What a reader does with a 64-bit integer column (Arrow Int64 / UInt64, Parquet INT64, driver BigInts):
'error' (default): values within ±(2^53 − 1) become an exact f64 column; if any value is outside, the read
fails with PrecisionLossError naming the column, row and value — nothing is silently rounded;
'string': the whole column becomes utf8 decimal strings — exact for every value (IDs, join keys);
'number': explicit lossy opt-in — the nearest double (2^53 + 1 → 2^53).
columna has no i64 dtype; see docs/compatibility.md.
What a reader does with a 64-bit integer column (Arrow Int64 / UInt64, Parquet INT64, driver BigInts):
'error'(default): values within ±(2^53 − 1) become an exact f64 column; if any value is outside, the read fails withPrecisionLossErrornaming the column, row and value — nothing is silently rounded;'string': the whole column becomes utf8 decimal strings — exact for every value (IDs, join keys);'number': explicit lossy opt-in — the nearest double (2^53 + 1 → 2^53). columna has no i64 dtype; see docs/compatibility.md.