What is a sliver polygon, and why does it break your cadastral layer

If you've ever run a topology check on a parcel layer and gotten back a thousand "polygon too small" flags, you already know what a sliver polygon is, even if nobody ever gave it that name. It's the thin, usually curved strip of a polygon that shows up where two boundary lines that are supposed to be the same line aren't quite the same line.
Where slivers actually come from
A sliver comes from a mismatch between two versions of a boundary that should match exactly and don't.
The most common cause is a mosaic seam. When a boundary layer gets assembled from adjacent tiles, or from two different years of digitising, the shared edge between neighboring parcels gets traced twice, once from each side. Even a trained digitiser tracing the same hedgerow or field edge off two different imagery tiles will land a node a pixel or two off from where they landed it the first time. Snap that edge into a topology and the gap between the two versions becomes its own polygon: a sliver.
Coordinate reprojection does the same thing. A layer digitised in one datum and reprojected into another doesn't always reproject a boundary vertex-for-vertex onto itself, so an edge that was clean before the transform picks up a hairline offset after it. Edge-matching between adjacent registry sheets, between a county layer and a state layer, or between this year's parcel edits and last year's baseline, all produce the same artifact for the same reason: two lines meant to be identical, drawn or transformed slightly differently.
Manual digitising at inconsistent zoom levels adds to it. Trace a field boundary at 1:2,000 on Monday and the adjoining field at 1:5,000 on Tuesday and the vertices along the shared edge won't fall in the same place, because you were working at different pixel scales when you placed them.
Why one thin sliver wrecks more than you'd think
A sliver polygon is small, sometimes a few square meters, sometimes under a square meter. That doesn't make it harmless.
In a parcel registry tied to subsidy payments or loan collateral, every polygon carries an area, and that area usually ties to a payment calculation or a lien value. A sliver sitting between two real parcels either overlaps one of them, which inflates that parcel's area, or opens a gap, which shorts it. Multiply that across a few thousand boundary edges in a county mosaic and you get area errors that don't cancel out, because overlaps and gaps don't fall evenly.
Topology validation tools flag slivers as errors, not warnings, in most GIS platforms, because a sliver usually means the layer fails "must not overlap" and "must not have gaps" rules at the same time, on the same edge. A registry layer that can't pass topology validation can't be published, which means every sliver is a manual fix before the layer ships: open the edge, snap the vertices, rebuild the polygon, re-run the check.
There's a downstream cost too. If the cadastral layer feeds a lending decision or a subsidy claim, an area discrepancy on a specific parcel shows up in an audit, and an auditor who finds a topology error on a payment-bearing parcel will ask why it wasn't caught before the layer shipped.
Cleaning a layer with slivers already in it
Most GIS software has a sliver-removal or "eliminate" tool that merges polygons under a given area threshold into the larger neighbor. That removes the symptom for the current layer, but it doesn't touch the edge-matching problem that created the sliver, so the same cleanup pass runs again on the next update, whenever digitising continues across tiles or years.
The edge-matching problem mostly goes away when boundaries get traced once, consistently, across a whole region in a single pass instead of tile by tile or hand-off by hand-off. That's the approach behind a parcel boundary layer built straight from satellite imagery, generated per region per year without the seam-by-seam digitising that creates slivers in the first place.
If your mosaic is fighting topology errors every update cycle, it might be worth seeing what a cleaner source layer looks like.