Standing up quality on one domain takes a working week: profile the sources, write the rules for the costliest fields, stamp them across the scope, review the first corrections and enable the write-back, then turn on the publish gate and sign the first contract.
Times are from the showcase supplier domain (two sources, 41,800 records); an estate with more sources adds profiling time, little else.
| Day | Work | Output | Test |
|---|---|---|---|
| Monday · profile | Connect the sources; run the profiler; read the column scores and the cross-source comparison | A scorecard per source and the cost view | The owner names the ten costliest fields from the cost view alone |
| Tuesday · rules | Write the directives for the ten fields as sentences; dry-run each on live rows; set every agent to L1 | Ten rule agents with their dry-run counts | Each dry run flags a number the steward recognises as right |
| Wednesday · packs and scope | Group the rules into packs; stamp them across plants and regions; set thresholds, schedules, owners and routes | The pack register with instances and inherited facets | The first scheduled run at 02:00 Thursday produces findings for every instance |
| Thursday · correction | Review the first proposals in the queue; set L2 for the two standardisations with the best acceptance; enable the write-back per field with a test write | The first corrections applied and acknowledged | A test write per field per system is acknowledged and visible in the source |
| Friday · enforcement | Turn on the publish gate; read the held records; sign the first contract with the main consumer; send the scorecard to the owner | Golden, in-flow and quarantined counts; one contract live | The consumer's feed reads golden records only, with the state visible |
The showcase supplier domain on day thirty, weighted overall 91% from 74% on Monday. Timeliness is the remaining work: the aged bank verifications. Showcase estate.
| Measure | Day 1 | Day 30 |
|---|---|---|
| Weighted score | 74% | 91% |
| Fields under rules | 0 | 38 |
| Rule agents | 0 | 27 (4 at L3, 9 at L2, 14 at L1) |
| Corrections applied | 0 | 12,400 · 96% accepted |
| Records held or quarantined | 312 on day 5 | 41 |
| Contracts | 0 | 3 of 3 met |
| Alerts a week | 41 (fixed thresholds) | 6 (learned patterns) |
The second domain skips most of the learning: the team knows the sentence form, the dry run, the pack register and the gate. Its week is shorter, and its packs often start from the first domain's templates: the country, address and tax-id rules of the supplier domain are the customer domain's rules with a different scope. On the showcase estate the customer domain reached its publish gate in three days.
The order of domains follows the cost view across the estate, which the profiler produces once every source is connected. The domains page gives the six groups and the effort per domain, and the catalogue carries sample rules per dimension for each of the thirty.
One steward full time, the domain owner for two decisions a day, the application owner for the write-back tests on Thursday, and the main consumer for the contract on Friday.
Monday's profiling takes longer, the cross-source comparison has more columns, and the write-back tests on Thursday are per system. The rest of the week is the same.
Yes, and many programmes do for the first month: the gate runs in observe mode, showing what it would hold, so the owner can see the work list before consumers feel it.
The monthly review on the domain page: scores and trend, drifted instances, agents due for promotion or demotion, contracts and their breaches, and the next domain.
Bring one domain and two systems. In a week the sources are profiled, the ten costliest fields are under rules, the first corrections are written back with their acknowledgements and the publish gate is on.
