- Expand benchmark surface — add benchmarks for Panel analytical methods and the
TemporalSeriesgeneric path; decide coverage for storage backends. - Parquet I/O —
read_parquet/write_parquetbehind an optionalparquetfeature flag, with tests, benchmarks, and an example. -
RollingSeriesmethod expansion — add.sum(),.min(),.max(),.std()to complement the existing.mean(); keep the same lazy-handle design. -
ChunkedBackend— implement the stub incrates/storage/chunked.rs; fixed-size chunks suited for append-heavy workloads; full test coverage mirroringColumnarBackendandRowBackend. - Panel I/O — wide-format CSV reader/writer for
Panel(one column per symbol, shared index column), so multi-asset data can be loaded without constructing each series by hand. - NaN handling utilities —
forward_fill,backward_fill,fill_with(value)onTimeSeries. - Serde support — optional
serdefeature flag; deriveSerialize/DeserializeforTimeSeriesandPanel.
- Forecasting module (
crates/forecasting/) — ARIMA, exponential smoothing; large enough scope to warrant its own crate boundary. - Index-based slicing and resampling —
slice(from, to), frequency conversion; requires a design decision on index-sorted invariant.