Governance, privacy and meaning

somebody has to be able to answer what data you hold, where it came from, who may see it and when it gets deleted.

Data governance

you search for 'revenue', find eleven tables with that word in them, and can see which one is actually used by the finance dashboard.

Data catalogue

a new analyst rebuilds a model that already existed because they had no way to find it.

Data discovery

the definition of active user lives in one place that every dashboard, notebook and export queries through, instead of being re-implemented in each.

Semantic layer

two dashboards show different revenue because one includes refunds and neither of them says so anywhere.

Metric definition

the same metric name means something slightly different than it did in March, and the year-on-year chart is comparing two definitions.

Metric drift

there is one table everyone agrees is the answer for orders, and the other four are explicitly labelled as derived or deprecated.

Canonical dataset

the table has an owner, a documented schema, a freshness commitment and a stated audience, like an API rather than a by-product.

Data product

each domain team owns and publishes its own data products rather than throwing rows over the wall to a central data team.

Data mesh

a free-text notes column turned out to contain names and phone numbers, and it had been copied into six downstream tables.

Personally identifiable information

every column is tagged as public, internal or restricted, and the access rules are generated from the tags rather than written per table.

Data classification

the card number is replaced by a meaningless token, and the only system that can turn it back is one you deliberately cannot query from the warehouse.

Tokenisation

the analyst sees the email column as four asterisks and an at sign, while the support tool sees the real thing, from the same table.

Data masking

the user id is replaced with a hash so you can still count sessions per person, and rejoining it to a name is a deliberate act with a key.

Pseudonymisation

you removed the id entirely and aggregated to groups of at least fifty, so no row can be traced back to a person even with the other tables.

Anonymisation

the regional manager runs the same query as everyone else and only ever sees their own region's rows.

Row-level security

everyone can query the customer table and only three people can select the date-of-birth column.

Column-level security

raw events older than thirteen months are deleted automatically, and nobody has to remember to do it.

Retention policy

a person asks to be deleted and you have to find them in the warehouse, the lake, every backup and last quarter's exports.

Right to erasure

the EU customers' rows may not physically leave the EU, so your one global warehouse is no longer allowed.

Data residency

two companies join their customer lists to measure overlap, and neither ever sees a row from the other side.

Data clean room

somebody asks who read the salary table last March and you can answer with a query rather than with a guess.

Audit trail