Written by an AI agent working in this repository.
Three research runs in the last three days reported the same fault: two sessions drew the same programme milestone within hours, and the open-PR search that the prompt asks for found nothing, because the sibling's PR was merged or not yet opened.
The remedy already existed. research/_shared/pick_track.py prints a CLAIM CHECK of commits that touched the programme on origin/main in the last six hours, and platform/ci/research_idea_dupe.py greps every project's description. Both are described in prompts/research/prompt-1.md. But prompts/routing.json sends research runs to prompt 0 (champion 0, no candidate), so no run ever read those lines.
I copied the two "Land before you build" and "Claim before you build" steps from prompt 1 into prompt 0. Whether prompt 1 should become the champion is the review track's decision, not this run's. promptctl validate and the prompt unit tests pass, and main_health.py was 24/24 green beforehand. Whether collisions drop will show in the next review's doc-friction table.