Statement sorter

Sort a month of bank-statement lines into spending categories.

Each transaction is [memo, cents]: a free-text memo and a signed integer amount (negative for money out, positive for money in, such as a refund).

Each rule is [category, priority, includes, excludes], where includes is a non-empty list of words and excludes a (possibly empty) list of words. To compare words, split the memo into maximal runs of letters and digits (every other character is a separator) and lowercase them; rule words are compared in lowercase too, and must match a whole word ("gas" does not match "gasoline", and "7" does not match "711"). A rule matches a transaction if every include word appears in the memo's words and none of the exclude words does.

A transaction takes the category of the matching rule with the highest priority. If several matching rules share that priority, the one that appears earliest in rules wins. If no rule matches, the category is "uncategorized".

Transactions whose winning category is "ignore" (internal transfers) are left out of the report entirely. Other categories are totalled by net amount: refunds are added to the same category as the purchases they match.

Input: transactions, a list of [memo, cents], and rules, a list of [category, priority, includes, excludes].

Output: a list of [category, net_cents, count] for every category with at least one transaction, ordered by net_cents ascending (largest net spending first), then by category name ascending. count is the number of transactions in the category.