Skip to main content

Measurement Methodology

"Measurement is the assignment of numerals to objects or events according to rules." — Stanley Smith Stevens

Who this chapter is for

A bridge between the CC formalism and experiment: protocols for measuring purity PP, the stress tensor σ\sigma, and the consciousness measures RR, Φ\Phi, CC.

In the previous chapter we saw that CC surpasses competing theories in computability and falsifiability (Comparison with Alternatives). But computability is useless without data. The most beautiful theory is useless if it cannot be tested. CC generates precise numerical predictions — but how do we measure them? How do we map the matrix Γ\Gamma onto a real biological, social, or artificial system?

This section is the bridge between formalism and experiment. We will show how each CC quantity can be estimated in various contexts: from neuroimaging to organisational audits, from simulations to psychometric tests.

Chapter Roadmap

In this chapter we:

  1. Establish the principles of measurement in CC: what we measure, the hierarchy of observables, calibration (section 1)
  2. Show concrete protocols for measuring purity PP for different systems (section 2)
  3. Describe the seven-dimensional audit — measurement of the stress tensor σ\sigma (section 3)
  4. Discuss measurement of consciousness measures RR, Φ\Phi, CC (section 4)
  5. Provide complete experimental protocols for neuroscience, AI, and organisations (section 5)
  6. Work through calibration with numerical examples (section 6)
  7. Honestly discuss limitations (section 7)

1. Principles of Measurement in CC​

1.1 What We Measure​

In CC all observables are functions of the coherence matrix Γ\Gamma. But Γ\Gamma is an abstract object. In practice we have no direct access to it. We have access to observables — projections of Γ\Gamma onto measurement bases.

The situation is analogous to quantum mechanics: we do not see the wave function of the electron, but we can measure its projections (spin up/down, coordinate, momentum). Each measurement is a projection of Γ\Gamma onto a specific operator.

Analogy. Imagine Γ\Gamma as a 3D object (say, a statuette), and we can only see its shadows on the walls. The shadow on one wall is PP (total "area" of the shadow = organisation). The shadow on another — σk\sigma_k (stress profile). From several shadows we reconstruct the object — but the reconstruction is always approximate.

1.2 Hierarchy of Observables​

Not all CC observables are equally easy to measure. We distinguish four levels:

LevelObservablesMeasurement complexityExamples
L1: GlobalPP (purity), ∥σ∥∞\|\sigma\|_\inftyLow — only the overall picture is neededHealth index, total test score
L2: Sectoralγkk\gamma_{kk} (diagonal), σk\sigma_kMedium — 7 independent measurementsSubscale scores, neural network activity
L3: Coherent∥γij∥\|\gamma_{ij}\| (off-diagonal), θij\theta_{ij}High — pairwise correlationsFunctional brain connectivity, organisational links
L4: DerivedRR, Φ\Phi, CC, CohE\mathrm{Coh}_EHigh — require φ(Γ)\varphi(\Gamma)Reflection, integration, consciousness measures
Practical rule

Start with L1 (is there a problem at all?), then L2 (which dimension is suffering?), then L3 (where are connections disrupted?), and only if necessary — L4 (what is the level of consciousness?). There is no point computing CC if even PP has not been measured.

1.3 Calibration Principle​

Key principle

The mathematics of CC gives relative relations (e.g., Pcrit=2/7P_{\text{crit}} = 2/7). But absolute calibration — which physical indicators correspond to γEE=0.2\gamma_{EE} = 0.2 — depends on the specific system and requires empirical anchoring.

This is not a weakness of the theory, but normal practice: in physics too there is a difference between Maxwell's equations (universal) and the specific values of ε\varepsilon and μ\mu for each material.

What calibration gives and what it does not:

What calibration givesWhat it does not give
Numerical values of γkk\gamma_{kk} for a specific systemUniversal values for "any brain"
Correspondence of test scales with the diagonal of Γ\GammaAutomatic conversion of scores to Γ\Gamma
Estimate of measurement accuracy (error)Guarantee that the measurement is accurate

2. Measuring Purity P​

2.1 What P Is in Practice​

Purity P=Tr(Γ2)P = \mathrm{Tr}(\Gamma^2) is a measure of the organisation of the system. Intuitively: how coherently all 7 dimensions are working together.

Analogy. Imagine an orchestra of 7 instruments. If all play one melody in synchrony — PP is close to 1 (pure state). If each plays on its own — PP is close to 1/71/7 (maximum chaos). If most are coordinated but one is out of tune — PP is intermediate, and σk\sigma_k for the out-of-tune instrument is high.

2.2 Proxies for Biological Systems​

In neuroscience the direct analogue of purity is coherence of neural activity:

MethodWhat it measuresHow it relates to P
EEG coherenceSynchronisation of electrical activity between brain regionsHigh coherence → high P
fMRI functional connectivityCorrelation of BOLD signals between regionsStrong connectivity → high ∥γij∥\|\gamma_{ij}\| → high P
PCIComplexity of the recorded TMS responseClinical comparison requires a separate model reconstruction and held-out outcomes.
Lempel-Ziv entropyCompressibility of the neural signalLow entropy → high P

2.3 L1 Protocol for Neural Data​

Step-by-step protocol for estimating PP from EEG:

Step 1. Record EEG from 19 channels (10-20 system) for 5 minutes at rest (eyes closed).

