Fix a Fact Table With Mixed Grain

You inherit a warehouse table called fact_orders from a previous engineer. Its DDL:

CREATE TABLE fact_orders (
    order_key      INT PRIMARY KEY,
    order_id       INT NOT NULL,
    customer_key   INT,
    product_key    INT,       -- NULL for order-level rows
    order_date_key INT,
    amount         NUMERIC,
    row_type       TEXT       -- 'order_total' or 'line_item'
);

Looking at the data, you find that for a single order with 3 line items, there are 4 rows in this table: one with row_type = 'order_total' (amount = the full order total) and three with row_type = 'line_item' (amount = each line's amount, product_key populated). An analyst ran SELECT SUM(amount) FROM fact_orders WHERE order_date_key = 20260301 to get "total revenue for March 1st" and got a number roughly double the real total.

1. Explain exactly why the analyst's query produced a number too high, in terms of grain.
2. Redesign this into a correct schema — decide whether you need one fact table or two, and write the DDL.
3. What would you tell the analyst to change about how they write the revenue query against your redesigned schema, and how would you prevent the next analyst from making the same mistake the previous design invited?