No Support Beyond Reach: The Measure Centrality Index 2.0 and Adversarial Comparisons of Consciousness Theories

arXiv:2611.06925 · q-bio.NC · Submitted 2026-09-15 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: We're looking at the paper: "No Support Beyond Reach".

Marcus: Here's the gist of the paper. A success in adversarial testing of scientific theories reaches only as deep as its matching failure would have cut, and this paper introduces the Measure Centrality Index 2.0 to enforce this constraint The Measure Centrality Index 2.0 extends the original version to the positive-evidence side, recording for every possible outcome what it would count against and what its proponent has agreed to change The One-Way Evidential Ratchet Adversarial tests of theories of consciousness (ToC) can acquire a characteristic asymmetry where a preregistered prediction passing may be claimed as evidence for the whole research program, while failure is absorbed by auxiliary apparatuses This asymmetry is termed the one-way evidential ratchet, which operates whenever the attribution of an outcome to a theoretical target is settled after the outcome is known.

Ines: The first experiment in the adversarial collaboration between GNWT and IIT serves as an example of this attribution risk, where theory proponents could anticipate revising implementations while retaining cores.

Marcus: What the Measure Centrality Index is for The MCI 2.0 specifies how far a success reaches, holding it to the same level as how deep a failure cuts.

Ines: It builds machinery that keeps this record, specifying what is compared, under which assumptions, when data suffice to settle an outcome, and which targets of a dated theory version each outcome exposes.

Title and authors: Marcus: So we're looking at this paper now titled "No Support Beyond Reach: The Measure Centrality Index two point zero and Adversarial Comparisons of Consciousness Theories <ref:2611.06925#pg1,No Support Beyond Reach: The Measure Centrality Index 2.0 and Adversarial>." It sounds like they’re building a way to make sure when you test a theory of consciousness, the support you get is actually tied to how deep the failure would have cut.

Ines: Exactly. It’s about fixing this problem where a passing prediction gets treated as proof for the whole research program, while if it fails, they just blame some tool or measurement setup. This paper introduces something called the Measure Centrality Index two point zero to enforce that constraint on the positive evidence side too <ref:2611.06925#pg1,the Measure Centrality Index 2.0>.

Yuki: It sounds like they are trying to create a fairer way to measure support for different theories of consciousness, making sure you don't claim more support than what's actually there based on how sensitive the test is.

Marcus: Right. Before this, we had this one-way evidential ratchet where you could get credit for the whole program just because a prediction passed, even if it failed to rule out some parts of it. This new index is designed to stop that by recording what each outcome would count against and what the proponents have already agreed to change.

Ines: It’s about making the test object itself record all that stuff before any data comes in, which is a big deal for controlling how we interpret experimental results in this area.

The paper's summary: Ines: Basically, the Measure Centrality Index two point zero extends the original version to look at success instead of just failure, and it specifies exactly how far that success reaches by holding it to the same level as where a failure would have cut <ref:2611.06925#pg1,the Measure Centrality Index 2.0 extends the original version to>. They build this machinery to track what is being compared, under what assumptions, and which parts of an older theory get exposed by a certain outcome.

Marcus: What's interesting about the summary is that they’re not just looking at one thing; they are building something that records the two things: what each possible outcome would count against, and what the proponent has already agreed to change based on that test. That’s a lot of bookkeeping before you even get any data.

Yuki: I think this means they are trying to define a concrete boundary for where support can be claimed in public debate about consciousness theories, tying it directly to the depth of potential failure.

Ines: That's right. The core idea is the Reach Principle, which says support for a theory can only be claimed as deep as the matching failure would have cut under the same conditions. They’re using this index to record that comparison, which includes things like what is compared and what targets of a dated theory version each outcome exposes, in one single document before we look at the actual results.

Marcus: And they use this to stop that ratchet because by recording all those options side-by-side before the data arrives, they ensure that the success travels exactly as far as the matching failure would have along a path that theory itself agreed to. It’s about timing attribution before you know what happened.

