Governance that is not a separate product
Most governance tools are catalogues that observe a pipeline they cannot see into, and drift from it within a quarter. In DataLens the catalogue and the pipeline are the same system, so the lineage is a record of what ran rather than a diagram someone maintained.
What governance covers here
Profiling
Types, null rates, cardinality, distributions and candidate keys, recomputed as the data changes rather than captured once at onboarding.
Quality rules
Expectations expressed against a dataset and evaluated on every run, with a history of when each one last passed.
Lineage
Where a dataset came from and what was applied to it, at column level, derived from the steps that actually executed.
PII detection
Columns that look like personal data are flagged automatically, so a classification decision can be made early.
Policies
Rules about what may be done with which data, enforced by the platform rather than documented in a wiki.
Stewardship
Ownership recorded per dataset, with the review and approval steps that go with it.
Data dictionary
Generated from the profile and then edited, so it starts complete instead of starting empty.
AI operations
Cost, tokens, latency, failure taxonomy, quarantine and cache behaviour for every AI step — the governance question about the platform itself.
Why lineage from the pipeline is different
Lineage that is documented by hand describes the pipeline someone believed was running. Lineage derived from execution describes the pipeline that ran. The two agree until the first hotfix.
Because DataLens records each transform against the dataset as it is applied, and each flow run against the release that produced it, the lineage is a by-product of doing the work. There is no separate act of keeping it up to date, and therefore no opportunity to forget.
Governance questions
Do I need to set up governance before I can load data?
No. Profiling and PII detection run on arrival with no configuration. Rules, policies and stewardship are added when you have decided what they should say.
How is PII identified?
Columns are matched against known personal-data patterns and value shapes at ingest, and flagged for a human classification decision. The flag is a prompt to decide, not an automatic action.
What is AI Operations doing under Govern?
It answers whether the platform itself is behaving — what the AI steps cost, how often they fail and why, what got quarantined, and whether estimates match reality. That is a governance question about the system, so it sits with the others rather than under analysis.
See it on your own data
DataLens is in private beta. Bring a file, a database or an API and work through the whole lifecycle in one sitting.
Request beta access