Product
What a run can read
Every source on this page is on for production runs today. Each tile says what a run gets from it. The planner picks the sources a question needs, so a run reads some of these, not all of them.
- Data sources
- 47
- Subjects
- 13
- Licensed publishers
- 2,200+
The web and your vault
03Your vault
Sources, notes and claims from earlier runs in the same project, searched before any web request.
Web search
Finds candidate pages. A run cites the page it read, never a search snippet.
Public web pages
Fetched logged out, directly first, then in a rendering browser when the page needs JavaScript. CSV, Excel and ZIP files are read as tables.
Papers and scholarly indexes
07- Open data
OpenAlex
Papers, books and chapters across every field, with authors, venues and citation counts.
DOI records for the papers a run cites, including retraction and correction notices.
- US Government
PubMed
Biomedical literature search, used by the scholarly lane alongside OpenAlex.
Unpaywall
Finds a legal open-access copy of a paper, so the run reads the whole paper rather than the abstract.
Europe PMC
Full text of open-access life-science papers.
- US Government
PubMed Central
Full text of open-access papers archived at the US National Library of Medicine.
CORE
Full text of open-access papers from university and research repositories.
Developer communities and social posts
04
Repositories and issues, found by search and read from the page itself.
Hacker News
Discussion threads on a product, company or technical question.
Stack Exchange
Questions and answers, Stack Overflow by default.
X (Twitter)
Posts from the last seven days, read through the official API when a question turns on what people said.
Company filings
02- US Government
SEC EDGAR
Full-text search of company filings, five-year financial statements from XBRL data, and Form D offering notices.
- State government
Minnesota franchise filings
Franchise disclosure documents filed with the Minnesota Department of Commerce, read Item by Item.
US official statistics
07- US Government
U.S. Bureau of Labor Statistics
Jobs and wages by industry and county (QCEW), wages by occupation (OEWS), consumer prices (CPI) and local unemployment.
- US Government
U.S. Census Bureau
Population, income and housing (ACS, PEP), business counts (CBP, NES, Economic Census, BDS), building permits and new business applications.
- US Government
U.S. Bureau of Economic Analysis
Personal income and GDP by county and metro area.
- US Government
U.S. Energy Information Administration
Electricity and natural gas prices by state and sector.
OpenEI Utility Rate Database
Published utility tariffs for a location.
- US Government
HUD
Fair market rents by county, metro and ZIP, and how a ZIP splits across tracts and counties.
- US Government
U.S. Small Business Administration
7(a) and 504 loan counts, sizes, lenders, franchise brands and charge-offs by industry and state, from SBA's FOIA data.
Local records and places
06City open-data portals (Socrata)
City and county records: building permits, business licences, code cases and property sales.
City and county GIS (ArcGIS)
Map layers such as zoning and parcels.
- US Government
FEMA flood maps
The flood zone at a site, from the National Flood Hazard Layer.
- US Government
FHWA traffic counts
Traffic volumes on roads near a site, from the Highway Performance Monitoring System.
- Open data
OpenStreetMap
Mapped businesses near a point, as a floor for competitor counts.
Places near a point, used to cross-check competitor counts.
Powered by Foursquare
Search demand and local listings
01DataForSEO
Search demand and map listings: monthly search volume for a term in a US area, and nearby businesses with ratings and review counts.
International statistics and reference
03- Open data
Our World in Data
One indicator for named countries over time, cited to the original producer.
CC BY 4.0
DBnomics
Series from national statistics offices, central banks and international agencies: NBS China, INSEE, Destatis, ONS, IMF, ILO, BIS, ECB and others.
- Open data
Wikidata
Reference facts about a named company or organisation: founding date, headquarters, parent, tickers, official website.
Compliance and enforcement
06- US Government
EPA ECHO
Facility inspections, violations and penalties for a named company.
- US Government
CFPB complaints
Consumer complaints filed against a named company.
- US Government
CPSC recalls
Consumer product recalls.
