[mohitj1988]

ppg:staging:18 and 3 of its tarballs. None of the container projects ran (ppg:staging:18:containers, :extras:containers, and the PG 18 entries of common:tools:containers, common:containers, common:extras:containers).

The nightly listed QA projects from one OBS server (vars.OBS_APIURL) and used that server for every job, but one server does not host every project: container projects live on another instance. obs-qa-run.yml already resolves each project's instance from OBS_INSTANCES; the nightly now does the same.

• list_qa_matrix.py: with OBS_INSTANCES set (the nightly), take the projects from this checkout's qa: blocks, resolve each to its instance with resolve_qa_instance.resolve_instance, and carry the instance in every combo (name, apiurl, user, registry, rootprj, env). Release snapshots are skipped as before; a project no instance hosts is reported, not silently dropped. Without OBS_INSTANCES (PR check) discovery is unchanged.
• obs-nightly-qa.yml: pass OBS_INSTANCES to discovery; each qa job connects to its combo's instance (OBS_PASSWORD_) and writes that instance's CI profile exactly as obs-qa-run.yml does.
• belongs_to_major: a major's own project (ppg:staging:18...) belongs to that major only; the -pg / -to- name rules apply to version-independent projects only, so no build can run on two nights.

Dry run for PG 18 with a packages instance and a containers instance: 45 combos, 25 on the first (as observed), 20 on the second (previously missing).