Badge pipeline

Evaluating a criterion

Every criterion page states what a package must show for the criterion to be met, partially met or not met. The note under each rating lists the sources it is based on; the #IDs link to the Sources page.

A criterion is not applicable only under the four conditions C1–C4 defined below. Anything else that is missing counts as not met.

N/A package condition Definition
C1 (simulation) All data are generated by the code itself (e.g. a simulation). Data the code downloads or scrapes do not count.
C2 (all data in package) All data needed are in the package (included, or deposited separately under a PID/DOI) or generated by the code.
C3 (no tables/figures) The paper has no tables or figures.
C4 (no numerical results) The paper reports no numerical results, only figures.

Rules of the badge system

  1. In order to get a badge, a reproduction package must fully meet all applicable core criteria of that tier and of every lower tier.
  2. Each reproduction package gets a summary of how many criteria (core and recommended) are met, partially met, not met, and not applicable, and a badge if rule 1 is met.
  3. If not all input datasets are in the package and there is no third-party certification, only the Tier 1 criteria are assessed, so the package can get at most Bronze. The summary then covers only the Tier 1 criteria; findings on the other data-independent criteria can be reported as comments.
  4. If not all input datasets are in the package but the package has third-party certification, the data-dependent, core criteria (path resolution, coding error handling, numerical output match, and table and figure match) are rated from the certificate or the signed letter.
    • A criterion is fully met only if the certificate or a signed letter, issued by an independent party, explicitly confirms it, e.g. that the code ran to completion (Silver) or that all results were reproduced (Gold).

Order of checking

A reproduction package contains the manuscript, a README, the code (R or Stata) and a codebook (not needed if all data are generated by the code), plus, optionally, the datasets (included in the package, or deposited separately under a PID/DOI) and a log file (Stata). It is checked in four steps:

  1. Data situation check. This sorts the package into one of four routes:
    • Route 1a: all datasets are generated by the code (not downloaded). All tiers are possible, and N/A conditions C1 and C2 apply.
    • Route 1b: all input datasets are in the package (included or under a PID/DOI). All tiers are possible, and N/A condition C2 applies.
    • Route 1c: not all input datasets are in the package, but there is third-party certification or a signed letter from an independent party. All tiers are possible.
    • Route 1d: not all input datasets are in the package, and there is no certification. The package can get at most Bronze.
  2. Data-independent criteria check. These can be checked without the data being in the package:
  3. Data-dependent criteria check. These can only be checked with the data, so this step depends on the route:
  4. Award the badge. Apply rule 1:
    • No badge: not all applicable Tier 1 core criteria are fully met.
    • Bronze: all applicable Tier 1 core criteria are fully met, but the package is on route 1d or not all applicable Tier 2 core criteria are fully met.
    • Silver: all applicable Tier 1 and Tier 2 core criteria are fully met, but not all applicable Tier 3 core criteria.
    • Gold: all applicable core criteria of all three tiers are fully met.

Within criteria checks (step 2 and 3), always check criteria with a lower check order first. The number counts the prerequisite steps above a criterion (0 = no prerequisite coded, — = no prerequisite links coded at all). Each criterion page lists what it builds on and what it is a prerequisite for; the Dependency graph shows all links at once.

Visual illustration of the pipeline