When you have 0 data but can sketch out what every figure should look like, that is when you can author a manuscript confidently. For me getting to this level took high-level authorship contributions to 8 published manuscripts.
This is what Whitesides means when he says to do experiments according to an outline -- the earliest stage in the project when you know it "works" is when you make some sort of outline or figure list describing in complete detail what data are needed to finish the paper. Then you actually write the paper while you collect your data. Before people reach this level, what they end up doing is just working really hard at doing "every possible" experiment and only publishing the ones that fit into the PI's desired narrative. They only start writing once 100% of the data is collected. This way is less productive than Whitesides way. If you have a data outline you can start writing with only 0-25% of the data already collected. That way the timelines for experiments and writing actually coincide.