Pointblank | Pointblank

5 min read Original article ↗

Logo

Pointblank

Find out if your data is what you think it is.

AI / Agents

Skills
llms.txt
llms-full.txt

Developers

Richard Iannone

Maintainer

Posit, PBC

Posit Software, PBC

Copyright holder, funder

Meta

Requires: Python >=3.10
Provides-Extra: pd, pl, pyspark, generate, mcp, otel, excel, cdisc, bigquery, databricks, duckdb, mysql, mssql, postgres, snowflake, sqlite, docs

Pointblank is a data validation framework for Python that makes data quality checks beautiful, powerful, and stakeholder-friendly. Instead of cryptic error messages, get stunning interactive reports that turn data issues into conversations.

Here’s what a validation looks like (click “Show the code” to see how it’s done):

Show the code 
import pointblank as pb
import polars as pl

validation = (
    pb.Validate(
        data=pb.load_dataset(dataset="game_revenue", tbl_type="polars"),
        tbl_name="game_revenue",
        label="Comprehensive validation of game revenue data",
        thresholds=pb.Thresholds(warning=0.10, error=0.25, critical=0.35),
        brief=True
    )
    .col_vals_regex(columns="player_id", pattern=r"^[A-Z]{12}[0-9]{3}$")
    .col_vals_gt(columns="session_duration", value=20)
    .col_vals_ge(columns="item_revenue", value=0.20)
    .col_vals_in_set(columns="item_type", set=["iap", "ad"])
    .col_vals_in_set(
        columns="acquisition",
        set=["google", "facebook", "organic", "crosspromo", "other_campaign"]
    )
    .col_vals_not_in_set(columns="country", set=["Mongolia", "Germany"])
    .col_vals_between(
        columns="session_duration",
        left=10, right=50,
        pre=lambda df: df.select(pl.median("session_duration")),
        brief="Expect that the median of `session_duration` should be between `10` and `50`."
    )
    .rows_distinct(columns_subset=["player_id", "session_id", "time"])
    .row_count_match(count=2000)
    .col_count_match(count=11)
    .col_vals_not_null(columns="item_type")
    .col_exists(columns="start_day")
    .interrogate()
)

validation.get_tabular_report(title="Game Revenue Validation Report")

Game Revenue Validation Report

Comprehensive validation of game revenue data

Polarsgame_revenueWARNING0.1ERROR0.25CRITICAL0.35
STEP COLUMNS VALUES TBL EVAL UNITS PASS FAIL W E C EXT
#4CA64C 1

col_vals_regex

col_vals_regex()

Expect that values in player_id should match the regular expression: ^[A-Z]{12}[0-9]{3}$.

 player_id ^[A-Z]{12}[0-9]{3}$ ✓ 2000 2000
1.00		 0
0.00		 ○ ○ ○ —
#EBBC14 2

col_vals_gt

col_vals_gt()

Expect that values in session_duration should be > 20.

 session_duration 20 ✓ 2000 1418
0.71		 582
0.29		 ● ● ○
#FF3300 3

col_vals_ge

col_vals_ge()

Expect that values in item_revenue should be >= 0.2.

 item_revenue 0.2 ✓ 2000 1192
0.60		 808
0.40		 ● ● ●
#4CA64C 4

col_vals_in_set

col_vals_in_set()

Expect that values in item_type should be in the set of iap, ad.

 item_type iap, ad ✓ 2000 2000
1.00		 0
0.00		 ○ ○ ○ —
#4CA64C66 5

col_vals_in_set

col_vals_in_set()

Expect that values in acquisition should be in the set of google, facebook, organic, and 2 more.

 acquisition google, facebook, organic, crosspromo, other_campaign ✓ 2000 1975
0.99		 25
0.01		 ○ ○ ○
#AAAAAA 6

col_vals_not_in_set

col_vals_not_in_set()

Expect that values in country should not be in the set of Mongolia, Germany.

 country Mongolia, Germany ✓ 2000 1775
0.89		 225
0.11		 ● ○ ○
#4CA64C 7

col_vals_between

col_vals_between()

Expect that the median of session_duration should be between 10 and 50.

 session_duration [10, 50] ✓ 1 1
1.00		 0
0.00		 ○ ○ ○ —
#4CA64C66 8

rows_distinct

rows_distinct()

Expect entirely distinct rows across player_id, session_id, time.

 player_id, session_id, time — ✓ 2000 1978
0.99		 22
0.01		 ○ ○ ○
#4CA64C 9

row_count_match

row_count_match()

Expect that the row count is exactly 2000.

 — 2000 ✓ 1 1
1.00		 0
0.00		 ○ ○ ○ —
#4CA64C 10

col_count_match

col_count_match()

Expect that the column count is exactly 11.

 — 11 ✓ 1 1
1.00		 0
0.00		 ○ ○ ○ —
#4CA64C 11

col_vals_not_null

col_vals_not_null()

Expect that all values in item_type should not be Null.

 item_type — ✓ 2000 2000
1.00		 0
0.00		 ○ ○ ○ —
#4CA64C 12

col_exists

col_exists()

Expect that column start_day exists.

 start_day — ✓ 1 1
1.00		 0
0.00		 ○ ○ ○ —

2026-07-03 20:26:25 UTC< 1 s2026-07-03 20:26:25 UTC

Notes

Step 7 (pre_applied) Precondition applied: table dimensions [2,000 rows, 11 columns] → [1 row, 1 column].

That’s the kind of report you get from Pointblank: clear, interactive, and designed for everyone on your team.

What is Data Validation?

Data validation makes sure your data is what you think it is before it reaches analysis, reports, or downstream systems. Pointblank gives you a structured way to declare what good data looks like, run those checks against a real table, and communicate the results to technical and non-technical audiences alike. You build a plan with a fluent, chainable API that draws on more than 45 validation methods, set warning, error, and critical thresholds, attach actions that fire when a threshold is crossed, and get back a report anyone on the team can read. Because Pointblank runs on Narwhals and Ibis under the hood, the same plan executes unchanged across Polars, Pandas, DuckDB, Spark, Snowflake, BigQuery, Databricks, PostgreSQL, MySQL, SQLite, and Parquet.

More than a checker

The reporting goes well beyond a single pass or fail. Any step can be opened in a focused step report that drills into the exact rows that failed, and those failing rows can be pulled out as their own table for debugging. The source data can even be split into passing and failing pieces for quarantine or reprocessing. Reports are localized in 40 languages, and results roll up into quality dimensions and a single health score, so completeness, validity, uniqueness, consistency, timeliness, and volume become one number you can watch over time.

A step report drills into the specific rows behind a failing validation step.

Authoring a plan does not have to start from an empty file. Pointblank can draft a starting plan from a natural-language prompt, and from there you can revise and iterate on it in plain English or ask it to suggest improvements. When your standards already live somewhere else, you can bring them with you. Pointblank will import contracts written as JSON Schema or Frictionless, pull column metadata straight from SPSS, SAS, and Stata files, and for clinical work validate against CDISC SDTM and ADaM templates or read a Define-XML specification. It can also model structured missingness, encoding why a value is absent instead of treating every gap identically.

Point Pointblank at a table and let it draft a starting validation plan for you.

Getting all of this into production is where Pointblank earns its place. You can define reusable data contracts and enforce them at both the source and target of a transformation, keep plans as YAML for version control and review, and run the whole thing from a command-line interface inside CI. Pointblank also speaks to machines: it ships an MCP server and llms.txt files for AI agents, emits OpenTelemetry traces and metrics for observability, and can generate synthetic test data when you need something to validate against.