The paper's improvements: Ines: The authors suggest several ways to improve this framework, focusing on how it handles objections and neighbors from other theories of consciousness. They specifically address the Bayesian modeling objection where a theory's core might make a successful implementation look more expected, while failure updating lands on a fragile auxiliary apparatus.

Marcus: And they also tackle the second objection that preregistered exposure limits room for theories to grow by proposing a successor rule. This rule says that when you revise a theory or test with new data, it becomes a new object and only inherits support through what stayed unchanged after testing.

Yuki: That successor rule sounds like an important mechanism for tracking evolution; it means if you change something, the old support doesn't just vanish; it has to pass on through the parts that weren't modified. That’s crucial for understanding how concepts evolve in this field.

Ines: And they suggest incorporating a path pi into the test object. This records exactly which links a given measure depends on, which helps you differentiate between different versions of the theory based on their specific dependencies, addressing issues where structural centrality can’t settle what a result may be taken to show.

Marcus: It sounds like they are trying to make the record much more detailed about the evidence sensitivity itself. They want to prevent situations where we can't settle what a result might actually be pointing toward because of how deeply it depends on certain parts of the measurement setup.

Conclusion: Ines: So, to wrap up, this paper with its Measure Centrality Index two point zero is about enforcing the Reach Principle by recording what each possible outcome would count against and what the proponent has agreed to change before any data is seen <ref:2611.06925#pg1,Measure Centrality Index 2.0>. It’s designed to stop that one-way evidential ratchet by tying success directly to the depth of potential failure, making sure support travels only along paths the theory itself agreed to.

Marcus: Right. The improvements they suggest, like having a successor rule for revisions and recording exactly which links measures depend on via path pi, help make this record much more robust against objections from things like Bayesian modeling or concerns about theories stagnating when you test them repeatedly.

Yuki: I think the main implication for us is that we need this kind of pre-registration to be extremely detailed about the entire comparison setup, including all those negative targets from which each deepest exposed target follows, so we can actually see where the support ends.

Ines: Exactly. The whole point is to make sure that when a prediction passes, it doesn't automatically mean you’ve proven the whole research program; it has to be demonstrably as deep as what a matching failure would have cut in the first place. That’s what "No Support Beyond Reach: The Measure Centrality Index two point zero and Adversarial Comparisons of Consciousness Theories" is all about <ref:2611.06925#pg1,No Support Beyond Reach: The Measure Centrality Index 2.0 and Adversarial>.

Sussex Centre for Consciousness Science · University of Sussex

q-bio.NC

Submitted: 2026-09-15

Updated: 2026-09-15

Comments: 23 pages

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 80/100

The gist: A success in adversarial testing of scientific theories reaches only as deep as its matching failure would have cut…

Key concepts

Measure Centrality Index 2.0 (MCI 2.0)
A tool that extends original indices to the positive-evidence side. It records for every possible outcome what it would count against and what the proponent has agreed to change, ensuring success depth matches failure depth.
One-Way Evidential Ratchet
An asymmetry in adversarial testing where a preregistered prediction can be claimed as evidence for the whole program, while failures are absorbed by auxiliary tools. This happens when attribution is settled after the outcome is known.
Reach Principle
The rule stating that support for a theory may only be claimed to the depth where a matching failure would have cut under the same assumptions. It constrains where public support can be asserted.
Success/Failure Depth Matching
The core constraint: a success counts for a target only if the matching failure would have counted against that same target. This is enforced by recording test objects that define the comparison and depth order.

Terminology

Summary

A success in adversarial testing of scientific theories reaches only as deep as its matching failure would have cut, and this paper introduces the Measure Centrality Index 2. p. 0

Where the paper says it

No Support Beyond Reach: The Measure Centrality Index 2. 0 and Adversarial Comparisons of Consciousness Theories Robert Chis-Ciure, PhD Sussex Centre for Consciousness Science Department of Informatics, University of Sussex, United Kingdom Abstract A success in adversarial testing of scientific theories reaches only as deep as its matching failure would have cut.

