Veridelta

Semantic diffing for data pipelines. Define expected variance, validate intentional changes, and detect regressions with confidence.

Veridelta is a declarative data comparison engine powered by Polars. It applies semantic and mathematical rules to filter out expected pipeline noise, such as floating-point drift or formatting changes, and programmatically isolates true data regressions at scale.

Documentation & API Reference

Installation

uv add veridelta

Warehouse and lakehouse drivers are optional extras (snowflake, databricks, delta, iceberg, or all):

uv add 'veridelta[snowflake]'

YAML examples, routing rules, and time travel live in the configuration guide.

Usage

Veridelta supports configuration-driven CLI execution and programmatic Python API integration.

CLI Configuration (YAML)

Declare comparison rules and tolerances via YAML:

# veridelta.yaml
primary_keys: ["transaction_id"]
schema_mode: "intersection"

source:
  path: "legacy_system.parquet"
  format: "parquet"

target:
  path: "modern_system.parquet"
  format: "parquet"

rules:
  - column_names: ["grand_total"]
    relative_tolerance: 0.01

  - column_names: ["contact_number"]
    regex_replace: {"[^0-9]": ""}

Execute the comparison:

veridelta run --config veridelta.yaml

Python API

Integrate directly into programmatic workflows (e.g., Airflow, Databricks). The DiffEngine natively consumes Polars LazyFrame objects for memory-safe, big-data execution.

import polars as pl
from veridelta import DiffConfig, DiffRule, DiffEngine

source_lazy = pl.scan_parquet("legacy_data.parquet")
target_lazy = pl.scan_parquet("modern_data.parquet")

config = DiffConfig(
    primary_keys=["user_id"],
    default_treat_null_as_equal=True,
    rules=[
        DiffRule(
            pattern="^AMT_.*",
            absolute_tolerance=0.05
        )
    ]
)

engine = DiffEngine(config, source_lazy, target_lazy)
summary = engine.run()

if not summary.is_match:
    print(f"Regression detected: {summary.changed_count} rows differ.")

Core Capabilities

Structural Alignment: Map legacy column names to modern schemas automatically.
Semantic Normalization: Coerce string markers to nulls, standardize whitespace, and cast types dynamically before mathematical comparison.
Warehouse Pushdown: Compile comparison SQL for Snowflake and Databricks so diffs run in-warehouse (changed, added, and removed counts) without extracting full tables.
Lakehouse Native: Scan Delta Lake and Apache Iceberg tables as unevaluated Polars LazyFrames, including optional version and snapshot time travel.
Discrepancy Artifacts: Export isolated Parquet files detailing added, removed, and changed records for downstream auditing.

Roadmap

Upcoming work:

Advanced Heuristics: Fuzzy string matching and ML-driven schema mapping.
Reporting: Interactive HTML diff dashboards and CI/CD status checks.
Additional warehouses: BigQuery pushdown.

View Detailed Roadmap

License

Distributed under the Apache 2.0 License.

Metadata

Release files for veridelta 0.3.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for veridelta 0.3.0
File	Size	Uploaded
veridelta-0.3.0.tar.gz	338.6 kB	Sep 8, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for veridelta 0.3.0
File	Interpreter	ABI	Platform	Reset
veridelta-0.3.0-py3-none-any.whl 39.5 kB Sep 8, 2026	Python 3	none	any	Details

Total release size: 378.0 kB

Release files / veridelta-0.3.0.tar.gz

Download URL	veridelta-0.3.0.tar.gz
Size	338.6 kB
Tags	Source
SHA-256 checksum How to use checksums	`fcfb85558c4531fcf0954d9e0ebd775d3c4452ea85838607f9fdd9bdced44a67`
BLAKE2b-256 checksum How to use checksums	`156639006f637f99cf5a43a9e0bf9abba67ebb71dd2ac47b6b77a73ff5cca209`
Upload date	Sep 8, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`uv/0.12.10 {"installer":{"name":"uv","version":"0.12.10","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}`

Release files / veridelta-0.3.0-py3-none-any.whl

Download URL	veridelta-0.3.0-py3-none-any.whl
Size	39.5 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`aafce6bd01eae49c0b1a6b1fff15267de921f976381b2daab1e4dac643748dc6`
BLAKE2b-256 checksum How to use checksums	`dd4ffc6e24197ded540f6eaf8266c4b9b61156dbe33992fd36db8b41be6044a0`
Upload date	Sep 8, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`uv/0.12.10 {"installer":{"name":"uv","version":"0.12.10","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}`