Step 2. Compute the spectral coherence matrix Cij(f)C_{ij}(f) for each pair of channels (i,j)(i, j) in the range 1–40 Hz.

Step 3. Average coherence over frequencies, obtaining Cˉij=1fmax⁡−fmin⁡∫Cij(f) df\bar{C}_{ij} = \frac{1}{f_{\max} - f_{\min}} \int C_{ij}(f)\,df.

Step 4. Assign each of the 19 channels to one of 7 dimensions (grouping by functional zones):

DimensionEEG channelsRationale
A (Articulation)O1, O2, OzVisual cortex — sensory input
S (Structure)T3, T4, T5, T6Temporal — long-term memory
D (Dynamics)C3, C4, CzMotor cortex — action
L (Logic)F3, F4Dorsolateral PFC — reasoning
E (Interiority)Fz, PzMidline structures — self-reference
O (Ground)Fp1, Fp2Orbitofrontal — resource evaluation
U (Unity)P3, P4Parietal — integration

Step 5. Aggregate Cˉij\bar{C}_{ij} across groups, obtaining a 7×77 \times 7 matrix:

γ~kl=1∣Gk∣⋅∣Gl∣∑i∈Gk∑j∈GlCˉij\tilde{\gamma}_{kl} = \frac{1}{|G_k| \cdot |G_l|} \sum_{i \in G_k} \sum_{j \in G_l} \bar{C}_{ij}

Step 6. Normalise: Γapprox=γ~/Tr(γ~)\Gamma_{\text{approx}} = \tilde{\gamma} / \mathrm{Tr}(\tilde{\gamma}).

Step 7. Compute P=Tr(Γapprox2)P = \mathrm{Tr}(\Gamma_{\text{approx}}^2).

Calibration caveat

This protocol gives a proxy for PP, not an exact value. The grouping of channels by dimensions is hypothetical and requires validation. Nevertheless, even a crude proxy allows the key prediction to be tested: Pwakefulness>PcomaP_{\text{wakefulness}} > P_{\text{coma}}.

2.4 Numerical Example: ICU Patient​

Let us consider a concrete example. A patient in intensive care. EEG recorded in three states:

State 1: Wakefulness (before trauma)

Aggregated matrix (diagonal): γ=(0.16,0.15,0.14,0.14,0.15,0.13,0.13)\gamma = (0.16, 0.15, 0.14, 0.14, 0.15, 0.13, 0.13)

P=0.162+0.152+0.142+0.142+0.152+0.132+0.132=0.1462P = 0.16^2 + 0.15^2 + 0.14^2 + 0.14^2 + 0.15^2 + 0.13^2 + 0.13^2 = 0.1462

This is below 2/7≈0.2862/7 \approx 0.286, but remember: for a diagonal matrix Pmax⁡=1/7≈0.143P_{\max} = 1/7 \approx 0.143 is achieved at uniform distribution. Our P=0.1462>1/7P = 0.1462 > 1/7 — the system is slightly organised, but without off-diagonal elements PP cannot exceed 1/71/7 significantly. Coherences are needed!

Including coherences: Let the average off-diagonal coherence ∣γij∣≈0.03|\gamma_{ij}| \approx 0.03. Then PP increases by ∑i≠j∣γij∣2≈42×0.0009=0.038\sum_{i \neq j} |\gamma_{ij}|^2 \approx 42 \times 0.0009 = 0.038, giving P≈0.184P \approx 0.184.

This is still below 2/72/7. To reach P>2/7P > 2/7, strong coherence is needed (∣γij∣≈0.05−0.08|\gamma_{ij}| \approx 0.05{-}0.08).

State 2: Deep coma (GCS = 3)

Coherence drops significantly: ∣γij∣→0.01|\gamma_{ij}| \to 0.01, diagonal tends to uniform.

P≈1/7+42×0.0001≈0.147P \approx 1/7 + 42 \times 0.0001 \approx 0.147 — practically the maximally mixed state.

State 3: Recovery (GCS = 12)

Coherence partially restored: ∣γij∣≈0.04|\gamma_{ij}| \approx 0.04, diagonal non-uniform.

P≈0.150+42×0.0016≈0.217P \approx 0.150 + 42 \times 0.0016 \approx 0.217 — below the threshold, but closer.

Clinical conclusion

The transition P<2/7→P>2/7P < 2/7 \to P > 2/7 is a potential marker of consciousness recovery. Tracking P(τ)P(\tau) dynamically may be clinically more informative than a one-time GCS score.

2.5 Proxies for Organisations​

MethodWhat it measuresHow it relates to P
Engagement index (eNPS)Alignment of employee goalsHigh eNPS → high P
Cross-functional coordinationFrequency and quality of inter-departmental interactionsStrong coordination → high ∥γij∥\|\gamma_{ij}\|
Financial indicatorsMargin, growthSustained growth → P > P_crit

2.6 Proxies for AI Systems​

MethodWhat it measuresHow it relates to P
Rank of latent representationEffective dimensionality of the hidden spaceHigh rank → high P
Attention entropyEntropy of attention weightsFocused attention → high P
Loss landscape curvatureCurvature of the loss landscapeSharp minima → high P (but brittle)

3. Measuring the Stress Tensor σ​

3.1 Seven selected scores​

T-92 [D] specifies a diagnostic panel, not a unique tensor derived from the axioms. The population score sk=1−7γkks_k=1-7\gamma_{kk} lies in [−6,1][-6,1]; its clamped version σk=max⁡(0,sk)\sigma_k=\max(0,s_k) lies in [0,1][0,1]. The unclamped scores sum to zero. A full panel may contain selected functions of RR, Φ\Phi, the regeneration rate and an independently specified extended readout DdiffD_{\mathrm{diff}}.