0 to enforce this constraint The Measure Centrality Index 2. p. 3
Where the paper says it

Section 2 builds the working parts, namely, the measure-role that bears an index value, the test object that goes on record, and the profile listing, for each possible outcome, what it would count against and what its proponent has agreed to change. Section 3 gives the argument behind the Reach Principle, first as a lemma inside standard confirmation theory, showing that a success whose matching failure was sent to an auxiliary confirms nothing deeper than that auxiliary, then as a reciprocity requirement carrying the same constraint into public attribution.

0 extends the original version to the positive-evidence side, recording for every possible outcome what it would count against and what its proponent has agreed to change

The One-Way Evidential Ratchet

Adversarial tests of theories of consciousness (ToC) can acquire a characteristic asymmetry where a preregistered prediction passing may be claimed as evidence for the whole research program, while failure is absorbed by auxiliary apparatuses This asymmetry is termed the one-way evidential ratchet, which operates whenever the attribution of an outcome to a theoretical target is settled after the outcome is known. The first experiment in the adversarial collaboration between GNWT and IIT serves as an example of this attribution risk, where theory proponents could anticipate revising implementations while retaining cores. p. 1

Where the paper says it

At that point, the interpreter can choose where the result lands, and the incentive is to send success up to the program and failure down to auxiliaries. The first experiment in the adversarial collaboration between GNWT and IIT is a good example of this attribution risk.

What the Measure Centrality Index is for

The MCI 2.0 specifies how far a success reaches, holding it to the same level as how deep a failure cuts. p. 2

Where the paper says it

0, extends the index to the positive-evidence side. Where first version settled how deep a failure cuts, the new one specifies how far a success reaches, and it holds the two to the same level.

It builds machinery that keeps this record, specifying what is compared, under which assumptions, when data suffice to settle an outcome, and which targets of a dated theory version each outcome exposes. p. 0
Where the paper says it

The ratchet turns wherever attribution is settled after the outcome is known. Against it, I define a preregisterable test object, putting the attribution policy on the record before the result, comprising what is compared, under which assumptions, when data suffice to settle an outcome, and which targets of a dated theory version each outcome exposes.

The index assigns a measure-role (µ) to empirical measures by the theory-relative consequences of disconfirming measure-based predictions. p. 2
Where the paper says it

3, 6) proposed the Measure Centrality Index (MCI) to address problems of this kind. The index ranks empirical measures by the theory-relative consequences of disconfirming measure-based predictions, on a gradient from zero change to replacement of the framework.

Where the Gains Come From

The Reach Principle states that support for a ToC may be claimed only as deep as the matching failure would have cut, constraining where support can be claimed in public. The principle requires that a success counts for a target only if the matching failure would have counted against that same target under the same contrast and background assumptions. p. 2

Where the paper says it

Consider the symmetry the evidential ratchet breaks. A success counts for a target only if the matching failure would have counted against that same target, under the same contrast and the same background assumptions, and only if the adverse outcome is one the data could have warranted.

This constraint is enforced by recording the test object Vvτ, which includes the matching relation CtrK,H,Bτ(E+) and the depth order ⪯v. p. 5
Where the paper says it

The last four elements are what the Reach Principle runs on. CtrK,H,Bτ (E +) is the matching relation: for each favorable cell E +, it lists the adverse cells the protocol treats as that success’s matching failures, meaning the results that would have counted as this very claim failing, under the same contrast and the same background.

