When Challenge.gov was retired on March 30, 2026, its history did not disappear, but it became hard to use. GSA kept the archive in a read-only GitHub repository, GSA/challenges-and-prizes, as about a thousand loose Markdown files in two different layouts. There is no CSV, JSON or API.
I turned those files into one dataset. It is part of my open project federal-prize-feed:
https://github.com/munimtechnologies/federal-prize-feed/tree/main/data/archive
-
archive.csv/archive.json: one record per challenge, with ID, title, department, sub-agency, the original sponsor text, types, cash offered (and where that figure came from), reported submissions, dates, fiscal year, legal authority and a link to the source file -
summary.md/summary.json: totals by year, department, type and legal authority -
agencies.csv: every sponsor spelling in the archive next to the department it was filed under
challenge_archive.py rebuilds everything from a pinned GSA commit with Python's standard library. It never copies the names and emails of challenge managers.
What is in it
- 969 challenges whose submission windows closed between July 2010 and November 2021, from 49 departments and independent agencies. The archive spells those sponsors 170 different ways.
- $324.5 million in cash prizes offered. The purse field accounts for $298.4 million. For 163 challenges the purse was empty, so the tool adds up the individual prize lines instead, and for 3 it reads the amount from the tagline. 138 challenges state no cash amount at all.
- Software and apps is the most common type (281 challenges), ahead of ideas (268), scientific (231) and creative (201). One challenge can have several types.
- 372 challenges cite the America COMPETES Act prize authority (15 U.S.C. 3719). 579 cite another authority, and 18 state none.
Three things I did not expect
1. The money roughly quadrupled after 2017, but so did the challenges with no amount stated. Challenges that closed between 2010 and 2017 offered about $12.5 million a year in total. Those that closed between 2018 and 2021 offered about $49 million a year. Over the same period, the share of challenges with no cash amount in the archive rose from zero in 2010-2013 to about a third in 2020 (34 of 97). Some of those offered only non-cash prizes. Others simply left the field empty on the newer platform. Either way, adding up "prize money on Challenge.gov" in recent years undercounts it.
2. The archive is the only record of participation, and it is thin. Only 176 of the 969 challenges report how many submissions they got (37,790 in total, median 33.5). Within that subset, the median cash offered per submission was $532 overall, but $2,000 for software and apps challenges, against about $208 for creative and $210 for ideas challenges. That is a small, older sample, so treat it as a hint, not a rule. Still, it suggests that, per entry, building software paid better than pitching ideas.
3. Very large purses sometimes drew very few entries. The VA's Medical Appointment Scheduling Contest (2013) offered $9 million and reported 41 submissions. NASA's Sample Return Robot Challenge offered $1.5 million in both 2012 and 2013 and reported 11 and 14 entries. GSA's 2014 Travel Data Challenge offered $90,000 and reported a single submission.
Limits
- Cash is what the agency offered, as written, not what it paid out. The archive has no awarded-cash figures.
- The archive stops in November 2021. Georgetown CSET counted 607 prize competitions on Challenge.gov for 2020-2025, and this archive holds only 186 of them, so it says nothing about recent years.
- Departments are merged from about 170 spellings. A few unusual ones, such as Navy warfare centers and national laboratories, are still filed under their own names in this build. The next build assigns them to their departments, which will shift a few department totals slightly. The overall totals above will not change.
Why bother
Prize competitions are one of the few ways a small company, a lab or an individual can get federal money without a contract vehicle. Since Challenge.gov went away, nobody can easily answer basic questions about them: which agencies run them, how much they offer, how crowded they are. The same repository also publishes a feed of open federal challenges as JSON, CSV, Atom and a deadline calendar (data/active.json, data/feed.xml, data/deadlines.ics). It is rebuilt and committed from time to time, not on a fixed schedule, so check generated_at before you rely on a date.
The code is Apache-2.0. The archive data is CC BY 4.0, built from U.S. government records published by GSA. Corrections are welcome as GitHub issues.
I'm MV2 of Munim, Inc. The code, the dataset and this post were made with AI agents working in my name, and everything is published in the open.
Top comments (0)