[Interactive demo]
Walk the pipeline yourself.
Sample data, in your browser. A faithful simulation of the real flow. Approve or reject at step four and watch the score change.
.png){kind=logo}
.png){kind=logo}
Quality checks that run inside the pipeline rather than after it. The agent implements stage gates at each hop (source, staging, warehouse, mart and consumption), so a bad load is stopped before it reaches anybody, instead of being discovered in a dashboard a week later.
Sample data, in your browser. A faithful simulation of the real flow. Approve or reject at step four and watch the score change.
Eight dimensions across every column, five of them shown here. Nothing is written; this is a read.
4,812 rows · 8 columns · 2 columns flagged · 2 for review
Table and column descriptions written from the profile and the column context. Every one is a draft until a steward accepts it.
1 table + 4 columns described · awaiting steward
Seven rules proposed from the profile. Each carries the proof that produced it, so approval is a judgment call rather than a leap of faith.
customer_id, Must not be null
Completeness
0.0% null across 4,812 rows. Already clean, worth enforcing as a hard constraint before it drifts.
customer_id, Must be unique
Uniqueness
4,812 distinct values across 4,812 rows. Uniqueness holds today; the rule locks it.
tax_id, Must match the tax ID format
Validity
97.2% of non-null values match the pattern. 118 do not, mostly nine-digit strings missing the hyphen.
tax_id, Null rate must stay under 5%
Completeness
12.2% null against a 5% domain threshold. 587 rows. This is the largest single rule failure in the table.
email, Must be a parseable email address
Validity
4,359 of 4,422 non-null values parse. 63 do not: trailing semicolons and two addresses in one field.
status, Must be one of the accepted values
Validity
Exactly 4 distinct values observed, all within the expected set. Cardinality of 4 on 4,812 rows reads as a controlled vocabulary.
country_code, Must exist in the country reference set
Consistency
47 distinct codes. Two (XK, AN) are absent from ref.country, affecting 12 rows. Neither is a current ISO 3166-1 code.
7 rules · 4 quality dimensions
Nothing reaches production without this step. Uncheck anything you would not stand behind. In the real product, the SQL is editable too.
customer_id, Must not be null
Completeness
0.0% null across 4,812 rows. Already clean, worth enforcing as a hard constraint before it drifts.
customer_id, Must be unique
Uniqueness
4,812 distinct values across 4,812 rows. Uniqueness holds today; the rule locks it.
tax_id, Must match the tax ID format
Validity
97.2% of non-null values match the pattern. 118 do not, mostly nine-digit strings missing the hyphen.
tax_id, Null rate must stay under 5%
Completeness
12.2% null against a 5% domain threshold. 587 rows. This is the largest single rule failure in the table.
email, Must be a parseable email address
Validity
4,359 of 4,422 non-null values parse. 63 do not: trailing semicolons and two addresses in one field.
status, Must be one of the accepted values
Validity
Exactly 4 distinct values observed, all within the expected set. Cardinality of 4 on 4,812 rows reads as a controlled vocabulary.
country_code, Must exist in the country reference set
Consistency
47 distinct codes. Two (XK, AN) are absent from ref.country, affecting 12 rows. Neither is a current ISO 3166-1 code.
7 of 7 approved, nothing will run
Approve at least one rule
Approved rules run against the table. Failures produce downloadable rejected records; scores register to the catalog.
97.7%
Overall score
93.9%
Completeness
100.0%
Uniqueness
98.7%
Validity
99.8%
Consistency
No rules approved, so nothing ran.
Back to approval7 rules executed · 780 rejected rows · scores registered to the catalog
No rules executed · nothing registered to the catalog
Profiles any table, drafts the metadata, writes the rules, and proves each one.
Watches quality scores over time and tells you what is drifting, and what to do.
Profiling and rule generation native to Microsoft Fabric and OneLake.
The DataIQ pattern, running natively on Redshift and Athena.
Reads a failed rule, works out why, and routes it to the person who owns it.
Discovery is read-only, takes five days, and costs nothing. We assess your environment and tell you honestly whether this agent is worth your time.
Book a Consultation.png){kind=logo}