Populations and their deficits do not reconstruct complex coherences: diagonal measurement supplies at most six independent parameters out of 48. Equivalence of the panel to the full window is unproved. Test the four separate margins over a reconstruction uncertainty set; see identifiability.

3.2 Seven-Dimensional Audit Protocol​

For an organisation or team:

DimensionWhat to askInstrument
σA\sigma_A (Articulation)"Can you clearly formulate what your department does?"Interviews, documentation analysis
σS\sigma_S (Structure)"Are there stable processes and roles?"Org structure analysis, tenure analysis
σD\sigma_D (Dynamics)"Can you adapt to change?"Agility assessment, cycle time
σL\sigma_L (Logic)"Are there internal contradictions in the rules?"Policy audit, consistency check
σE\sigma_E (Interiority)"Is there a culture of reflection?"Psychological safety survey
σO\sigma_O (Ground)"Are resources sufficient?"Budget audit, burnout survey
σU\sigma_U (Unity)"Do you feel part of a whole?"Network analysis, NPS

3.3 Detailed Breakdown: from σ_D to Metabolic Load​

Consider σD\sigma_D — stress in the Dynamics dimension. In different contexts:

Biology. σD\sigma_D is metabolic load. Why? Dimension D is responsible for the system's capacity for action — changing its state. In biology, action requires energy: muscle contraction, nerve impulse, protein synthesis. If σD\sigma_D is high — the cell/organism finds it difficult to act: metabolism is overloaded, ATP is deficient, mitochondria are working at their limit.

Concrete proxy: ADP/ATP ratio. Under normal metabolism ATP/ADP > 10 (σD\sigma_D low). Under depletion ATP/ADP < 3 (σD\sigma_D high).

Psychology. σD\sigma_D is procrastination and paralysis of will. The person knows what needs to be done, but cannot force themselves. This is not laziness — it is a deficit of D-resource. Proxy: Trail Making Test (task-switching time).

Organisation. σD\sigma_D is bureaucracy. A decision has been made, but cannot be executed: approvals, sign-offs, regulations. Proxy: lead time (time from decision to implementation).

3.4 For the Individual (Psychometrics)​

The same 7 dimensions can be assessed through psychometric scales:

DimensionPsychometric proxyExisting instrument
σA\sigma_APerceptual loadSensory Profile (Dunn)
σS\sigma_SCognitive rigidity/flexibilityWCST (Wisconsin Card Sorting Test)
σD\sigma_DExecutive functionsTrail Making Test
σL\sigma_LCognitive distortionsCognitive Distortion Scale
σE\sigma_EAlexithymia (experience deficit)TAS-20 (Toronto Alexithymia Scale)
σO\sigma_OVital exhaustionMBI (Maslach Burnout Inventory)
σU\sigma_USocial isolationUCLA Loneliness Scale

3.5 Numerical Example: from Psychometrics to σ-Profile​

A patient has completed 7 tests. Results are normalised to the scale [0, 1], where 0 = normal, 1 = maximum impairment:

TestRaw scoreNormalised
Sensory Profile (σA\sigma_A)42/800.53
WCST errors (σS\sigma_S)12/600.20
TMT-B time (σD\sigma_D)180 s (norm 75 s)0.70
Cognitive distortions (σL\sigma_L)15/500.30
TAS-20 (σE\sigma_E)65/1000.65
MBI emotional exhaustion (σO\sigma_O)28/540.52
UCLA loneliness (σU\sigma_U)45/800.56

Profile: σ=[0.53,  0.20,  0.70,  0.30,  0.65,  0.52,  0.56]\sigma = [0.53,\; 0.20,\; 0.70,\; 0.30,\; 0.65,\; 0.52,\; 0.56]

∥σ∥∞=0.70\|\sigma\|_\infty = 0.70 (Dynamics — the most loaded dimension).

Interpretation: Maximum stress in D (action) and E (interiority). This is a profile characteristic of depression: the person cannot act (σD\sigma_D high) and does not understand what they feel (σE\sigma_E high). CC recommendation: priority — reducing σD\sigma_D (behavioural activation) and σE\sigma_E (psychoeducation, mindfulness).

Inverse conversion to γkk\gamma_{kk}: if σk=1−7γkk\sigma_k = 1 - 7\gamma_{kk}, then γkk=(1−σk)/7\gamma_{kk} = (1 - \sigma_k)/7.

