verdict

A high-performance data validation library written in Rust, with Python bindings via PyO3.

Verdict lets you define typed schemas and validation rules for tabular data, then run them against datasets loaded from CSV files or constructed in memory. All validation logic lives in pure Rust; Python integration is a thin binding layer on top.

---

Features

• 14 constraint types — null checks, uniqueness, numeric comparisons, string patterns, set membership, and more
• Column-to-column comparisons — validate that one column's values are greater than another's
• Null-aware operations — all comparisons preserve null information without false failures
• 4 typed column kinds — Int, Float, Str, Bool, each with appropriate operations
• CSV loading — feature-gated CSV reader with typed schema enforcement
• Python bindings — clean PyO3 API with a fluent RuleBuilder
• CLI binary — zero-dependency binary for CI/CD pipelines, JSON output, exit codes
• Zero I/O in core — verdict-core has no filesystem dependencies by default

---

Architecture

verdict-core  ←  verdict-py
verdict-core  ←  verdict-cli

Crate	Description
verdict-core	Pure Rust validation engine. CSV support behind the csv feature flag.
verdict-py	PyO3 bindings that expose verdict to Python.
verdict-cli	CLI binary for CI/CD pipelines. Reads CSV + JSON schema, outputs results.

---

Installation

cd crates/verdict-py
pip install maturin
maturin develop

---

Quick Start

from verdict_py import Dataset, Schema, DataType, RuleBuilder, py_validate

schema = Schema([
    ("user_id", DataType.integer()),
    ("score",   DataType.float()),
    ("country", DataType.string()),
])

dataset = Dataset.from_csv("data.csv", schema)

rules = [
    *RuleBuilder("user_id").not_null().unique().build(),
    *RuleBuilder("score").between(0.0, 100.0).build(),
    *RuleBuilder("country").is_in(["US", "UK", "DE", "FR", "JP"]).build(),
]

results = py_validate(dataset, rules)

for r in results:
    status = "PASS" if r.is_passed else "FAIL"
    print(f"[{status}] {r.column} / {r.constraint} — {r.failed_count} failures")

---

Constraints

Constraint	Applies to	Description
not_null()	All	No null values in the column
unique()	All	All values are distinct
gt(value)	Int, Float	Every value > threshold
ge(value)	Int, Float	Every value >= threshold
lt(value)	Int, Float	Every value < threshold
le(value)	Int, Float	Every value <= threshold
equal(value)	Int, Float, Str	Every value == target
between(min, max)	Int, Float	min <= value <= max
matches_regex(pattern)	Str	Value matches regex pattern
contains(substr)	Str	Value contains substring
starts_with(prefix)	Str	Value starts with prefix
ends_with(suffix)	Str	Value ends with suffix
length_between(min, max)	Str	String length in [min, max]
is_in(values)	All	Value is a member of the given set

Pass a column name instead of a literal to compare two columns row-by-row:

Rule("high", Constraint.gt("low"))  # validates high > low for every row

---

CLI

cargo build --release -p verdict-cli
./target/release/verdict-cli data.csv schema.json

Schema format (schema.json):

{
  "columns": [
    { "name": "user_id", "dtype": "int", "constraints": [
      { "constraint": "not_null", "value": true },
      { "constraint": "unique",   "value": true }
    ]},
    { "name": "score", "dtype": "float", "constraints": [
      { "constraint": "between", "value": [0, 100] }
    ]},
    { "name": "country", "dtype": "str" }
  ]
}

• Columns without constraints are still required for schema — they're just not validated
• Exit code 0 — all constraints pass; 1 — at least one fails
• Default output is JSON; --format text for human-readable

# CI usage
verdict-cli data.csv schema.json && echo "data quality OK"

---

CSV Loading

Dataset.from_csv(path, schema) parses each column according to its declared type. Empty cells become None. Booleans accept true/True/TRUE/1 and false/False/FALSE/0.

---

Performance

CSV loading benchmarks (Criterion, release build, Apple Silicon):

Dataset	Load time
10k rows	~1.5 ms
100k rows	~15.5 ms
1M rows	~173 ms

---

Development

cargo build --all-features   # build
cargo test --all-features    # test

cd crates/verdict-py
maturin develop && pytest tests/

---

License

MIT