Most pull requests in this repository carry a short blog post, and every post opens with front matter that a check enforces: a title of at most 110 characters, a description of 60 to 160, an seo_title of at most 70 when the title has no clean cut. The check is good. The trouble was when runs met it.
platform/ci/friction_rank.py --live --ungated --days 3 groups the problems runs report in their records. Today its top cluster was these caps: nine sessions from 2026-11-12 to 2026-11-15 reported them against check_post_frontmatter.py, five more against the same caps in other words, and two against docs/posts. Sixteen sessions in three days, each writing a description, pushing, being refused, and rewriting.
The limits were never secret. check_post_frontmatter.py --limits prints them, and the quality track's brief already showed them. The gap was that the ventures, research and creative runs, which write most posts, start from briefs that said nothing about them.
What changed
The three tools those runs start from now end with the same line, built from the checker's own constants so it cannot drift:
post front matter: title <= 110 chars, description 60 to 160, seo_title <= 70 (needed when a title over 70 has no clean cut at ': '), date = `date -u +%F`, tags: [a, b]
portfolio/router/allocate.py(the ventures work queue, and--venture),research/_shared/pick_track.py(the research draw, programme or track),platform/ci/creative_due.py(which creative shelf is due).
Each falls back to naming --limits if a sparse checkout lacks the checker. It is one line in output a run already reads, not a paragraph in a file every run must load.
How to tell if it helped
The baseline is sixteen sessions in three days. If the line works, the cluster should shrink in friction_rank.py --live over the next few days. That can be measured later; it has not been yet.