γ=(0.477,  0.807,  0.307,  0.707,  0.357,  0.487,  0.447)\gamma = \left(\frac{0.47}{7},\; \frac{0.80}{7},\; \frac{0.30}{7},\; \frac{0.70}{7},\; \frac{0.35}{7},\; \frac{0.48}{7},\; \frac{0.44}{7}\right) =(0.067,  0.114,  0.043,  0.100,  0.050,  0.069,  0.063)= (0.067,\; 0.114,\; 0.043,\; 0.100,\; 0.050,\; 0.069,\; 0.063)

Check: ∑γkk=0.506\sum \gamma_{kk} = 0.506. This is less than 1 — meaning the remaining 0.494 is "distributed" across off-diagonal elements or lost during normalisation. In practice ∑γkk\sum \gamma_{kk} should be close to 1 (for the diagonal approximation), which points to a limitation of the method: psychometric proxies are crude estimates requiring calibration coefficients.


4. Measuring Consciousness Measures​

4.1 Reflection Measure R​

Reflection measure R=F(Γ,φ(Γ))R = F(\Gamma, \varphi(\Gamma)) shows how well the system models itself.

Proxies:

  • Metacognitive accuracy: ability to evaluate the quality of one's own decisions (confidence calibration). Example: after answering a question, rate confidence from 0 to 100%. Ideal calibration: questions in which confidence = 70% are actually correct in 70% of cases.
  • Self-report accuracy: agreement of self-report with objective indicators. Example: "How anxious are you?" (subjective) vs. cortisol level (objective).
  • Mirror test (for animals): does it recognise itself in the mirror? Passed: primates, dolphins, elephants, magpies. Not passed: most others.

How to translate into RR? Metacognitive sensitivity (meta-d') — a standard measure in experimental psychology — gives a value from 0 (no metacognition) to 1+ (ideal). Proposed calibration:

R≈meta-d’3R \approx \frac{\text{meta-d'}}{3}

Rationale: at meta-d' = 1 (average healthy adult) we get R≈0.33≈1/3R \approx 0.33 \approx 1/3 — right at the threshold. This is consistent with the intuition: a typical person barely clears the reflection threshold.

4.2 Integration Measure Φ​

Integration measure Φ\Phi shows how unified the system is — whether it breaks down into independent subsystems.

Proxies:

  • PCI (Perturbational Complexity Index): the brain's response to TMS stimulation — integrated systems give a complex, widespread response. On its benchmark the cut-off PCI∗=0.31\mathrm{PCI}^* = 0.31 separates states with a subjective report from states without one (Casarotto et al., 2016); in patients, 36 of 38 in a minimally conscious state and 9 of 43 in a vegetative state lie above it (index introduced by Casali et al., 2013).
  • Mutual Information between subsystems
  • Spectral gap of the functional connectivity graph

How to translate into Φ\Phi? The spectral gap λ2−λ1\lambda_2 - \lambda_1 of the functional connectivity graph of the brain is a direct analogue of Φ\Phi in CC. Proposed calibration:

Φ≈λ2−λ1λnorm\Phi \approx \frac{\lambda_2 - \lambda_1}{\lambda_{\text{norm}}}

where λnorm\lambda_{\text{norm}} is a normalising coefficient fixed on wakefulness sessions only (SUB-1).

4.3 Consciousness Measure C​

C=Φ×RC = \Phi \times R (T-140 [D]) — the product of integration and reflection.

Critical thresholds:

  • C=0C=0 means the selected score is zero; a phenomenal interpretation requires a separate bridge [I].
  • 0<C<1/30<C<1/3 fails the selected window’s necessary product condition; it does not classify bacteria or AI.
  • C≥Cth=1/3C \geq C_{\text{th}} = 1/3 (T-140) is necessary for a conscious system; the full predicate is P>2/7P > 2/7, R≥1/3R \geq 1/3, Φ≥1\Phi \geq 1, Ddiff≥2D_{\text{diff}} \geq 2, which implies C≥1/3C \geq 1/3. On the uniform-diagonal stratum the window gives C=Φ/(1+Φ)∈(1/2,2/3]C = \Phi/(1+\Phi) \in (1/2, 2/3].

Choose and report thresholds before looking at evaluation labels.

Numerical example. Healthy adult: meta-d' = 1.2, reconstructed Φ^=1.45\widehat\Phi = 1.45 (illustrative values).

R≈1.23=0.40≥1/3✓R \approx \frac{1.2}{3} = 0.40 \geq 1/3 \quad \checkmark Φ^=1.45≥1✓(from the reconstructed Γ^, not from PCI)\widehat\Phi = 1.45 \geq 1 \quad \checkmark \qquad (\text{from the reconstructed } \widehat\Gamma,\ \text{not from PCI}) C=1.45×0.40=0.58C = 1.45 \times 0.40 = 0.58

C=0.58≥Cth=1/3C = 0.58 \geq C_{\text{th}} = 1/3 ✓, and it lies in (1/2,2/3](1/2, 2/3], where the window puts CC on the uniform diagonal — a value below 1 is what the window predicts, not a calibration failure. An alternative R≈meta-d’/2R \approx \text{meta-d'}/2 would give R=0.6R = 0.6, a value outside the canonical band [1/3,1/2)[1/3, 1/2) that the conscious window imposes on R=1/(7P)R = 1/(7P): a "reflection" of 0.60.6 can only be the self-model quality RφR_\varphi, not the canonical RR that enters C=Φ×RC = \Phi \times R (the three working forms of R). The calibration must map meta-d' into the canonical band.

