Record-to-Truth
A run starts whenever records arrive from somewhere: a form, an import, an enrichment feed, a person typing. The processes here decide what a record must contain before it is usable, deduplicate it against what is already held, enrich what is thin, correct what is wrong, and keep the definitions everything else counts against. This is the stream that decides whether an account means the same thing in two reports and whether a number can be trusted twice. It owns the shape of the record and the rules that keep it honest; what anybody does with a record belongs to whoever is doing it.
vs/rev/record-to-truth v1The processes
The processes in this stream
Dedupe and Merge
Starts withrecords are proposed as one thing
Ends whenone record stands, or the two are marked as different
Activities inside
- Take in the Duplicate Set
- Read the Match Rule That Fired
- Pull What Each Record Holds
17 activities · 46 blanks to fill · 2 human gates
Define a Record
Starts withsomebody proposes a new object, field or rule
Ends whenthe definition is published and its rules are live
Activities inside
- Take in the Definition Request
- Read How the Record Is Used Today
- Write What the Object Means
17 activities · 44 blanks to fill · 1 human gate
Enrich a Record
Starts withrecords fall short of what their definition asks for
Ends whenthe fields are filled and each one names where it came from
Activities inside
- Take in the Thin Records
- Read the Field Definitions
- Check What May Be Filled
16 activities · 47 blanks to fill · 1 human gate
Fix a Data Quality Failure
Starts witha data quality rule crosses its threshold
Ends whenthe source is fixed, the records are corrected, the rule passes
Activities inside
- Take in the Failing Rule
- Check the Failure Is Real
- Measure How Far It Spread
16 activities · 58 blanks to fill · 1 human gate