v3.3.0

Changed

• numericQuantity return type is now inferred from the options object: verbose: true yields NumericQuantityVerboseResult, bigIntOnOverflow: true yields number | bigint. Options typed as a plain NumericQuantityOptions variable now infer the full union number | bigint | NumericQuantityVerboseResult and must be narrowed; use satisfies NumericQuantityOptions instead of : NumericQuantityOptions to retain literal inference.
• isNumericQuantity accepts unknown instead of string | number.

Fixed

• Added "sideEffects": false to package.json.
• vulgarFractionsRegex no longer matches }.
• bigIntOnOverflow now evaluates decimals, exponents, fractions, mixed numbers, and percentages exactly, rounding half-up instead of discarding the tail.
• round is now applied to the value as written, before percentage division, on all code paths. Previously '1%' and '1.0%' could produce different results.
• The percentage option now applies to Roman numeral results (e.g. 'L%' → 0.5).
• verbose.trailingInvalid is now the complete unconsumed suffix of the original input. Previously, when decimalSeparator was ",", the internal separator marker could leak into it and part of the suffix could be dropped (e.g. '10,00.,0' reported '&'; it now reports '.,0').
• symbol input returns NaN instead of throwing. In verbose mode, input is the stringified symbol (e.g. 'Symbol(1)').
• Input that cannot be coerced to a string (null-prototype objects, throwing toString/valueOf) returns NaN instead of throwing TypeError. In verbose mode, input is an empty string.
• bigIntOnOverflow no longer throws RangeError for absurdly large exponents (e.g. '1e' + '9'.repeat(400)). Exponents beyond ±10,000 fall back to the number path, yielding Infinity/0.
• verbose.trailingInvalid is now populated for multi-comma input when decimalSeparator is ',' and allowTrailingInvalid is false, matching the behavior of every other trailing-invalid path.
• Non-finite round values (NaN, ±Infinity) are now treated as false (no rounding) instead of silently rounding to 0 decimal places. Negative finite values still clamp to 0.
• Very large round values no longer produce incorrect results. The rounding factor was built by string concatenation, so values in exponential range were misread (e.g. round: 1e21 became "1e1e+21" → a factor of 10, rounding '1.23456' to 1.2). Factors beyond Number.MAX_VALUE now fall back to no rounding, as does a rounding operation that would overflow an otherwise finite value (e.g. '1.5e300' with round: 308 returned NaN, now returns 1.5e300).
• Currency and percentage affixes are now stripped in any order (e.g. '100€%' → 1, matching '50%€'). At most one % is stripped, so '50%%' is still NaN.

Full Changelog: v3.2.2...v3.3.0