Change taxonomy

The nine categories every change is filed under. The list is public on purpose: a score you cannot inspect is a score you cannot argue with.

The nine categories

01Stock openers18%
Scene-setting first clauses that carry no information.
02Hedges15%
Qualifiers aimed at the reader rather than the subject.
03Filler phrases14%
Preambles that announce a point instead of making it.
04Vague claims12%
Abstract benefits with no number, name or date in them.
05Clichés11%
Stock metaphors standing in for a plain verb.
06Padding10%
Several words doing one word of work.
07Repeated shapes8%
Three sentences in a row built the same way.
08Listicle reflex7%
A list where a sentence would do.
09Over-signposting5%
Connectives the reader does not need to follow the point.

Phrase and structural

Seven categories are found by matching a published phrase list, so the fix is deterministic and needs no model. Two — repeated shapes and listicle reflex — are visible only in the shape of the text. Those are reported with after: null: the API names the problem and leaves the rewrite to you.

Shares

The share next to each category is the proportion of changes it accounts for across polished drafts. It is published so that a surprising distribution in your own writing is something you can notice.