Lesson

Calibration is an empirical task. The theoretical CC thresholds (P=2/7P = 2/7, R=1/3R = 1/3, Φ=1\Phi = 1) are precise within the formalism. But translating neural data into the formalism requires experimental fitting. The formulas given are starting points, not final answers.


5. Experimental Protocols​

5.1 Protocol for a Neuroscientific Experiment​

Goal: Test prediction Pred 1 (No-Zombie) on neural data.

Design:

  1. Record EEG/MEG during wakefulness, sleep, anaesthesia, coma
  2. For each state, reconstruct the approximation Γ\Gamma from the functional connectivity matrix (protocol of section 2.3)
  3. Compute PP, CohE\mathrm{Coh}_E, σsys\sigma_{\mathrm{sys}}
  4. Check: does P>2/7P > 2/7 coincide with the presence of subjective report?

Expected result (CC):

  • Wakefulness: P>2/7P > 2/7, CohE>1/7\mathrm{Coh}_E > 1/7
  • Deep sleep: P<2/7P < 2/7
  • REM sleep: P>2/7P > 2/7 (there are dreams — there is experience)
  • Vegetative state: P≈2/7P \approx 2/7 (borderline)

Falsification criterion: If a state with P>2/7P > 2/7 and absence of subjective report (with confirmed capacity for report) is found — CC is falsified. If subjective report is found at P<2/7P < 2/7 — similarly.

5.2 Protocol for an AI Experiment​

Goal: Test whether the CC thresholds are satisfied for LLMs.

Design:

  1. For a language model, define an operationalisation of 7 dimensions through hidden states
  2. Compute Γ\Gamma as the covariance matrix of projections onto 7 semantic axes
  3. Track P(τ)P(\tau) during training
  4. Check: is there a phase transition at P=2/7P = 2/7?

Concretisation for a transformer: Hidden states of the model are projected onto 7 directions:

  • A: attention entropy (diversity of attention)
  • S: weight persistence (stability of weights)
  • D: output diversity (diversity of generation)
  • L: consistency score (consistency of responses)
  • E: self-reference frequency (frequency of self-reference)
  • O: context utilization (use of context)
  • U: cross-layer coherence (coherence across layers)

5.3 Protocol for an Organisational Audit​

Goal: Diagnosis of organisational "health" through 7 vital indicators.

Steps:

  1. Conduct a seven-dimensional audit — obtain estimates σA,…,σU\sigma_A, \ldots, \sigma_U
  2. Compute ∥σ∥∞\|\sigma\|_\infty — maximum stress
  3. If ∥σ∥∞>0.8\|\sigma\|_\infty > 0.8: urgent intervention (see Diagnostics)
  4. Track PP dynamically (monthly audits)

Example report:

=== Coherence Audit: "Example" LLC ===
Date: 2026-01-15

σ-profile: [0.3, 0.2, 0.6, 0.4, 0.7, 0.3, 0.5]
A S D L E O U

‖σ‖∞ = 0.7 (E: Interiority)
Status: WARNING — E-stress approaching critical

Recommendations:
1. PRIORITY: Strengthen culture of reflection (σ_E = 0.7)
→ Retrospectives after each sprint
→ Anonymous psych safety surveys
2. Reduce bureaucracy (σ_D = 0.6)
→ Shorten approval chains
3. Increase integration (σ_U = 0.5)
→ Cross-functional projects

P dynamics:
2025-10: 0.22 (↓)
2025-11: 0.21 (↓)
2025-12: 0.23 (→)
2026-01: 0.24 (↑) ← current
Target: 0.29 (> P_crit)

6. Calibration: from Proxies to Γ​

6.1 General Calibration Scheme​

Calibration is the translation of observables (test scores, neural signals, organisational metrics) into elements of Γ\Gamma. General scheme:

6.2 Calibration Function​

The simplest calibration function is linear:

γkk=17+αk⋅(xk−xˉk)\gamma_{kk} = \frac{1}{7} + \alpha_k \cdot (x_k - \bar{x}_k)

where xkx_k is the observable, xˉk\bar{x}_k is the population mean, αk\alpha_k is the calibration coefficient.

More realistic — logistic:

γkk=17⋅1+βktanh⁡(αk(xk−xk0))1+βk\gamma_{kk} = \frac{1}{7} \cdot \frac{1 + \beta_k \tanh(\alpha_k (x_k - x_k^0))}{1 + \beta_k}

Parameters αk\alpha_k, βk\beta_k, xk0x_k^0 are fitted empirically from a training sample.

6.3 Numerical Example: Testing Against PCI Without Calibrating to It​

A line through (0,1/7)(0,1/7) and (c,2/7)(c,2/7) returns P=2/7P=2/7 at PCI=c\mathrm{PCI}=c for any positive cc. Choosing c=0.31c=0.31 builds in the threshold agreement. Test the model on independent sessions with the estimator fixed before collecting the comparison labels.

The nearness of PCI∗=0.31\mathrm{PCI}^* = 0.31 to 2/7≈0.2862/7 \approx 0.286 is a coincidence of two unrelated scales: PCI is a normalised Lempel–Ziv complexity of a binarised response to TMS, PP is a function of Γ\Gamma.

