Back the unglamorous bet when it is the real constraint. At MLSE reliability was the growth thesis, because the interesting work only gets funded once the boring work holds.

Fund the experiment properly or do not run it. Protected time is what turns curiosity into evidence; a rushed prototype only ever proves that it was rushed.

Remove the decision instead of explaining it. Fans wanted to collect, not learn a category, so the work was cutting steps rather than writing better copy.

Say what the result will mean before it arrives. Every proposal named what strong, weak and inconclusive would look like, so nobody argued the scoreboard afterwards.