Stopping the Ratchet

The MCI 2.0 profile assigns to each possible outcome what it would count against (N) and what its proponent has agreed to change (R). p. 6

Where the paper says it

0 is the profile that assigns, to each possible outcome, what that outcome would count against and what someone has agreed to change: MCIv(µ, τ ; H, Bτ ) = {Ei 7→ ⟨N(Ei ; H, Bτ ), R(Ei)⟩}n i=1. Conceptually, the equation says that, if we take a measure-role µ at preregistration time τ , a contrast H, and the background Bτ , the index is then a table with one row per cell of the outcome partition, and the row for Ei carries two entries, namely what that outcome would count against (N) and what its proponent has agreed to change (R).

The test object is designed so that revision creates successors rather than erasing what was tested, and support carries over only through what stayed unchanged after testing. p. 2
Where the paper says it

However, what they cannot do is undo the record. The completed adversarial test reached what it reached, and a successor, whether a new test object of the same theory or updated theory-complex, begins with support only for the parts it carried over unchanged from the previous testing.

The final rule states that a success reaches only as deep as its matched failures would have cut, and the timing is the whole point because attribution must be settled before the outcome is known. p. 9
Where the paper says it

When the warrant rule calls an adverse outcome, that outcome counts against the preregistered version at the targets identified in N, the adverse rule from the test object. A success reaches only as deep as its matched failures would have cut; the same contrast, the same background, the same measure-role.

Objections, Neighbors, and Limits of MCI 2. p. 1

Where the paper says it

3 The Reach Principle under a partial order . 10 4 Why structural centrality cannot set the ceiling 11 5 Putting reach on record without locking theories in place 12 6 Prediction 3 and the reach of the test 14 7 Objections, neighbors, and limits of MCI 2.

0

Bayesian modeling brings the strongest objection to the Reach Principle because a theory’s core can make a successful implementation more expected while on failure rational updating lands mostly on a fragile auxiliary. The second objection is that preregistered exposure leaves no room for theories to grow, but the successor rule ensures that revision creates new objects and inherits support only through what stayed unchanged. p. 3

Where the paper says it

Section 4 explains why the position of a prediction in a theory’s graph, however central, cannot settle how far support travels. Section 5 shows that keeping this record leaves revision entirely free, since a revised theory or an amended protocol simply starts a successor that inherits only what it kept.

The MCI 2.0 records the last two concerns, as it sets options side by side before the data arrive. p. 3
Where the paper says it

Cogitate’s Prediction 3 shows the diagnostic value of these consequences, and it shows what goes missing when a preregistration sets pass–fail criteria without saying which target answers for which outcome. The next preregistration is where all of this becomes public before the data arrive.

Prediction 3 and the Reach of the Test

The framework is exemplified by Cogitate’s third prediction, where Tcore was outside both theories’ testable implementations. The MCI 2.0 machinery acts as a diagnostic tool saying what the attempted test did and did not reach. p. 3

Where the paper says it

An adversarial collaboration arranged this way can still run an informative experiment; but the support it yields, and the claims made from it, stay at that level. Cogitate’s Prediction 3 shows the diagnostic value of these consequences, and it shows what goes missing when a preregistration sets pass–fail criteria without saying which target answers for which outcome.

The asymmetry between core bracketed for the rival and core confirmed for one’s own theory is performed within a single record document. p. 6
Where the paper says it

The two subsections that follow carry that symmetry out of the model and into public attribution, and then generalize it to targets that depth cannot rank on a single ladder. Within a fixed model, probability theory already enforces a version of the symmetry the Reach Principle demands.

The next preregistration is where all of this becomes public before the data arrive. p. 3
Where the paper says it

Cogitate’s Prediction 3 shows the diagnostic value of these consequences, and it shows what goes missing when a preregistration sets pass–fail criteria without saying which target answers for which outcome. The next preregistration is where all of this becomes public before the data arrive.

