When your merge works on one table but fails on the next, using the same column from the same source, it's natural to feel like you're going crazy. But the problem the user encountered here isn't a mystery; it's a classic ETL footgun that reveals how fragile traditional spreadsheet tools become when you push them past their designed limits. The column 'DonorID' was clearly there, visible in the wizard, present in the data, yet Power Query refused to find it during an in-place merge. The fix, merging as a new query instead, works around the bug, but it doesn't solve the underlying brittleness that makes tools like Power Query so frustrating for real-world data work.
This user's story echoes a broader pattern we see across the industry: incremental complexity breaks your pipeline. First, a simple deduplication and merge works fine. Then another merge fails despite identical column names and types. The issue likely stems from how Power Query handles query dependencies and lazy evaluation, the activities table may have been transformed in a way that strips the column at runtime, even if the preview shows it. Compare this to the experience of another user who found that Row insertion no longer breaks your spreadsheet formulas, a small but meaningful step toward predictability. Or consider the developer who built incremental pipelines using dbt to work around How One Developer Built a Smart Pipeline to Outsmart a Tricky Job Market, deliberately designing for the kind of silent data errors that plague tools like this.
What matters here isn't the specific workaround, it's what the workaround tells us about the tools we rely on. When a user has to debug why a column that visibly exists throws an error during a merge, the tool has failed its core promise: to make data manipulation intuitive and reliable. The user did everything right: same data source, consistent column names, no type changes. Yet the system still punished them with an opaque error. This is the consequence of legacy architecture designed for static data, not the dynamic, multi-step transformations that modern ETL demands. The user's frustration is valid, and the solution of creating a new query is a bandage, not a repair.
The takeaway here is direct: if your data tool requires you to know its internal evaluation order to perform a simple merge, it's not empowering you, it's making you an amateur compiler debugger. Look for tools that treat column existence as a guarantee, not a suggestion. The next time you encounter an error like "The column wasn't found" when it clearly exists, ask yourself whether the software is working for you, or whether you're working around its limitations. That question, not the fix itself, is the real opportunity to transform how you handle data.