Check it yourself
Every issue commits to four things, so you can verify rather than trust.
| 1 | Collection code. Scripts that re-download every raw input from its public source, with URL, date and checksum. |
|---|---|
| 2 | Analysis code. Scripts that turn the collected data into the numbers printed in the article, with a check that re-derives them. |
| 3 | Figure code. Scripts that redraw each figure from the processed data, in both languages. |
| 4 | These pages. Full-resolution figures, and interactive versions where the printed one had to be simplified. |
What to do
- Open the issue's repository from the Issues page.
- Follow its README. Where the data is small enough, the intermediate data is included, so checking takes a few commands and no account.
- Run the number check. Each number prints PASS or FAIL against the article.
If something does not match
A FAIL, or a figure that differs, is a finding. Open an issue on that issue's repository, titled Mismatch with what you ran and what you got. Reproduced it cleanly? Open one titled Reproduced. Confirmed errors go in the issue's bug log, and the article is corrected.
What cannot be checked
Each README says plainly which statements rest on inputs that are not included. Honest gaps are listed, not hidden.