Contrast pairs
The single most reliable steering technique: show the model a failure and its correction as a pair, same artifact, same subject, one axis of difference. "Here is the hedged ending; here is the committed one. Here is the emoji grid; here is the paragraph." The model moves further on one good pair than on three paragraphs of instructions.
Why pairs beat rules
A rule ("avoid hedging") names a category; a pair defines its boundary. Models generalize from boundaries. The pair also survives translation across models and versions, because it carries its meaning in the artifact rather than in the instruction vocabulary.
Building a pair library
Collect failures from real output, not synthetic strawmen: hedged conclusions and emoji grids both began as folders of pasted transcripts. Write the correction yourself, keep the subject identical, and label the axis ("commitment", "density"). Measure whether the library works with the slop score. Background reading on example-based steering: https://arxiv.org/abs/2005.14165