jeondaehong/lake-sift data diff

Diff two datasets (Parquet / Iceberg / Delta) at the cell level and gate a PR or pipeline on the result.

Trust Signals

name	description	required	default
left	Base (before) source — Parquet path, iceberg:<catalog>/<ns>.<table>[@<snap\|ref>], or delta:<path>[@<ver>].	yes	—
right	Compared (after) source — same forms as `left`.	yes	—
key	Row identity key (comma-separated column names).	yes	—
columns	Compare only these columns (comma-separated).	no	""
exclude	Columns to exclude from comparison (comma-separated).	no	""
tolerance	Numeric tolerance — cells are equal when abs(l - r) <= tolerance.	no	""
ignore-case	Case-insensitive comparison for text columns.	no	false
allow-duplicates	Allow duplicate keys instead of erroring.	no	false
sample	Max rows to show per change kind in the report.	no	""
extras	lake-sift extras to install for the source formats used, e.g. 'iceberg', 'delta', or 'iceberg,delta'. Empty installs Parquet-only.	no	""
version	lake-sift version to install (a PyPI version specifier). Defaults to the latest release.	no	""
fail-on-diff	Fail the step (non-zero exit) when differences are found. A comparison error always fails.	no	true
comment	Post/update a sticky comment with the diff report on the pull request (requires pull-requests: write and a PR context).	no	false

name	description
diff	'true' when differences were found, 'false' when the datasets are identical.
exit-code	The lake-sift exit code (0 identical, 1 differences, 2 error).
report	Path to the Markdown report file.