What is here, what is at the Comparative Agendas Project, how the coding works, and the terms under which any of it can be reused.
Three collections, with different provenance, coverage and freshness. They are not one dataset and should not be treated as one.
The long-run US series — congressional hearings, bills, statutes, executive orders, Supreme Court cases, budgets, front pages, party platforms and the rest — are hosted, documented and versioned by the Comparative Agendas Project. They are free and public, and this site links straight to those files rather than mirroring them.
Browse the domains →This site's own contribution: contemporary US press articles classified to CAP major topics. Searchable by topic, source and year, and reported in aggregate on the domain and election pages. Titles and links only — no article text is reproduced. Bulk export is not yet available; write to us if you need it for a project.
Search the corpus →A separate, continuously updating collection: state news coverage for the 2026 Senate races. Different sources and a different window from the searchable corpus above — these run to the last few hours, and are not part of the search index.
Midterms 2026 →The open-source components behind the collection and classification layers, including the classifier service and the published models.
Code repositories ↗What the codes mean, and which of them this site actually reports. Method in full on Data & methods.
All policy coding on this site follows the Comparative Agendas master codebook — 21 major topics, codes 1–10, 12–21 and 23, each opening into subtopics. Codes 11 and 22 were retired from the original US scheme and never reused, so the gaps are deliberate and decades of coded data stay comparable. The codebook is maintained by CAP, not by this site, and is the authoritative reference for what any code means. Media sources carry a further extension — codes 24 to 41, for obituaries, sport, weather, advertising and the like, plus a ‘no policy content’ outcome. Those are not master-codebook topics and are excluded from every policy figure here.
Aggregates and article-level labels on this site use the CAP major-topic set. Where a figure reports a process share, that is major topic 20, Government Operations. Treat it as an UPPER BOUND, not a clean measure of campaign coverage: 20 is a substantive policy topic in its own right — appointments, procurement, tax administration, the civil service — and campaign coverage is only the 2012 Political Campaigns subtopic beneath it. Measured on the Maine press for the 90 days to 24 August 2026, 973 of 1,893 code-20 articles carried subtopic 2012; the remainder is largely government-operations policy. A subtopic-level process share would be the sharper measure, and is not what the figures on this site currently report.
Two audits of the Maine corpus, both run on 25 August 2026, put limits on the subtopic field. First, nesting: 2,452 of 8,672 policy-coded articles carrying a subtopic — 28.3% — were given one that belongs to a different major topic than the article's own, which is not a valid code combination. Second, stability: among 2,165 groups of byte-identical article text republished across outlets, the same classifier returned a different major topic for 10.9% of them and a different subtopic for 17.1%. Both numbers are self-consistency measures, so they bound reliability from above rather than measuring accuracy. Read any single subtopic as a hint, not a fact, and treat the major-topic figures — which is all this site reports — as the trustworthy layer.
The archival series are human-coded. The contemporary corpus is classified automatically, using models trained on human-coded examples and checked against held-out human-coded sets. The two are not interchangeable, and figures on this site say which they rest on.
The group tags on article lists are dictionary matches over each headline and the opening of the article, not classifier output and not part of the CAP scheme. They are a working draft of a category scheme still being designed.
Set by the Comparative Agendas Project, not by this site. Read it before reusing anything.
© Comparative Agendas Project 2015. The CAP topic-code variables and other Project-generated variables are distributed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license ↗. Any reuse must abide by those terms and include the applicable academic references.
Where a dataset's codebook says so, the original copyright in the primary source material applies to reuse as well. Check the codebook for the specific series you are using.
Topic codes reported here are CAP topic codes, so the same non-commercial share-alike terms carry over to them. Article titles and links belong to their publishers and are shown, not relicensed; no article text is reproduced anywhere on this site.
Cite the dataset you actually used, not this site, whenever the numbers come from an archival series.
Each US series carries its own citation — CAP's guidance for the United States is to refer to the respective dataset. The citation and the academic references to include are given with the file and on CAP's How to Cite page ↗.
For figures or coded output taken from these pages:
The US Policy Agendas Project. August 2026. The Policy Agendas Project data site, v0.1 (preview). Syracuse University Maxwell School and the University of Texas at Austin. https://pap.poltextlab.com — accessed [date].
No DOI has been registered yet, so please include the access date. Version 0.1 (preview), last updated August 2026.
If you use the automatically coded corpus, cite the method as well — the peer-reviewed workflow papers are listed under published methodology on Data & methods.
uspolicyagendas@syr.edu — including for bulk access to the coded corpus. More ways to reach the project on About.
The figures on this site are computed from the underlying collection of coded articles. The methods below say how the headline numbers are defined; for bulk or programmatic access to the data, write to us.
The election cards report CAP major topic 20, Government Operations — the code that campaign, procedural and horse-race coverage falls into — as a share of all policy coding for the same articles. Topic shares elsewhere on the site count CAP major topics 1–24 over the chosen window.
For bulk or programmatic access to the coded data behind these pages, write to us — see About.
The classifier model cards are public in the model library ↗. The model weights are access-restricted and currently closed to new requests, and the collection and classification code is not yet public. Write to us for access.
These are live figures: a reading taken tomorrow will not match today's numbers, so record the date with any figure you take.