Public Money Watch: Starter Kit, Round 1

The task

Circle the row. From one year of U.S. Department of Labor contracts, flag the awards a reviewer should open first. Tie each flag to its public record and to a rule you can state in one sentence.

Start in two minutes

No installs. Python 3.8 or newer, standard library only.

git clone https://github.com/Multi-Sync/public-money-watch.git
cd public-money-watch
python3 baseline.py           # 470 flags on 2,358 awards -> flags.csv
python3 check.py flags.csv    # passes, and warns: 19.9% of awards is too many to read

Fork it, or start a repo of your own and copy these files in. Either is fine.

The baseline flags competed awards that drew a single offer. It is correct, and it is too blunt: it circles one award in five. Round 1 is about doing better than that. Fewer, sharper flags, each one you can explain.

What's in the box

• data/dol_contracts_fy2024.csv: the frozen data, 2,358 awards. Every entry is read against it.
• data/DATA.md: where the data came from, what each column means, and the traps.
• baseline.py: one red flag, end to end. Copy it and add your rules to flags().
• check.py: run it before sending. It fails on untraceable flags and on accusation words.
• fetch_data.py: rebuilds the data from the USAspending API. Optional.
• LICENSE: MIT, for the code. The data in data/ is derived from USAspending.gov, a work of the U.S. federal government, and is in the public domain.

What to send

Email hello@interworky.com with:

1. A link to a public repo with your code and your flags.csv. python3 check.py flags.csv must pass.
2. One paragraph. For each flag: the rule in one sentence, why it signals risk (cite a source), and one honest reason an award could trip it.
3. The name you want credited.

Rules

1. A flag is a lead, not an accusation. Describe the data, not the people. Use the legal words only where a source says so: an auditor estimates, a prosecutor charges, a court convicts. You flag.
2. Every flag points to one row in the frozen file and that award's USAspending page.
3. State each rule in one sentence. The code must do what the sentence says.
4. Use the frozen file so entries can be compared. You may join other public, free data if you name the source and your repo fetches it.
5. Keep people out of it. Publish nothing about a private person beyond the public record, and name no one as a suspect.

How entries are read

A person reads every entry, in this order:

1. Does it run? Your repo produces your flags.csv from the frozen file.
2. Is it traceable? Every flag points to a row, every rule is stated.
3. Is it sharp? Would a reviewer want to open the rows you circled? Fewer, better flags beat many.
4. Is it honest? Your paragraph says what the flag cannot tell you.

Credit goes to the name you give. Round 1 is open until 31 October 2026.

Where to look next

• U.S. federal awards, contracts and grants: https://www.usaspending.gov
• Inspector General reports from every agency: https://www.oversight.gov
• Any country: the Open Contracting Partnership's guide to 73 procurement red flags, with formulas, for data in the Open Contracting Data Standard: https://www.open-contracting.org/resources/red-flags-in-public-procurement-a-guide-to-using-data-to-detect-and-mitigate-risks/
• Cardinal, an open-source library that computes those red flags: https://github.com/open-contracting/cardinal-rs

Why this exists

The sixty-second brief at https://interworky.com asks who is circling the row. This kit is a place to start.