True if the distinct-cap was hit → uniqueness NOT provable from this scan.
Distinct non-null values seen (exact, up to the cap).
Generously-bounded inferred type (both dialects), string width covering MaxObservedLength.
TRUE maximum observed string length across EVERY scanned row (rsuplan "largest string" stat). Tracked separately from the capped SampleValues — the sample may miss the widest value, so string sizing must widen to this, not the sample max.
Rows where the key was present + non-null.
Bounded sample of observed values (for type inference).
Total rows scanned.
Per-column evidence gathered from the stream.