Skip to content

crdunwel/inspectors-general

 
 

Repository files navigation

Inspectors General

A project to collect reports from the offices of Inspectors General across the US federal government.

So far we have over 20 inspector general offices scraped, downloaded, and stable. See our safe list for details.

How you can help: Matt Rumsey kindly compiled a spreadsheet of IG offices. The left-hand column tracks who's completed what IG.

The best way to "claim" an IG is to create an open pull request with an empty scraper as soon as you've begun work -- you don't need to wait until the scraper is finished to submit a pull request.

Eric Mill has written a piece explaining what the project is and why it's being done. From it:

What's an inspector general?

Just about every agency in the federal government has an independent unit, usually called the Office of the Inspector General, dedicated to independent oversight. This includes regular audits of the agency's spending, monitoring of active government contractors and investigations into wasteful or corrupt agency practices. They ask tough questions, carry guns and sue people.

Scraping IG reports

Python 3: This project uses Python 3, and is tested on Python 3.4.0. If you don't have Python 3 installed, check out pyenv and pyenv-virtualenvwrapper for easily installing and switching between multiple versions of Python.

Dependencies: You'll need to have pdftotext installed. On Ubuntu, apt-get install poppler-utils. On OS X, install it via MacPorts with port install poppler, or via Homebrew with brew install poppler.

To run an individual IG scraper, just execute its file directly. For example:

./inspectors/usps.py

This will fetch the current year's reports from the Inspector General for the US Postal Service and write them to disk, along with JSON metadata.

If you want to go back further, use --since or --year to specify a year or range:

./inspectors/usps.py --since=2009

If you want to run multiple IG scrapers in a row, use the igs script:

./igs

By default, the igs script runs all scrapers. It takes the following arguments:

  • --safe: Limit scrapers to those declared in safe.yml. The idea is for "safe" scrapers to be appropriate for clients who wish to fully automate their report pipeline, without human intervention when new IGs are added, in a stable way.
  • --only: Limit scrapers to a comma-separated list of names. For example, --only=opm,epa will run inspectors/opm.py and inspectors/epa.py in turn.
  • --data-directory: The directory path to store the output files. Defaults to data in the current working directory.

Using the data

Reports are broken up by IG, and by year. So a USPS IG report from 2013 with a scraper-determined ID of no-ar-13-010 will create the following files:

/data/usps/2013/no-ar-13-010/report.json
/data/usps/2013/no-ar-13-010/report.pdf
/data/usps/2013/no-ar-13-010/report.txt

Metadata for a report is at report.json. The original report will be saved at report.pdf (the extension will match the original, it may not be .pdf). The text from the report will be extracted to report.txt.

Common options

Every scraper will accept the following options:

  • --year: A YYYY year, only fetch reports from this year.
  • --since: A YYYY year, only fetch reports from this year onwards.
  • --debug: Print extra output to STDOUT. (Can be quite verbose when downloading.)
  • --dry_run: Will scrape sites and write JSON metadata to disk, but won't download full reports or extract text.

Contributing a Scraper

The easiest way is to start by copying scraper.py.template to inspectors/[inspector].py, where "[inspector]" is the filename-friendly handle of the IG office you want to scrape. For example, our scraper for the US Postal Service's IG is usps.py.

The template has a suggested workflow and set of methods, but all your task needs to do is:

  • start execution in a run(options) method, and
  • call inspector.save_report(report) for every report

This will automatically save reports to disk in the right place, extract text, and avoid re-downloading files. options will be a dict parsed from any included command line flags.

You will also need this line at the bottom:

utils.run(run) if (__name__ == "__main__") else None

You should use inspectors.year_range(options) to obtain a range of desired years, and to obey that range during scraping. See an example of creating it and using it.

Scrapers are welcome to use any command line flags they want, except those used by the igs runner. Currently, that's --safe and --only.

Finally, scraper authors are encouraged to note a few things in comments at the top of the scraper:

  • The earliest available year for reports.
  • Any additional command line options you've chosen to support, besides --since and --year.
  • Any notes to pass on to the IG's web team, about how they can make their website better and more reliable.

Report metadata

The report object must be a dict that contains the following required fields:

  • inspector - The handle you chose for the IG. e.g. "usps"
  • inspector_url - The IG's primary website URL.
  • agency - The handle of the agency the report relates to. This can be the same value as inspector, but it may differ -- some IGs monitor multiple agencies.
  • agency_name - The full text name of an agency, e.g. "United States Postal Service"
  • report_id - A string usable as an ID for the report.
  • title - Title of report.
  • published_on - Date of publication, in YYYY-MM-DD format.

Additionally, some information about report URLs is required. However, not all report contents are released: some are sensitive or classified, or require a FOIA request to obtain. Use these fields to handle report URLs:

  • url - URL to the report itself. Required unless unreleased is True.
  • landing_url - URL to some kind of landing page for the report.
  • unreleased - Set to True if the report's contents are not fully released.

If unreleased is True, then url is optional and landing_url is required.

You can also include any other fields you think worth keeping.

The report_id only needs to be unique within that IG, so you can make it up from other fields. It does need to come out the same every time you run the script. In other words, don't auto-increment a number -- if the IG doesn't give you a unique ID already, append other fields together into a consistent, unique ID.

Finally, err towards errors: have your scraper choke and die on unexpected input. Better to be forced to discover it that way, then for incomplete or inaccurate data to be silently saved.

Review process

Suggested instructions for people reviewing new scrapers:

  • Verify that the scraper is getting every possible and useful category of report.
  • Verify that the scraper is using HTTPS, if it's available.
  • Run the scraper for its full archive but without downloading the reports, using --dry_run and --since.
  • Review scraped metadata for a representative sample of years and report types to ensure sanity and quality.
  • Keep an eye out for announcements, press releases, and other non-report data.

If the full dry run ran without errors, and the data looks good and complete:

  • merge the scraper
  • update the pending section of safe.yml with a commented-out line for the scraper.

Adding to safe.yml

The only people who should update safe.yml with uncommented lines are those who run servers managing complete synced archives of IG data.

Before adding a scraper to safe.yml, it's suggested that you allow a full download of the archive to complete (which will also test PDF download and metadata/text extraction code).

Removing from safe.yml

If a scraper is throwing persistent errors, remove it (comment it out) from safe.yml and open a ticket to discuss it.

Ephemeral errors (for example, from connection errors, or other erratically reproducible situations) should be reported as issues first, to be discussed.

Public domain

This project is dedicated to the public domain. As spelled out in CONTRIBUTING:

The project is in the public domain within the United States, and copyright and related rights in the work worldwide are waived through the CC0 1.0 Universal public domain dedication.

All contributions to this project will be released under the CC0 dedication. By submitting a pull request, you are agreeing to comply with this waiver of copyright interest.

Releases

No releases published

Packages

No packages published