- US Government
NHTSA recalls
Vehicle and equipment recalls.
- US Government
OFAC sanctions list
Screening of a named company against the Specially Designated Nationals list, refreshed nightly.
- US Government
U.S. Department of Labor
OSHA inspections and citations, and wage-and-hour cases, for a named company.
Software vulnerabilities
02- US Government
National Vulnerability Database
CVEs recorded against a named product or vendor, with CVSS severity.
- US Government
CISA Known Exploited Vulnerabilities
Whether a CVE is known to be exploited in the wild.
State legislation
01- Open data
Open States
State bills on a topic with status, sponsors and latest action, and current state legislators.
Patents
02EPO Open Patent Services
Patent publications worldwide, including China, Japan and Korea, with applicants, dates and family countries.
- Open data
Google Patents Public Data
Patent-landscape counts by year, applicant and patent office.
Google Patents Public Data by IFI CLAIMS Patent Services and Google, CC BY 4.0
Attention and adoption
03PyPI downloads
Downloads of a named Python package over the last 12 months, by country and installer.
- Open data
Wikipedia pageviews
Pageviews of English Wikipedia articles on a named company, product or topic, over time.
- Open data
GDELT
Weekly volume of news articles mentioning a named company or topic.
Licensed news
2,200+2,200+ news publishers
A run buys an article per page under the publisher's own licence when it answers the question, writes its own note, and cites the article's URL. The article itself is never stored.
How a run reaches a source
A run filed under a project searches that project’s vault before it makes any network request, so a question that overlaps an earlier one in the project starts with the sources, notes and claims already there. Sources that came from the vault are labelled as such on the receipt. A post on X and a licensed article are never served from the vault, because their bodies are never stored. The vault page describes what is kept.
A public page is fetched logged out, by a direct request first and then in a rendering browser when the page only exists after JavaScript runs. If no rung gets the page, the run raises an escalation, and you resolve it by uploading a copy you are entitled to. Each rung is recorded on the source, so the receipt says which one got the page. Everything fetched is treated as untrusted input before a model reads it.
For any paper with a DOI, the run asks Unpaywall, Europe PMC, PubMed Central and CORE for a legal open-access copy and reads the whole paper. The note says which version it got, and a preprint is labelled as a preprint. Where no open copy exists, the note says so.
The data sources return tables, not pages. A run keeps each table as a note that names the producer, the dataset, its last update, the licence and the URL to cite, with the figures in the note itself, so the claim check can find a number the report quotes.
Posts on X
The X lane opens only when a question turns on what people said as something happened, such as an outage, a recall, a launch or a public statement. It is never part of the default search. It reads through the official X API over the seven-day recent-search window, with at most two searches of up to 25 posts per run. A post becomes a citation to its x.com URL with the run’s own note about what it said. The post body is never stored in the vault, never exported and never reused by a later run.
What a run does not read
- Anything behind a login. The service holds no credentials, never signs in anywhere and never solves a CAPTCHA. A page you can only see when logged in is a page you upload.
- Paywalled articles, and paywalled papers with no open-access copy. The note records the refusal instead of guessing at the content.
- X posts older than seven days.
- Video transcripts. A video is cited by its page, not by what was said in it.
- Private company and funding databases such as Crunchbase and PitchBook. There is no connector and no licence for them. A run reads what a company publishes on the web and in its filings, and a customer who holds a licence to such a database can upload the record, where it is kept in that workspace only.
- Your own documents, unless you put them there. An upload is scoped to the workspace it was made in and is never reused across workspaces.
What every source records
Every source in a report carries the route that read it, the provider, the time of the fetch, the rights signal the source or its index reported, and whether its text is kept in the vault or was read once and discarded. Copies of one article are clustered, so syndicated versions count as one voice in the independence audit. All of this is on the verification receipt for every run, and the trust page states the data handling for each retention class.
Logos and names belong to their owners. They show where data comes from and do not imply endorsement.
Updated 2026-09-30