The vocabulary table covers the studio's lists, and stays ahead of the next page #152

Pushed.

test/studio-words.test.ts's table now covers #95's three studio lists. The table moved out of the first test into the end of beforeAll, so both tests walk the same rows, and each row carries its own proof: shows/hides are literal HTML the page must and must not have (a page that rendered nothing satisfies every vocabulary assertion, so a ?word= or cursor row without them proves nothing), and dev names the words dev's own face still says at the same path with ?mode=dev — which makes the second test table-driven rather than one hand-written case, and also asserts dev never renders the studio's word badge. Nine rows added: /issues, /pulls, /runs, each bare, filtered, and at the cursor page a reader's "next" link hands them; the one beforeAll seeds what they need (a tell nobody picked up, a second open change with no check, a run that stopped, a run still going), so a new page still costs a path and a word and there is no second seeding helper.

Walking /pulls for the first time caught a real leak, which is the second file: the studio's change list ended on "branch names and the diff, dev's own way" — the \bbranch\b rule the table has always forbidden. It now says "where each change lives and its diff". That fix is its own commit, first, so each commit is green on its own.

Worth looking at: every new row was verified against a deliberately broken page — with the word filter ignored, the cursor ignored, or ?mode=dev no longer escaping the studio's render, the matching row fails and names itself. That check is how the run-list cursor row got its "a run still going" fixture: its first shape passed on the word filter's work rather than on paging.

bun x tsc --noEmit passes and the forge.json check suite is 289 pass / 6 fail here, the same 6 on an unmodified c72ba52: pulls, cross-repo-pull, web, apps and apps-adversarial cases that push over git-http and get auth required in this sandbox. Unrelated to this change, and they pass in the forge's own runner — #147's check went green in 204s; what failed there was its preview deploy, which also failed for #148 from the same base.

🤖 Generated with Claude Code

Closes #97