Compare with previous run

Compares the data of the current run with the previous run of this same step and fails the test if it changed: monitoring a price, stock level, catalogue contents or the version in the footer.
Four fields: selector|key|tolerance|threshold. An empty selector takes the dataset of the nearest preceding “Extract data” step. The key is the field that identifies a row across runs. The tolerance is the percentage of changed rows below which the step stays quiet.
The threshold (“fail if less than 60000”) works instead of the comparison and only for a single number; tolerance and threshold disable each other.
On the first run there is no baseline yet - the step passes and establishes one. It spends no AI tokens. The fields and examples are covered in the action help.

Compare the data of the current run with the previous run of this same step. If it changed, the test fails (with the usual alarm and notification). This is how you monitor a price, stock level, catalogue contents or the version in the footer. Four fields: selector|key|tolerance|threshold. The step spends no AI tokens or money.

A single value - price on a product card:

• .product-price|| - fails as soon as the price differs from the previous run
• .stock-badge|| - watch availability
• footer .version|| - catch an unplanned deploy

A table - right after an “Extract data” step:

• |Link| - the selector is empty (take the dataset of the previous step), rows are identified by the Link field: the step reports what was added, removed and whose price changed
• |SKU|5 - the same, but while fewer than 5% of rows changed the step stays quiet (a catalogue always churns a little)
• || - no key: rows are compared as a whole, so only additions and removals are visible

A threshold instead of a comparison - minimum advertised price (MAP):

• .product-price|||<60000 - fails if the price drops below 60000; any change above the threshold is ignored
• .stock-count|||<1 - the item ran out
• .rating|||<4.5 - the rating slipped
• .discount|||>30 - a discount larger than allowed

Tolerance and threshold are mutually exclusive - in the form they disable each other. With a tolerance filled in the step works the old way (tables and lists included). If both end up filled anyway (after editing the step by hand, or after an import), the tolerance wins.

With a threshold filled in, the baseline of the previous run is not used: the step only checks the current value. The amount is read out of the text automatically - 63,000 USD/mo means 63000. If a table is being watched, or the value holds no digits, the step passes and states in the report that no check was made.

On the first run or after a scenario change there is nothing to compare against - the step passes and establishes the baseline. That is expected: monitoring starts working from the second run on.

A typical chain: “Go to page” → “Extract data” (catalogue cards) → “Compare with previous run” (|Link|3) → run the test hourly on a schedule.