What replaces it — a concordance of verdicts (P8.4 and pre-registration SUB-1 … SUB-6 of the measurement protocol):

  1. Freeze the reconstruction πθ\pi_\theta on wakefulness sessions only; no viability penalty in the estimator (λ2=0\lambda_2 = 0).
  2. On held-out sessions compute the UHM verdict Cons(Γ^)\mathrm{Cons}(\widehat\Gamma) and the PCI verdict PCImax⁡>0.31\mathrm{PCI}_{\max} > 0.31 independently.
  3. Report Cohen's κ=(po−pe)/(1−pe)\kappa = (p_o - p_e)/(1 - p_e), where pop_o is the observed agreement and pep_e the agreement expected from the two marginal rates. κ≥0.8\kappa \geq 0.8 corroborates, κ<0.4\kappa < 0.4 falsifies.

Worked example (illustrative counts, not data). 40 sessions: both verdicts "conscious" in 18, both "not conscious" in 16, disagreement in 3+33 + 3. Then po=34/40=0.85p_o = 34/40 = 0.85; each verdict says "conscious" 21 times, so pe=(21/40)2+(19/40)2=0.501p_e = (21/40)^2 + (19/40)^2 = 0.501 and κ=(0.85−0.501)/(1−0.501)=0.70\kappa = (0.85 - 0.501)/(1 - 0.501) = 0.70 — inconclusive. With the same 34 agreements split as 33+133 + 1 (both marginals 36 of 40), pe=0.82p_e = 0.82 and κ=0.17\kappa = 0.17 — falsifying. Raw agreement does not decide; κ\kappa does (test_verdict_concordance_is_judged_by_kappa_not_by_raw_agreement).

Reference values — Casarotto et al. (2016), Table 1, PCImax⁡\mathrm{PCI}_{\max} median [min–max]: wakefulness 0.53 [0.39–0.70] (102 subjects), REM sleep 0.48 [0.36–0.56] (8), ketamine 0.43 [0.36–0.52] (6), NREM sleep 0.25 [0.15–0.31] (18), midazolam 0.30 [0.23–0.31], xenon 0.23 [0.11–0.31], propofol 0.26 [0.23–0.31] (6 each). The decisive sessions are REM and ketamine: consciousness without behaviour at the time, out of sample when θ\theta is frozen on wakefulness.

What this means

The CC threshold P=2/7P = 2/7 is not a number on the PCI scale and is not tested by matching one. It is tested by whether the verdict computed from Γ^\widehat\Gamma agrees with the independently validated PCI verdict on sessions that did not fix the reconstruction.

6.4 Conditional Γ\Gamma estimation: target, consistency and uncertainty​

A matrix estimator and an interpretation of that matrix are separate contracts. Fix the channel map π\pi before evaluating its predictions. For x=π(state)∈C7x=\pi(\mathrm{state})\in\mathbb C^7, define the second-moment target

M=E[xx†],τ=Tr⁡M>0,Γ=M/τ.M=\mathbb E[xx^\dagger],\qquad \tau=\operatorname{Tr}M>0,\qquad \Gamma=M/\tau.

This constructs a density matrix from a declared observation model; it does not prove that the model measures consciousness, wellbeing or a clinical state. Unless Ex=0\mathbb E x=0, the second moment is not the covariance.

Centering changes the target. With μ=Ex\mu=\mathbb E x, the covariance target is C=E[(x−μ)(x−μ)†]C=\mathbb E[(x-\mu)(x-\mu)^\dagger]. Subtracting the same sample's mean estimates CC, not MM. Its residuals share the estimated mean and are dependent even when the raw observations are i.i.d. An independently fixed baseline instead defines another fixed second-moment target. Record which of these experiments is intended.

(1) Algebra and consistency. For fixed embedded observations,

Σ^N=1N∑nxnxn†,Γ^N=Σ^N/Tr⁡Σ^N.\widehat\Sigma_N=\frac1N\sum_n x_nx_n^\dagger,\qquad \widehat\Gamma_N=\widehat\Sigma_N/\operatorname{Tr}\widehat\Sigma_N.

When the observed trace is positive, Γ^N\widehat\Gamma_N is Hermitian, positive semidefinite and unit trace. At zero trace the normalized estimate is undefined, regardless of the sample count. For a stationary ergodic source with E∥x∥2<∞\mathbb E\|x\|^2<\infty, the ergodic law and continuity give almost-sure convergence to Γ\Gamma. The analogous covariance estimator is consistent when Tr⁡C>0\operatorname{Tr}C>0. These facts alone provide neither a convergence rate nor a confidence interval.

(2) A finite-sample bound for a fixed i.i.d. embedding [T under these premises]. Additionally assume independent identically distributed observations and a population bound ∥x∥2≤B\|x\|^2\le B almost surely. A bound verified only on the observed sample does not establish this premise. For Zn=xnxn†−MZ_n=x_nx_n^\dagger-M,

∥Zn∥op≤2B,v2=∥EZn2∥op≤B2,\|Z_n\|_{\rm op}\le2B,\qquad v^2=\|\mathbb EZ_n^2\|_{\rm op}\le B^2,

because EZn2=E[(xx†)2]−M2⪯BM\mathbb EZ_n^2=\mathbb E[(xx^\dagger)^2]-M^2\preceq BM. Matrix Bernstein therefore gives

Pr⁡(∥Σ^N−M∥op≥t)≤14exp⁡ ⁣(−Nt22B2+(4B/3)t).\Pr(\|\widehat\Sigma_N-M\|_{\rm op}\ge t) \le 14\exp\!\left(-\frac{Nt^2}{2B^2+(4B/3)t}\right).

