Validation
How these tools were tested
Every tool in this toolkit is validated against the established software for its discipline. Each scenario is run twice, once by this toolkit and once by the reference package, and the two sets of results are compared value by value. The full comparison is published below, including every value where the two do not agree.
105
scenarios
9
reference packages
2210
values agreeing
14
differing
Validation coverage by tool
| Tool | Validated against | Scenarios | Values agreeing | Detail |
|---|---|---|---|---|
| Descriptive Statistics | USEPA ProUCL, R EnvStats / nortest, R NADA | 33 | 472 / 472 | View scenarios |
| Trend Analysis | R EnvStats / nortest, GWSDAT, USEPA ProUCL, pymannkendall not in this run: GSI Mann-Kendall Toolkit | 16 | 318 / 320 | View scenarios |
| Background Threshold | USEPA ProUCL, R EnvStats / nortest | 9 | 40 / 42 | View scenarios |
| Normality Testing | R EnvStats / nortest, USEPA ProUCL | 4 | 34 / 38 | View scenarios |
| Outlier Identification | USEPA ProUCL, R EnvStats / nortest | 4 | 18 / 24 | View scenarios |
| Hypothesis Testing | R EnvStats / nortest, USEPA ProUCL | 4 | 15 / 15 | View scenarios |
| Sample Size and DQO | USEPA ProUCL, R EnvStats / nortest | 7 | 12 / 12 | View scenarios |
| Spatial Plume Mapping | Shepard inverse-distance weighting and the least-squares plane, Golden Software Surfer not in this run: GWSDAT | 6 | 1064 / 1064 | View scenarios |
| Groundwater Contour Mapping | Shepard inverse-distance weighting and the least-squares plane, Published worked example, Golden Software Surfer | 5 | 52 / 52 | View scenarios |
| Coordinate Conversion | PROJ / pyproj | 3 | 14 / 14 | View scenarios |
| Sieve Analysis and PSD | ASTM D2487 / D6913 (implemented from the published criteria) | 11 | 171 / 171 | View scenarios |
The reference packages
PROJ is the library behind almost every geographic information system in use, so for coordinate conversion it is the reference rather than a cross-check.
NADA is the reference implementation of regression on order statistics, written by the author of the method, so for ROS it is the primary reference rather than a cross-check.
classification and gradation criteria; no vendor build
· ASTM International
Shepard inverse-distance weighting and the least-squares plane
published methods; no vendor build
· Shepard (1968); ordinary least squares
How to read this
- Every comparison has a stated tolerance. Counts and categories must match exactly. Computed statistics are held to a relative band, and any wider band is stated on the row it applies to, with the reason.
- Differences are published, not hidden. Where a reference package and this toolkit genuinely disagree, the scenario page shows both numbers and explains why.
- A difference gets a third opinion, where one exists. Where this toolkit and a reference package disagree, we do not simply assert that we are right. We run the same scenario through a third, independent implementation and publish which way it falls - whether it backs this toolkit, backs the other package, or disagrees with both. For a few quantities no third implementation exists to ask. We say so on the scenario, and give our reasoning instead of implying corroboration we do not have.
- Reference versions are pinned. A validation result is a claim about a specific pair of versions, so the version of every package used is recorded above.
Disclaimer: evidence generated 2026-08-26 from a full comparison run against the versions named above. Results are provided without warranty and professional judgement remains your responsibility.