In the past two weeks, more than a dozen news organizations opened or created public repositories on GitHub. Highlights below.
Highlights
The Pudding published the code and data behind “A Love Story”, Alvin Chang’s remarkable scrollytelling piece on how the pandemic reshaped American relationships. It draws on Stanford’s “How Couples Meet and Stay Together” (HCMST 2017) survey, which followed thousands of people across three waves in 2017, 2020 and 2022, to trace how couples fared through lockdown, finding that strong relationships often grew stronger while fragile ones frayed.
CalMatters published the data and fact-checking notebook behind its reporting on green card adjustments of status, which appeared in a June 4 story, “5 things to know about the Trump administration’s new green card policy”. The analysis finds that California processed 112,100 green card adjustments in 2023, nearly one in five of all such adjustments nationwide. The repo is a Marimo notebook by Mohamed Al Elew, a journalism engineer at CalMatters. A README notes Claude Code helped write the documentation, though not the analysis.
USA Today published the data behind its analysis of ICE’s Alternatives to Detention program, part of its ongoing interactive DHS immigration enforcement tracker. Ignacio Calderon compiled more than 100 spreadsheets from the Wayback Machine’s archive of ICE’s detention-management page going back to 2021, then traced how the program’s use of ankle monitors has shifted. The repo ships the processed data alongside an R Markdown analysis and the team posted a handy rendered version as a GitHub Page.
The Guardian developer Andy Gallagher opened a prototype for fine-tuning language models to structure recipe data. The Go project trains Amazon Bedrock models on the Guardian’s own archive of human-edited recipes, pairing each recipe’s raw HTML with its published structured JSON to teach a model to turn messy recipe pages into the clean schema that feeds its Feast cooking app. A separate evaluation harness scores each model’s output on a 0–12 error scale across JSON validity, ingredients, method steps and titles, using Claude Sonnet 4.6 as an automated judge.
OpenSanctions released Pravda, a web-archiving service it calls “the evidence layer.” Built on Playwright, it captures web pages as MHTML archives and full-page screenshots, stores their response headers, and tracks snapshot history in content-addressed storage, running full Chrome in a virtual framebuffer so pages render as a real user would see them. The result is durable, addressable evidence that can be inspected and diffed long after the original page changes. The repo ships an AGENTS.md and is largely the work of developer Johan Schuijt. During the same window OpenSanctions founder Friedrich Lindenberg also published a client library and CLI for its Yente entity-matching API.
MuckRock‘s Mitchell Kotler published a DocumentCloud add-on that extracts tables from documents as Markdown. It runs Amazon Textract over uploaded documents, identifies tabular data, and writes it back into the document text as Markdown tables — aimed at the PDFs of financial reports and other structured records journalists routinely have to retype by hand. (Disclosure: I’m on the board of MuckRock.)
ICIJ developer Bruno Thomas published cceco, a command-line tool that tallies the environmental cost of using Claude Code. It reads local session history and reports estimated electricity, greenhouse-gas, water, metal and fossil-fuel usage broken down by model and date range, drawing its figures from the EcoLogits project. The tool was itself built using Claude Code.
iMEdD, the Athens-based investigative outlet, published a dataset of Attica’s water reserves stretching back to 1985. The data records the daily volume of extractable water, in cubic meters, across the four reservoirs that supply the Greek capital.
Buried Signals‘ Tom Vaillant released a benchmarking suite for the tools investigative journalists use to parse documents, automate browsers and scrape the web. It runs PDF-extraction tools, browser-automation tools, and scrapers over real investigative source material, then produces a static HTML report comparing them.
By the Numbers
Beyond new repos, 80 news organizations made a combined 3,737 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 | 969 |
| ICIJ | 290 |
| Freedom of the Press Foundation | 230 |
| MuckRock | 179 |
| ABC News (Australia) | 144 |
| PRX | 124 |
| OpenSanctions | 115 |
| OpenStates | 93 |
| OpenNews | 85 |
| The Pudding | 74 |
This roundup is assembled with the help of open-journalism-bot, which monitors public GitHub activity from ~360 news organizations and posts new repositories to @openjournalism.news on BlueSky. Commit counts shown here exclude commits we identified as automated.