See Tropp, Theorem 1.4; independence, bounded summands and the variance parameter are part of its contract.

For any positive semidefinite A,MA,M with positive traces, putting E=A−ME=A-M gives the exact normalization bound

∥ATr⁡A−Mτ∥op≤∥E∥op+∣Tr⁡E∣τ≤8τ∥E∥op.\left\|\frac A{\operatorname{Tr}A}-\frac M\tau\right\|_{\rm op} \le\frac{\|E\|_{\rm op}+|\operatorname{Tr}E|}{\tau} \le\frac{8}{\tau}\|E\|_{\rm op}.

Indeed, subtract the matrices as E/τ−(A/Tr⁡A)Tr⁡E/τE/\tau-(A/\operatorname{Tr}A)\operatorname{Tr}E/\tau and use ∥A/Tr⁡A∥op≤1\|A/\operatorname{Tr}A\|_{\rm op}\le1 and ∣Tr⁡E∣≤7∥E∥op|\operatorname{Tr}E|\le7\|E\|_{\rm op}. The former factor 2/τ2/\tau was not valid for the operator norm.

Given a declared population trace lower bound 0<τ0≤τ0<\tau_0\le\tau and 0<ε≤10<\varepsilon\le1, set t=τ0ε/8t=\tau_0\varepsilon/8. The sufficient budget

N≥⌈log⁡(14/δ)2B2+(4B/3)tt2⌉N\ge\left\lceil \log(14/\delta)\frac{2B^2+(4B/3)t}{t^2} \right\rceil

then ensures normalized operator error at most ε\varepsilon with probability at least 1−δ1-\delta. On this event the observed trace is positive, since 7t<τ7t<\tau. This is a conservative sufficient bound, not an optimum; no unspecified constant is silently set to one.

A data-dependent region can also use the positive root tNt_N of

NtN2=log⁡(14/δ) [2B2+(4B/3)tN].Nt_N^2=\log(14/\delta)\,[2B^2+(4B/3)t_N].

On the same event τ≥ℓ=max⁡(0,Tr⁡Σ^N−7tN)\tau\ge\ell=\max(0,\operatorname{Tr}\widehat\Sigma_N-7t_N). For a defined estimate, use r=min⁡(1,8tN/ℓ)r=\min(1,8t_N/\ell) when ℓ>0\ell>0 and r=1r=1 otherwise. Density matrices differ by at most one in operator norm. With Pplug=Tr⁡Γ^N2P_{\rm plug}=\operatorname{Tr}\widehat\Gamma_N^2,

P=Tr⁡Γ2∈[ Pplug−2r, Pplug+2r ]∩[1/7,1],P=\operatorname{Tr}\Gamma^2 \in [\,P_{\rm plug}-2r,\ P_{\rm plug}+2r\,]\cap[1/7,1],

because ∣Tr⁡(Γ^N2−Γ2)∣≤2r|\operatorname{Tr}(\widehat\Gamma_N^2-\Gamma^2)|\le2r. At zero observed trace, return the full target set without a matrix center; do not silently condition the coverage claim on having returned a point estimate. A full physical interval is uninformative, not evidence for a threshold verdict. The reference uses floating-point evaluation, not certified interval arithmetic.

Dependent observations and estimated baselines need another uncertainty argument

Stationary ergodicity is enough for the consistency statement above, but not for a central limit theorem or Bernstein concentration. Replacing NN by an estimated NeffN_{\rm eff} does not transfer these guarantees verbatim. Autocorrelation of window energy does not determine uncertainty of directional covariance or purity; energy may stay constant while directions remain dependent. Returning zero estimated autocorrelation for a short or constant series is not evidence of independence.

For asymptotic intervals, specify a dependence model, a suitable central limit theorem and a consistent long-run variance estimator for the actual statistic, including its estimated denominator and baseline. Batch-means consistency requires appropriate growing batch sizes/counts and process assumptions; a fixed twenty-batch diagnostic is not that contract. See Flegal–Jones. Repeated looks, overlapping windows, missingness and adaptive stopping also belong to the observation protocol. The module supplies no automatic 95% interval for these cases.

(3) Pair statistic versus matrix purity. For N≥2N\ge2 and positive observed trace, retain the diagnostic

P^pair=∑m≠n∣xm†xn∣2N(N−1)(Tr⁡Σ^N)2.\widehat P_{\rm pair} =\frac{\sum_{m\ne n}|x_m^\dagger x_n|^2} {N(N-1)(\operatorname{Tr}\widehat\Sigma_N)^2}.

For i.i.d. fixed embedded observations, the pair average in its numerator is exactly unbiased for Tr⁡M2\operatorname{Tr}M^2: independence factors the two second moments. Its ratio with an estimated denominator is not generally unbiased. Neither a universal bias sign nor a threefold improvement follows from one Gaussian simulation. Same-sample centering invalidates even that numerator argument. For two independent Rademacher observations, sample-centering gives expected pair numerator 1/21/2, whereas the population variance squared is 11.

