Quality is configured as packs of rule templates, stamped across a scope to mint the rule instances, with thresholds, schedules, owners, escalation routes and contracts as facets of the pack.
| Object | What it is | Who sets it | Example |
|---|---|---|---|
| Rule template | A rule sentence with parameters left open | Steward | “Lead time must be {min} to {max} days for active {object} in {scope}” |
| Pack | A bundle of templates that belong together, with an accountable function | Steward, approved by the owner | Planning-readiness pack: lead time, minimum order, planner, unit of measure, gross weight |
| Scope | The tables, rows and partitions the pack is stamped across | Owner | Plant materials in 14 plants, status active |
| Thresholds | Per dimension and per field: the score below which a record is held, and the weights | Owner with the consumer | Completeness of planning attributes ≥ 98% before a record is golden |
| Schedule | When the instances run: interval, calendar, event, on request | Steward | Nightly 02:00; on change for the weighted fields |
| Escalation route | The screen and the people a finding goes to, with a hop SLA | Owner | Planning centre review screen, 24 hours |
| Contract | The thresholds a consuming system relies on, and what happens on a breach | Owner with the consumer | Planning reads materials with completeness ≥ 98%; breach → notify and hold the feed |
Stamping a pack across a scope mints one rule instance per combination: three templates across fourteen plants give forty-two agents, each with its own findings, acceptance rate and eval. The instances inherit the pack's thresholds, schedule, owner and route, so a change to the pack flows to all of them at the next publish.
A plant that needs an exception edits its instance: a longer lead-time ceiling for a plant that imports by sea, say. The edit is tracked as drift. At the next publish of the pack, the drift is shown beside the pack's version, and the owner chooses per instance: keep the local edit, or adopt the pack's new template. The choice is recorded as an override contract with the reason.
Nothing is overwritten silently, and nothing is deleted: a retired instance is flagged off with its dependencies shown first, and can be restored with its history.
| instance | plant | state | local edit | decision |
|---|---|---|---|---|
| lead-time-range | 1010 | drift | max 180 → 240 (sea freight) | keep · override contract |
| lead-time-range | 1020 | in sync | — | adopt v7 |
| moq-positive | 1010 | in sync | — | adopt v7 |
| planner-assigned | 1030 | drift | escalation → plant controller | keep · override contract |
| uom-valid | all | in sync | — | adopt v7 |
Configuration follows the same path as any other change on the platform.
Domain scores on the showcase estate, weighted. Bank details are held back by timeliness: 312 records past the twelve-month verification. Showcase estate.
The domain page shows the score per dimension, the packs and their versions, the drifted instances, the agents by level, the open contracts and the trend since the programme started. Each number opens the register behind it, so 'why is bank details at 79%' is one click to the 312 records and the timeliness rule that holds them.
The same page carries the excusals, the budgets consumed today, the agents demoted by policy this month and the evaluations due, which is what a governance lead needs for the monthly review and what an auditor asks for.
Yes. Packs, templates, thresholds, routes and contracts are definitions with a version and an append-only history, and a change set is the unit that moves between environments.
An accountable function, named on the pack: planning for the planning-readiness pack, treasury for the bank-details pack. The owner stamps publishes and promotions and answers for the drift decisions.
From the profiler's numbers and the consumer's needs: a contract states what a consuming system relies on, and the threshold is set at or above it. Thresholds are facets with a history, so they can be tightened as scores rise.
Templates are sentences, scope is a selection, thresholds are numbers and routes are screens; the intended author is the steward with the owner's stamp. The technical gate is the write-back test per field, which the application owner runs.
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.
