In the past two weeks, 12 news organizations created or opened 13 public repositories on GitHub. Highlights below.
Highlights
The Baltimore Banner published the data and analysis behind “ICE sent her 2,000 miles from her Maryland home. She’s far from alone.,” by Daniel Zawodny and Sahana Jayaraman. It covers more than a million detentions from fiscal 2022 through 2026, drawn from records the Deportation Data Project obtained through litigation, and measures the distance between each person’s consecutive detention facilities. The team matched facilities for more than 99% of detainees against directories from the Vera Institute of Justice and the Deportation Data Project, then used a log-transformed regression to separate the effect of the administration from the effect of which facilities were available. The repo holds the R code, the pre-computed spatial data and the facility directories. The Banner also published a detailed methodology. You love to see it.
ProPublica revived a parser for federal financial disclosures that had sat untouched for nine years. Chris Zubak-Skees wrote it in 2017 at the Center for Public Integrity, where he was then data editor. Brandon Roberts picked it up at ProPublica last month, added tests and validation data, and made it survive malformed filings. It reads OGE Form 278e PDFs, the disclosures senior federal officials must file, and writes CSVs covering positions held outside government, assets and income, liabilities, transactions, gifts and employment agreements, for the filer and a spouse. Yet another reason to open-source your code, folks! Other newsrooms can adopt the code even after you’ve closed up shop.
Chalkbeat opened the reference library behind Votebeat’s Election Expert Desk, a pop-up Slack community that matches reporters with election-administration experts during election season. The desk, which started as a project of the Knight Election Hub (which I led) in 2023, returned for the 2026 midterms. Journalists who join the free program can put questions to vetted experts and read what other reporters have asked. The repo includes federal election laws and cases, documents of state election rules and guidance, poll-worker guides and state election codes.
The Data Liberation Project opened data on hospital adverse events — the medical errors and injuries that happen during care rather than from the underlying illness — from four states so far: California, Colorado, Michigan, and Washington. A federal inspector general survey found that 26 states and the District of Columbia require hospitals to report them, and the project has filed requests with more than 30 states to find out which do and what they will release. Each state returns a different set of fields, but every row is one event with a type, a date and a facility. It is part of the project’s broader patient safety data initiative, and the README lists public-records requests filed for the remaining states. (Note: I’m on the board at Muckrock, the organization that runs the Data Liberation Project)
USA TODAY opened its analysis of ICE hiring and departures, which backs “ICE went on a hiring spree. Then workers left in record numbers.” Ignacio Calderon compiled every monthly accessions and separations file the Office of Personnel Management (OPM) publishes, back to 2005, with ICE figures starting in August 2011 and running through June 2026. The repo reports that ICE hired nearly 12,000 full-time workers between January 2025 and June 2026, more than in the whole decade before, and that 4,800 left in the same period. Retirements usually account for most departures; in 2026 quits and terminations overtook them. More people quit in the first half of 2026 than in any full year since 2012. The raw files were too large to ship, so the repo carries the R script and the processed outputs behind the graphics.
Follow the Money, the Dutch investigative outlet, published the data and code behind “Overdag meldt Tata braaf ontsnappende gifwolken – ‘s nachts blijft het opvallend stil” (“By day Tata dutifully reports escaping toxic clouds; at night it stays notably quiet”). Jan Daalder, Sjors Hofstede and Marit Gijsberts examined more than 16,000 incident reports the steelmaker filed with a regional environmental authority between 2014 and 2025. Between 2020 and 2024 the plant reported incidents 2.4 times as often by day as by night, though it runs continuously. The pipeline converts PDFs obtained by records request into structured data with pdftotext, joins them to sunrise and sunset times from the Dutch meteorological institute, and filters for maintenance work to test whether scheduled repairs explain the gap. (Note: I can’t read Dutch so I’m leaning on Claude here more than usual.)
Correctiv opened a prototype community app that gathers its investigations, fact checks, podcast audio, membership features and crowdsourced callouts in one place, built with React Native and Expo for Android, iOS and the web. Some interesting ideas here: Every architectural decision is written down in a dedicated directory. Every data source the app uses is also inventoried, and a weekly job checks each one against the live source and opens a pull request with whatever changed. The README is pretty open about how it was built using “vibe coding,” with a person directing the work, and the repo ships a CLAUDE.md, an AGENTS.md and a .claude directory.
Verso released Big If True, a fact-checking plugin for Claude that works through a piece of text claim by claim. It picks out the checkable claims in an article, press release, draft or newsletter, searches primary records and the live web, and marks each one supported, not verified, or needing review, with sourcing attached. It runs in Claude Code, Cowork or the Claude app with web search on.
Fiquem Sabendo, the Brazilian public-records news agency, opened a hub for its reproducible data reports, published as a static site. Each report gets its own folder and documented pipeline, with a metadata file and a script that regenerates the index, so a new report does not mean new infrastructure. The first one is the federal tax-exemption dataset by company that we covered last issue. Nine of the repo’s first 10 commits credit Claude. (Note: My Brazilian Portuguese is no better than my Dutch so I’m taking Claude’s word for this.)
By the Numbers
Beyond new repos, 93 news organizations made a combined 5,745 public commits to GitHub during this period. The most active by commit count (excluding, as best we can, commits done by bots, gh-actions, or cron):
| Organization | Commits |
|---|---|
| The Guardian | 1,627 |
| Buried Signals | 908 |
| Freedom of the Press Foundation | 399 |
| Correctiv | 257 |
| Bloomberg | 228 |
| MuckRock | 143 |
| ICIJ | 134 |
| OpenSanctions | 125 |
| PRX | 94 |
| Open States | 92 |
This roundup is assembled with the help of open-journalism-bot, which monitors public GitHub activity from ~420 news organizations and posts new repositories to @openjournalism.news on BlueSky. Commit counts shown here exclude commits we identified as automated.