A model shaped for writing is normalised and a model shaped for reading is not, and insisting on one model means one of the two is wrong.
-- write model: eleven joins to answer the dashboard
-- read model: one table, shaped like the question
CREATE TABLE order_summary (
order_id BIGINT PRIMARY KEY,
customer_name VARCHAR(255), -- duplicated. deliberately.
line_count INT,
total_cents BIGINT,
refreshed_at DATETIME(6)
);
-- and the staleness is a number somebody agreed to.
The duplication is the point and is only defensible when the read model is derived rather than authoritative — it must be rebuildable from the write model at any time, and rebuilding it should be a routine command rather than an emergency procedure. The staleness budget has to be agreed with the people reading it rather than assumed: four seconds is fine for a dashboard and not for the page somebody lands on immediately after saving.