Every geo format converter, this site's included, ends the same way: "Conversion complete." That message is true and also incomplete. GPX has no polygon concept. KML keeps only a name and description per feature. WKT keeps no properties at all. A plain CSV holds one point per row. None of that is a bug — it's an inherent limit of the target format — but it means a conversion can succeed and still quietly leave out exactly the data you needed downstream.
What actually gets lost, by format
| Target | What survives |
|---|---|
| GeoJSON | Everything — every geometry type and every property. |
| KML | Any geometry type, but only name and description as properties. |
| GPX | name only, and Polygon/MultiPolygon are dropped entirely (GPX has no polygon). |
| CSV | Every property as a column, but only Point/MultiPoint geometry — a row is one coordinate pair. |
| WKT | Every geometry type, but no properties at all — WKT has no property mechanism. |
This table is worth knowing, but it's more useful as something a tool checks against your actual file than as something to memorize and apply by hand.
Inspecting a conversion automatically
The Geo Conversion Loss Inspector runs the real conversion — the same Geo Format Converter logic — reads the output back in through the same parser a real import would use, and diffs it directly against the source: which geometry types disappeared, which property keys didn't survive, how the feature count changed. There's no separate "predict what will happen" step distinct from this, because for a small, synchronous, text-based conversion, actually doing it and checking the result is both simpler and more accurate than a lookup table.
Comparing any two datasets, not just a conversion pair
Sometimes what you have isn't a source-and-its-conversion but two independently-produced files — an old export and a new one, a raw dataset and a version someone ran through the Geo Reprojection & Repair Workbench. The Geo Dataset Compare tool runs the same kind of diff without assuming one file came from converting the other: feature count, geometry type changes, and property schema, plus one thing a plain feature-count check would never catch — a property that's still present on both sides but has quietly gone emptier, present and populated in the older file, increasingly blank in the newer one.
The recommended order
For a conversion you actually need to trust: inspect the source with the Geo QA & CRS Inspector first, convert it (directly through the Loss Inspector, which does the diff for you at the same time), and if the result still needs a second look — a repair pass happened in between, say — run the before-and-after through Dataset Compare. Each step catches something the others don't: QA catches problems in the source itself, the Loss Inspector catches what a specific conversion changes, and Dataset Compare catches drift between any two versions regardless of how the second one was produced.
Where each step happens
- Checking the source first — the Geo QA & CRS Inspector catches CRS and geometry problems before you convert anything.
- Converting and seeing what changed — the Geo Conversion Loss Inspector is the tool this guide is built around for a source-to-conversion pair.
- Comparing two independent files — Geo Dataset Compare for a before/after that didn't come from a single conversion step.