If a formula that once worked suddenly breaks, most users assume they made a typo. In this case, the user did everything right, they used `=SUM(--TEXTSPLIT(Z3:Z48,","))` to split comma-separated values and coerce numbers out of text. The problem isn't their skill. It's that traditional spreadsheet tools were never designed to handle messy, human-entered data like `90, A, Y` reliably. That formula fails because `TEXTSPLIT` returns an array that includes text values, and the double-unary operator (`--`) throws an error when it tries to coerce `"A"` into a number. The user isn't struggling with a limitation of their own making. They're hitting a wall that legacy spreadsheet architecture built decades ago.
What this user needs isn't a more complicated workaround, it's a fundamentally different approach to data extraction. A modern, AI-native spreadsheet should be able to look at a cell containing `90, A, Y` and simply understand that `90` is a number worth summing while `A`, `G`, `Y`, and `R` are labels to ignore. This isn't a parsing problem. It's a pattern-recognition problem. When every cell has inconsistent text strings, no amount of manual `SUBSTITUTE` or `FILTER` gymnastics will scale. The solution is a function that can isolate numeric values from mixed content automatically, without requiring the user to predefine every possible text variant they might encounter.
This matters because the user's goal is straightforward: total minutes played across a football stats sheet. That's a common, practical need, tracking time, counting occurrences, aggregating scores. The friction they're experiencing isn't inherent to the task. It's imposed by a tool that forces them to fight the data's structure instead of working with its meaning. An AI-native spreadsheet should accept data as it is and let the user ask for what they want in plain terms: "Sum only the numbers in this range, ignoring everything else." That request should work on the first try, not require a forum post and a half-dozen comment threads.
Our position is clear: when a user has to abandon a working formula and search for alternatives because their spreadsheet can't distinguish `90` from `A`, the tool has failed them. The fix isn't a better regex or a clever array trick. It's a spreadsheet that treats numbers as numbers and text as text, even when they share a cell. That capability exists now. The only question is whether users will keep patching old tools or finally adopt one that meets them where their data actually lives.