This is a frustrating puzzle, and we think the answer is hiding in plain sight. You've done the right troubleshooting: permissions match, the source file is intact, and other workbooks in the same folder refresh fine. The fact that one team can refresh this specific file while another cannot points squarely at a metadata mismatch, not a broken connection. The error messages about a missing source or columns are symptoms, not the root cause.
What likely happened is that the file's query metadata became bound to a specific user's authentication context or local environment during an earlier edit. Power Query stores connection details in a way that can incorporate machine-specific paths or cached credentials. When the affected team opens the file, their local environment doesn't match that embedded reference. This is why recreating the file, your last-ditch plan, often works: a fresh file builds metadata from scratch, using the current user's context. The problem isn't the data or the permissions. It's the invisible fingerprints left behind by whoever last saved the queries.
For your team, the practical fix is to rebuild the queries in a clean file, but with one deliberate step: have someone from the *affected* team perform the initial setup. Open a new workbook, connect to the source, define the queries, and save it to the shared folder. That forces the metadata to align with their environment. If that person can then refresh without errors, share the file. Every user who opens it afterward should also test a refresh immediately. If the error reappears, the issue is likely tied to a specific user's cached credentials in Power Query's local settings, which can be cleared via the Data Source Settings dialog.
Do not treat this as a one-off glitch. It is a warning about how shared workflows interact with Power Query's metadata persistence. Moving forward, establish a rule: designate one user to own the query definitions and rebuild them after any major structural change to the source data. If your organization uses a central data gateway or a shared data model, migrate those connections there. That removes the dependency on individual environments entirely. You have a working solution in your hands, recreate the file with the affected team in the driver's seat. That is the concrete step that will turn this conflict into a closed case.