The design consequence for a successor test of Prediction 3 is that its preregistration must state the comparison background, contrast, full outcome partition, matching relation, warrant standard, and negative targets from which each deepest exposed target follows. The record being nothing grander than the preregistered profile itself makes this possible. p. 3

Where the paper says it

Section 2 builds the working parts, namely, the measure-role that bears an index value, the test object that goes on record, and the profile listing, for each possible outcome, what it would count against and what its proponent has agreed to change. Section 3 gives the argument behind the Reach Principle, first as a lemma inside standard confirmation theory, showing that a success whose matching failure was sent to an auxiliary confirms nothing deeper than that auxiliary, then as a reciprocity requirement carrying the same constraint into public attribution.

Success then travels exactly as far as the matching failure would have along a path the theory itself agreed to. p. 3
Where the paper says it

Section 3 gives the argument behind the Reach Principle, first as a lemma inside standard confirmation theory, showing that a success whose matching failure was sent to an auxiliary confirms nothing deeper than that auxiliary, then as a reciprocity requirement carrying the same constraint into public attribution. Section 4 explains why the position of a prediction in a theory’s graph, however central, cannot settle how far support travels.

The core that wants support from experiments must come within their reach. p. 6
Where the paper says it

The two subsections that follow carry that symmetry out of the model and into public attribution, and then generalize it to targets that depth cannot rank on a single ladder. Within a fixed model, probability theory already enforces a version of the symmetry the Reach Principle demands.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Enhanced theory comparison by implementing a Measure Centrality Index 2.0 to record what it would count against and what its proponent has agreed to change for every possible outcome before data collection begins. This allows systems to distinguish between support for a specific implementation and support for the underlying theoretical core, preventing success from being claimed at any depth beyond where the matching failure would have cut.

  2. Automated enforcement of the Reach Principle in experimental design by mandating that a preregistration can state in advance how far every possible outcome reaches. This ensures that support for a theory then reaches only what was exposed ex ante, preventing success from traveling upward through auxiliary apparatuses like instrument coverage or estimator choice without explicit attribution.

  3. Dynamic successor generation for scientific models by implementing a rule where a revised theory or test meets future data as a new object and inherits support only through what stayed unchanged. This allows the AI system to track how theories evolve, ensuring that support carries over only through what stayed unchanged after testing, rather than erasing previous results.

  4. Explicit mapping of evidence sensitivity by incorporating a path pi into the test object to record exactly which of those links a given measure depends on. This allows the system to differentiate between different theoretical versions based on their specific dependencies, addressing the issue where structural centrality cannot settle what a result may be taken to show.

  5. Automated verification of attribution policies by requiring that for any warranted success, the same pre-result attribution policy must identify at least one matched, warrant-eligible outcome of the same versioned measurerole that counts against K. This prevents the evidential ratchet from turning on after the result is known by requiring a pre-defined matching failure to be recorded in advance.

Abstract

A success in adversarial testing of scientific theories reaches only as deep as its matching failure would have cut. I call this the Reach Principle, and the goal of this chapter is to build the instrument that enforces it. Adversarial tests of consciousness theories can turn into a one-way evidential ratchet. When a preregistered prediction passes, the success is claimed for the research program; when it fails, the failure lands on an auxiliary involved in measurement or implementation. The ratchet turns wherever attribution is settled after the outcome is known. Against it, I define a preregisterable test object, putting the attribution policy on the record before the result, comprising what is compared, under which assumptions, when data suffice to settle an outcome, and which targets of a dated theory version each outcome exposes. The Measure Centrality Index 2.0 extends the original version to the positive-evidence side, recording for every possible outcome what it would count against and what its proponent has agreed to change. I run the framework on Cogitate's third prediction, where the preregistered target was the posterior-connectivity implementation and the mathematical or computational core of both Global Neuronal Workspace Theory and Integrated Information Theory stood outside the test. The record shows that a pass could have been claimed at any depth, because nothing fenced success the way the failure was bound. The next preregistration can set reach in advance, and support for a theory then reaches only what was exposed ex ante.

Related papers