On this page
In short. An organization brings its own numbers, declares what each column means, and then reads them like any other measure: on a dashboard, in the morning brief and in the data assistant. A stock summed across months is refused when the formula is saved, not computed wrong.
How a column becomes a measure#
A dataset is a table the organization declares: a source (a spreadsheet tab or an internal system), its columns, and a role per column. The platform reads a sample and suggests a role; an administrator confirms before the column is used in a calculation.
| Role | Means | Examples |
|---|---|---|
| Dimension | Group and filter by it | department, warehouse, cost center |
| Measure | Count it | headcount, overtime hours, inventory value |
| Time | Decides which period a row joins | month, snapshot date |
A measure then declares four things:
- Time behavior. A flow accumulates across periods. A stock does not; summing it is refused when the formula is saved.
- Lowest valid grain. A monthly measure cannot be charted daily.
- Declared filters. The measure names the dimensions it honors. Any other filter shows as ignored, never silently dropped.
- No-data reason. A measure that cannot compute says why (no data yet, sample too small, wrong grain, source not connected, missing input, no rule declared, not permitted). It never returns zero.
Derived measures (ratios, differences, period averages, period-end values, sums across columns, threshold classes) come from a closed set of named operations. There is no free-text formula field. A formula is validated at save and precomputed once data arrives, so two pages never disagree.
A card points at a measure by name: pick a dataset, a measure, a breakdown and a grain, then add a filter. The same card catalogue serves commerce and dataset measures, so one page can mix both (Dashboards).
Industry data packs#
A pack ships a domain's standard as versioned data: the datasets, each column's role, the derived formulas, the dashboard pages, the alert rules and the morning-brief entries. Every raw record set comes with a starter mapping template.
| Pack | Datasets | Dashboard pages | Morning brief |
|---|---|---|---|
| HR people analytics | 7 raw record sets, 5 monthly rollups | Overview, Recruitment, Turnover, Overtime, Staff cost, ROI per employee, Promotion and salary, Company-wide | headcount, turnover, staff cost variance, hires, departures |
| Finance P&L | 3 raw record sets, 2 monthly rollups | P&L overview, Budget vs actual, Cash flow, Cost centers | none |
| Warehouse ops | 3 raw record sets, 2 monthly rollups | Warehouse overview, Stock, Fulfillment, Movement and cost | stock on hand, on-time rate, backorder rate |
A pack's alert rules do nothing until an admin switches them on (Configure the morning brief and alerts).
Upgrading a pack never overwrites a card the organization edited. A pack-installed card carries the pack's mark and a content snapshot taken at install; the upgrade compares the card with that snapshot. An edited card is skipped, an untouched card is updated, and card position is never compared.
Where it stops#
- An empty-looking cell always carries a reason and is never zero by default.
- A card with no install snapshot counts as untouched, so the next upgrade overwrites it. Local edits survive only on a card that has its snapshot before the first upgrade.
- A salary or cost-per-head figure is not shown to a role without permission.
- Rollout state per pack is on Project status.
Reference#
Adding a measure without engineering
| Task | Who | New build |
|---|---|---|
| Attach another tab or source to an existing dataset | organization admin | no |
| Confirm a new column as a dimension, measure or time column | organization admin | no |
| Put a measure on a new card; change breakdown, grain or filter | organization admin | no |
| Define a new derived formula (ratio, difference, period average) | platform team, through a data pack | yes |
| Add a new industry data pack or internal connector type | platform team | yes |
What it reads and never reads
| Reads | Never reads |
|---|---|
| The organization's own tabs and system exports, per brand | Another organization's dataset, even under the same pack |
| A row with no brand, shown under every brand | A salary or cost-per-head figure for a role without permission |
| The install-time snapshot of a pack card | Card position, which is never compared on upgrade |