How do you decide which data sources are trustworthy enough to use?
Trust in a data source is assessed, not assumed. The practical question is whether you would be willing to defend a number built on it in front of the person whose budget depends on it.
Is it the system of record? — prefer the system where the transaction actually happens over a downstream copy, export or spreadsheet, since every hop adds a chance for logic to drift.
Completeness and coverage — check row counts over time, null rates on critical fields, and whether the date range and entity coverage actually span the question being asked.
Freshness and reliability — establish the refresh cadence, typical lag and failure history; a source that silently stops updating is more dangerous than one that is obviously missing.
Does it reconcile? — the strongest single test is tying a total back to a figure the business already publishes, such as finance revenue, and explaining any residual gap.
Ownership and stability — a named owner, documented schema and change notifications mean problems get fixed; an unowned table nobody will vouch for is a liability regardless of how clean it looks.
Key terms: system of record, data lineage, completeness, freshness, reconciliation, schema stability, data ownership, certified dataset
Frame the answer around defensibility — "would I stake a budget decision on this number" — because that reframes trust as a practical standard rather than a checklist. Reconciliation to an already-trusted figure is the point interviewers most want to hear, so make it the centrepiece and give an example of tying a warehouse total back to finance-reported revenue.
The stale-versus-missing distinction is a strong differentiator: a pipeline that stops silently leaves data that looks valid, which is worse than an obvious gap. Expect the follow-up "what if the only available source is a spreadsheet someone maintains?" — the answer is to use it while documenting the risk, reconcile what you can, and push for a proper source rather than refusing outright.