"Ordering is an ordering, never a ranking of importance."
A sorted list is not innocent
The dashboard has a "what moved" panel: each series' newest reading beside the one before it, with the percentage change, sorted by the absolute size of that change. It is a genuinely useful thing to look at, and it is also the single most dangerous widget in the product.
The danger is not in the arithmetic — a percentage change is a fact. It is in what a human does with a vertical list. Readers do not experience a sorted list as "these are ordered by a quantity I could name." They experience it as a leaderboard. The top item feels like the most important item, and the feeling arrives before any conscious reasoning about the sort key.
So a surface that would never say "this is the most important change today" can say exactly that by accident, through layout alone, and be entirely truthful in every string it renders.
Two defenses, one soft and one hard
The soft defense is presentation. Show the sort key. Show both dates. Show the previous value beside the new one. Every extra visible fact competes with the leaderboard reading, because it makes the list look like a table of observations rather than a chart of winners.
The hard defense is in the code, and it is the one worth stealing: the function's own docstring states what the ordering is and is not. Descriptive by construction: a sorted list of magnitudes, no thresholds, no "alerts", nothing engineered to provoke. Ordering is an ordering, never a ranking of importance.
That sentence is not documentation. It is a constraint aimed at a future contributor who opens the file with a reasonable-sounding ticket — "highlight the top mover", "add a badge to the biggest change" — and would otherwise have no idea that the innocuous-looking sort is load-bearing.
The absent features are the design
Look at what the digest does not have. No threshold that promotes an entry. No color that escalates past a certain magnitude. No notification. No "unusual" flag. Each of those would be defensible on its own and each would convert an observation panel into a signal panel.
It also caps the list and reports how many pairs were compared. That is a small honesty: a list of twelve out of ninety-seven is a sample, and a sample presented without its denominator quietly implies it is the whole.