Sources
Where the intelligence comes from.
What feeds the Research Console, and how it is checked before anything is published.
This page is a companion to Methodology, which covers how an individual AI Disruption Snapshot is built. This one covers the standing research behind the Research Console.
1. What we track
Five public tracks run continuously, plus a sixth that is specific to your own book of clients or prospects:
AI Frontier
What labs, hyperscalers and vendors shipped, priced or scored. A curated roster of 68 organizations across 10 countries and 10 segments, tracked continuously. The segments are the structure, separating frontier labs from open-weight developers, compute and silicon from the platforms built on it, and independent evaluators from the vendors they assess. Published prices, open-weight releases and independent benchmark results are treated as harder evidence than announcement coverage, because they say what a thing does and what it costs rather than what its maker says about it.
Adoption
Named organizations changing how they operate, collected separately in each of the five markets we track. Evidence of what is actually landing, not what is being sold, and always attributed to the market it happened in.
Regulatory
What governments and regulators are actually doing, tracked separately in five markets: the United States, Australia, China, the European Union and the United Kingdom. Constraint and permission, market by market, never one market’s rules presented as another’s.
Invention
Patent activity in AI and machine learning, a leading indicator of what is being industrialized now. Counts come from the US patent office, so they are a US-office measure and never a global one.
Federal spend
Public contract award data, showing where government AI money is actually going. United States only; there is no equivalent dataset in the other markets and we never imply one.
Client signals
Matched to your own book when you run the Console. Not part of any public feed.
2. Verification
Nothing is published unless it ties to a real, checkable source. No invented figures, no invented URLs, no invented dates or attributions: the same standard the platform applies to every deliverable it generates for a paying engagement.
Figures reported in the weekly briefs are carried through verbatim from what was found, not rounded, estimated or reworded for effect. Where the evidence only supports a reading rather than a fact, it is written as a reading, not stated as settled.
3. Cadence
Four tracks refresh weekly, each on a Monday. Federal spending and client signals refresh monthly, on the first of the month: a weekly read of either would mostly report no change. The regulatory track runs as five independent weekly collections, one per market, so a quiet week in one market never borrows another market's activity to fill the page. Where a market's collection turns up nothing that meets the citation standard, its previous week is kept rather than the gap being filled with weaker material.
Each week's evidence is collected once and shared by every brief that draws on it. Nothing is collected again per reader or per surface.
4. Selection
More evidence is collected each week than any one brief can carry, so what gets left out matters as much as what goes in. Selection reserves places for the hardest evidence first: model releases, published prices and independent benchmark results claim their places before announcement coverage fills the rest. Without that reservation, a week of heavy vendor announcements would push every shipped model and every price change off the end of the list.
The regulatory and adoption tracks reserve places per market on the same principle, so one loud market cannot crowd out the other four.
5. Citations
In the Research Console, the signals behind each weekly brief sit under it with their sources.
6. What this isn't
This is a research operation, not a data vendor. We don't publish raw feeds, the underlying datasets, or the process used to select, weight and cross-reference what makes it into a brief. That is the analysis, and it is the product.
It also isn't investment advice, and vendor claims, previews and patent filings are reported as what they are, not as proof of a shipped or verified outcome.