Two kinds of insight over one data platform. Static insights are curated executive pages that refresh with the nightly load. Dynamic insights are generated on demand from a governed dataset and a written brief, so an executive can ask a new question on Monday and have a challengeable, repeatable answer the same day.
Curated pages in the insights viewer, tied to a dataset and refreshed by the ETL. The page renders the data in the browser, so it is always current and never re-typed.
A row in the insights table: a name, a dataset, a brief. The engine sends the dataset and the brief to the model and stores the page it returns. Run it again next week and compare.
The brief is where the discipline lives. It names the audience, the grain of the dataset, what every column means, which columns are flows and which are positions, the formulas, the thresholds that define a finding, and the order of the page. Because the definitions are written down, the executive can challenge a ranking and the analyst can change one threshold and rerun.
One row per what, over which window, and which part of the window is complete.
Dollars or units, flow or position, code or name. Never leave the model to guess.
Numeric rules for white space, over-space, exceptions, idle capacity. Each finding carries its evidence in one line.
Executive summary, scorecard, tables, findings with actions, data notes stating exclusions.
The shape of a generated page. Figures are illustrative, not from a client.
A dataset for the engine is a governed view sized so the whole result fits one call: a merch plan cube by class and region, a SKU extract for size analysis, DC flow by day, store lanes by day, and a week of sales audit aggregated to store, day, terminal, cashier and exception. Line-grain sources get a week-scoped wrapper rather than a raw dump.