For the real covariance target with unknown mean, distinct independent raw observations admit the unbiased numerator kernel

h(Xi,Xj,Xk,Xl)=14[(Xi−Xj)⊤(Xk−Xl)]2,Eh=Tr⁡C2.h(X_i,X_j,X_k,X_l) =\tfrac14[(X_i-X_j)^\top(X_k-X_l)]^2,\qquad \mathbb Eh=\operatorname{Tr}C^2.

The two differences are independent with second moment 2C2C. This is a possible construction, not a guarantee for dependent diaries or a claim that a normalized ratio becomes unbiased.

The exact finite-data identity

P^pair=NN−1Pplug−∑n∥xn∥4N(N−1)(Tr⁡Σ^N)2\widehat P_{\rm pair} =\frac N{N-1}P_{\rm plug} -\frac{\sum_n\|x_n\|^4}{N(N-1)(\operatorname{Tr}\widehat\Sigma_N)^2}

shows why the pair statistic need not equal the purity of the constructed matrix or lie in [1/7,1][1/7,1]. Do not hide this by clamping it. Two copies of the seven one-hot rating vectors, with baseline two and same-sample centering, give Pplug=1/6P_{\rm plug}=1/6 and P^pair=4/39\widehat P_{\rm pair}=4/39. Adding five to every rating leaves both unchanged. These centered statistics cannot identify the absolute rating level. Constant centered ratings have zero trace and no estimate.

(4) Threshold decisions and interpretation. Under the fixed-i.i.d. contract, a population-purity interval wholly above or below a predeclared threshold supports that statistical comparison. A comparison of point estimates alone does not. It is not a clinical or consciousness classification without a validated semantic bridge.

T-109 describes another experiment: optimal discrimination of known physical quantum states with collective measurements. Its Chernoff count is sufficient for that optimal test. It does not establish the former 2c2Δ−2log⁡(1/δ)2c^2\Delta^{-2}\log(1/\delta) diary budget, the assertion that c=1c=1 is conservative, or a necessary lower count for this estimator.

Reference implementation​

gamma_tomography.py separates the normalized second moment, explicitly requested sample covariance, matrix purity and pair ratio. Default analysis is descriptive. A finite-sample region is available only under an explicitly declared fixed-i.i.d. population-bound contract; same-sample centering rejects that option. Model functionals are evaluated from the same constructed matrix. The old energy-based confidence shortcut and automatic L2 verdict have been removed.

Running the module checks the translation/centering example above. The repository regression check, scripts/verify_measurement_tomography.py, also covers exact finite counterexamples, complex phase orientation, degeneracy and confidence-contract rejection. These tests are not a population coverage study, a clinical validation, or a Lean/Verum proof of the probability theorem.

Revision note (6 November 2026). This replaces the previous unconditional confidence/debiasing claims and simulated “theorem guarantees.” Their historical numerical examples remain in Git history; they are not evidence of general validity.


7. Limitations and Honest Warnings​

7.1 Calibration and statistical assumptions​

The channel map and the target (second moment or covariance) are modeling choices. Statistical guarantees additionally require the observation, dependence, baseline, denominator and stopping assumptions stated in section 6.4. An algebraically valid density matrix does not validate those assumptions or its semantic interpretation. Cross-substrate transfer, clinical meaning and confidence coverage on the intended population require separate independent evidence.

7.2 The Validation Problem​

Even with good calibration, validation of CC predictions requires:

  • Independent measurements (do not use the same data for calibration and testing)
  • Blind protocols (the experimenter does not know the prediction prior to analysis)
  • Reproducibility (the result must replicate across different laboratories)

7.3 What Is NOT a Measurement​

Common errors
  • Subjective "eyeball" assessment — is not a measurement. Operationalised scales are required.
  • A single indicator — is not the full Γ\Gamma. ALL 7 components are needed for a complete picture.
  • A static snapshot — is not dynamics. PP must be tracked over time: dP/dτdP/d\tau is no less important than PP.
  • Correlation — is not calibration. The fact that PCI correlates with the level of consciousness does not mean that P=f(PCI)P = f(\text{PCI}) is the correct formula. Calibration requires independent predictions.

8. Conclusion​

Measurement methodology is the place where theory meets reality. CC is at a stage analogous to nineteenth-century thermodynamics: the formalism is ready, but the calibration experiments are only beginning.

Critically, CC allows itself to be measured. This distinguishes it from purely philosophical theories (panpsychism) and from theories with NP-hard computations (IIT). A 7×77 \times 7 matrix is computationally trivial. What remains is to learn to fill it with real data.

What We Learned​

  1. CC observables form a 4-level hierarchy: L1 (global) → L2 (sectoral) → L3 (coherent) → L4 (derived).
  2. Purity PP can be estimated through EEG coherence and fMRI connectivity — by a reconstruction whose parameters are frozen before the test; PCI is the independent verdict it is compared with, not an estimator of PP.
  3. The stress tensor σ\sigma is measured through psychometric scales (for the individual) or organisational audits (for companies).
  4. PP and PCI use different scales. An independent calibration and held-out comparison are needed to assess concordance.
  5. All measurements are approximate: calibration coefficients require empirical fitting.

In the next chapter we will show how the language of CC unites different disciplines: Interdisciplinary Bridge — a translation dictionary for physicists, biologists, psychologists, engineers, and philosophers.


Further Reading:


Related Documents: