcing gaps and surface source-health problems.
Agents use the Media Intelligence MCP to search evidence, inspect source health, evaluate trust and independence, create or extend Story Threads, and post research findings into Agent Activity.
Research actions are attached to identifiable sources, nodes, messages and audit events so a human can inspect why an agent reached a research state.
The system is intentionally built around explicit human gates.
| Decision | Who owns it | System role |
|---|---|---|
| Is this newsworthy? | Human editor | Agents can surface evidence and opportunity signals. |
| What is the message or framing? | Human editor | Agents can show competing interpretations and factual constraints. |
| Is the evidence sufficient? | Human editor | The sourcing gate makes gaps visible and blocks premature workflow progression. |
| Who should receive it? | Human editor | Future contact intelligence can propose audiences, not silently choose them. |
| Send or publish? | Human editor | No autonomous publishing or distribution is assumed by the newsroom workflow. |
The same permission model applies to the website and MCP actions.
Read Story Threads, inspect evidence and provenance, browse opportunities, view Source Trust and use read-only research tools.
Viewer access plus research mutations, Story Thread workbench actions, triage, status changes, evidence enrichment and Agent Activity actions.
Editor access plus newsroom policy/configuration actions such as editing the Source Trust policy. Unrecognised authenticated roles fail safe to viewer.
A thread cannot become editorially ready just because an agent sounds confident.
Automation helps organise evidence; it does not remove uncertainty.
A highly trusted source can still contain errors. A low-weight source can still publish an important fact. Weights guide research priority and corroboration, not verdicts.
Two newspapers repeating one press release are not genuinely independent evidence. Agents are instructed to check shared upstream provenance.
WAFs, blocked feeds, changed HTML, missing APIs and licensing restrictions can make sources temporarily unavailable. The Desk surfaces source failures instead of silently hiding them.
Passing the sourcing gate means the configured research requirements are met. It does not replace editorial judgement, legal review, subject-matter expertise or final fact checking.