Profiling

Profile a table before you write its rules. A profile is a summary of what is actually in the data — per column: type, how much is empty, how many distinct values, the range — plus findings: what a person should look at. Only counts and summaries leave the engine, never rows.

Try it in Colab: profile a claims table, agree its rules as a contract, and test every row — no warehouse needed.

From the command line

# A file or folder (local, s3://, abfss://, gs://); the format is detected
lakelogic profile landing/orders/ --findings

# A table, by SQL pushdown
lakelogic profile main.sales.orders --engine databricks --findings
lakelogic profile orders --engine duckdb --database lake.duckdb

# A large table: sample it
lakelogic profile main.sales.orders --engine databricks --sample-percent 10 --findings

--findings adds the checks; --contract marks a contract's personal-data columns as sensitive so their values are never shown; --output writes the JSON to a file. See lakelogic profile --help.

From Python

import duckdb
from lakelogic.core.profile_assess import assess_table

con = duckdb.connect("lake.duckdb")                  # or a SparkSession, or a SQLAlchemy engine
entry = assess_table("orders", con)                  # add contract="orders.yaml" to check its rules
print(entry["finding_counts"])                       # {'confirmed': 0, 'potential': 2, 'gap': 1}
for f in entry["findings"]:
    print(f["kind"], "—", f["title"], "—", f.get("evidence"))

assess_tables([...], connection) profiles several tables at once. For a large table pass settings={"mode": "sample", "sample_unit": "percent", "sample_percent": 10}.

Reading the findings

Kind	Meaning
confirmed	A rule you agreed (in the contract) failed.
potential	An observation to check with the table's owner — e.g. a repeated id, a mostly empty column. It may be legitimate.
gap	A check that is missing — no contract, no key, no freshness target, a categorical column without accepted values.

Profile without a contract first; agree the rules with the owner and write them into the contract; profile again with the contract, and the agreed rules become confirmed checks. Then run the contract on every load (lakelogic run) so each change is tested.