Coral

Coral is a typed dataframe shell for Chelis.

The central design choice is that numeric columns are tensors. That lets Coral reuse Chelis tensor lowering, fusion, and backend execution instead of building a separate dataframe execution engine.

Modules

Module	Purpose
Core	Typed column accessors, GPU-friendly numeric columns
Frame	AD-oriented dataframe pipelines (runtime-executed pandas parity lane)
GroupBy	Split-apply-combine over typed frames
Join	Equi-join and key-matching combinators
Io	Read/write for CSV, Parquet, and Arrow IPC
Reshape	Pivot, melt, stack, unstack
Window	Rolling sum/mean/std/min/max and EWM (runtime-executed pandas parity lane)
AsOf	Point-in-time joins for ordered time-series alignment