The map scored 87%. Would a reviewer believe it?

Paste your layer manifest, your validation confusion matrix and your methods text. Before you sign in, this page reads all three properly — it recomputes overall accuracy, kappa and every per-class figure from the matrix's own margins, converts a degree-sized pixel to metres at its own latitude, walks the extents for real coverage, runs the area-weighted Olofsson estimator with confidence intervals, and checks whether your validation folds share neighbours with your training folds. Then three lanes work the study itself.

Both examples ship with a saved model run for every lane, so you can see all three outputs end to end without signing in and without spending a credit.

The layer manifest — one block per layer, or a delimited table nothing pasted yet

Recognised keys: layer/name, kind, crs/epsg, pixel_size/resolution, extent/bbox (or xmin…ymax columns), bands, dtype, nodata, geometry, features, size, role. Drag a file in, or Read in your browser. Nothing uploads until you run a lane.

The validation confusion matrix — rows are map classes, columns are reference classes nothing pasted yet

Also accepted: map,reference,count lines; an orientation: rows = reference line to transpose it; an areas (ha): block instead of an area column. A total row or column is recognised and set aside, never counted as a class. The mapped area of each class is what makes the area-weighted estimate and its intervals computable. Drag a file in, or Without areas, the weighted figures report as not computable rather than guessed.

The methods and results text — how the sample was drawn, how it was validated, what was reported nothing pasted yet

Read sentence by sentence with clause-scoped negation, so “spatial cross-validation was used, no random folds” reads the first phrase as on and only the second as denied. Drag a file in, or A metric the text says was not computed is recorded as denied, not as absent.

What kind of product is it? — only fills a silence in the methods text

How was the validation sample drawn? — only fills a silence

What is the accuracy figure used to claim? — the lanes check the design against this, not against a generic standard

Anything the three blocks do not say — the reviewer's ask, the deadline, what you already tried

Paste the study to price the run.

The model's raw reply

The structured parse failed twice, so here is the reply verbatim rather than nothing.

The map scored 87%. Would a reviewer believe it?

Instant read — free, in your browser, no model

Where the lane and the browser disagree

Findings

Every free-read flag, accounted for

What you told it, checked against the study

Not assessable from what was supplied

Summary

The model's raw reply

Your past assessments

How this was produced