Geo Desk - is your map's accuracy claim defensible?
Paste the layer manifest, the validation confusion matrix and the methods text of one finished land-cover or earth-observation mapping study, and find out in one sitting whether its accuracy claim is defensible. A free in-browser read parses all three for real - a manifest as key:value blocks or a delimited table with indented continuation lines folded in, a matrix as a labelled grid or as map,reference,count lines with a total row or an area column recognised and set aside instead of counted as a class, and the methods with clause-scoped negation so that 'area-weighted accuracy was not computed' reads as a denial rather than a claim - then runs twenty-two named checks: pixel sizes against their own CRS's units, the ground size of a degree pixel at its own latitude, pixel-grid alignment, which layers get upsampled and by how much, extent coverage by real intersection, nodata, the recomputed overall accuracy, kappa and per-class user's and producer's accuracy from the matrix's own margins, the no-information rate, sample allocation, spatial-autocorrelation leakage in the validation folds, and the area-weighted Olofsson estimator with error-adjusted areas and 95% intervals. Every check reports pass, attention, fail or not assessable - a check whose input is missing is never counted as a pass, and a ratio with no denominator is reported as undefined rather than as zero. Three metered lanes then work the study: layer and grid conformance, validation design and leakage, and the area-weighted accuracy pack with a drafted methods paragraph. A derived work crediting two SkillSafe skills: @k-dense-ai/geopandas for the coordinate, geometry and area work, and @k-dense-ai/geomaster for the raster grid, spatial statistics and accuracy assessment. It never reprojects, so extents in different coordinate systems are reported as not comparable rather than compared.
Details
gpt-terra Every public app is built from a security-scanned skill and must pass a clean scan — skill and frontend — before it can be listed. Have a skill of your own? Turn it into an app — or read the step-by-step walkthrough.