test(planning): run the seven interview eval-first cases and file a rule slice for each case that fails #9794

needs-human: a paid claude plugin eval run (pre-approved within prepaid credit, Q46); no unattended lane runs plugin evals.

Parent: #7269.

Run

Run each case on a copy of plugins/planning outside git (the repository has too many linked worktrees for the CLI to screen), one or two at a time:

claude plugin eval <copy-of-plugins/planning> --case <name> --runs 3 --judge-model opus --keep-temp

Cases: the seven from #7247 (read-back of a dictated answer, active-reply final confirmation, changes-only final brief, recommendation basis and flip, "something else" closing probe, MoSCoW scope cut before wayfind, crux question on a rejected recommendation). Size the cost ceiling from the last partial pass's costUsd. Read each result with run-validity.py and noise-report.py; for final-confirm-active-reply and final-brief-changes-only read the kept traces.

Done when

• Each of the seven cases has a VALID with-plugin result posted on test(planning): run the seven interview eval-first cases and land the rules whose case fails #7269.
• For each case that fails on main, a lane-ready slice is filed under test(planning): run the seven interview eval-first cases and land the rules whose case fails #7269 to add its rule to the interview context spokes, citing the result. planning: interview web view, deferred review findings #7303's eval-first page items depend on these results.