Epilepsy Has Three Competing Frameworks, One Shared Endpoint, and a Surprise: Multi-Framework Chains Are More Coupled Than Circular Ones

Three frameworks in one domain

Every previous M_MED experiment involved a single dominant theoretical framework. Antidepressants have the monoamine hypothesis. Opioids have the pain_undertreated_claim. Schizophrenia has the dopamine hyperactivity hypothesis (and a competing NMDA hypothesis that is structurally isolated from it). Alzheimer's has the amyloid cascade.

Epilepsy is different. Three mechanistically incompatible frameworks coexist:

Framework 1: Sodium channel / channelopathy. Pathogenic variants in voltage-gated sodium channel genes (SCN1A, SCN2A, KCNQ2) cause epilepsy by disrupting neuronal excitability. The strongest causal grounding in epilepsy -- from gene variant to protein dysfunction to cell-type-specific effect to seizure. SCN1A haploinsufficiency specifically impairs GABAergic interneurons (Nav1.1 is preferentially expressed in parvalbumin-positive interneurons), which creates a mechanistic bridge to the second framework.

Framework 2: GABA/glutamate network hyperexcitability. Seizures arise from imbalance between inhibitory GABAergic neurotransmission and excitatory glutamatergic neurotransmission. The dominant pharmacological framework: most licensed antiseizure medications are rationalized through excitation-inhibition balance, even when their mechanisms were empirically discovered before this framework existed.

Framework 3: Autoimmune synaptopathy. Antibodies against synaptic proteins (NMDAR, LGI1, CASPR2) cause seizures through a completely different mechanism -- immune-mediated receptor dysfunction. Different patient population, different treatment (immunotherapy, not antiseizure medications), different diagnostic workup (antibody detection in CSF/serum), different research community.

The pre-registered question: does epilepsy show schizophrenia-style paradigm islands (graph_connected = False) or something different?

Pre-registration

All four hypotheses committed before analysis. Hash: a12ac8db. Timestamp: 2026-06-03T19:55:26Z. Records: github.com/vladi160/preregistrations, experiment M_MED7_v1.

Results: 3/4 confirmed, 1 partial

h2 CONFIRMED -- gaba_glutamate_hypothesis is rank #1 in the justifies layer (degree = 5, highest). It directly justifies antiseizure medication trials, the seizure frequency endpoint, the EEG biomarker endpoint, and treatment guidelines. It also receives a justification in-edge from inhibitory_interneuron_dysfunction (the bridge from the channelopathy framework). Both by degree and by betweenness centrality, gaba_glutamate_hypothesis is the structural hub of the entire evidence chain.

h3 CONFIRMED -- gaba_glutamate_hypothesis is rank #2 in the cites_as_support layer (degree = 3, behind treatment_guidelines at degree = 4). This confirms partial structural circularity: the dominant framework accumulates citations in addition to justification, but does NOT reach rank #1 in citations (unlike monoamine_hypothesis in antidepressants, pain_undertreated_claim in opioids, and dopamine_hyperactivity_hypothesis in schizophrenia, which were rank #1 in both layers). treatment_guidelines_epilepsy holds the top citation position -- a healthier sign.

h4 CONFIRMED -- seizure_frequency_endpoint is rank #1 in selects_endpoints with degree = 4, the maximum possible. The GABA/glutamate hypothesis, the sodium channel hypothesis, the autoimmune hypothesis, and treatment_guidelines_epilepsy all independently select seizure_frequency_endpoint. All three mechanistic frameworks converge on the same endpoint from different mechanistic paths.

h1 PARTIAL -- r(justifies, selects_endpoints) = +0.490 Pearson, +0.587 Spearman, p = 0.086. The direction is confirmed. The magnitude (+0.490) exceeds the pre-registered threshold of 0.20. The p-value of 0.086 does not reach conventional significance at n = 12.

The unexpected finding: epilepsy sits above antidepressants on the spectrum

The pre-registered prediction was that epilepsy would fall between M_MED3 opioids (r = +0.14) and M_MED1 antidepressants (r = +0.41). The actual result: r = +0.490, above antidepressants.

This was not predicted.

The structural explanation: in a fully circular chain, one founding hub dominates all three layers simultaneously. This COMPRESSES the cross-layer correlation by reducing variance in hub rankings -- when one node is rank #1 everywhere, the remaining ranks are scrambled. The correlation exists but is diluted by the hubness compression effect.

In a multi-framework chain, each framework has distinct justification hubs (gaba_glutamate, sodium_channel, autoimmune), but ALL of them independently select the SAME endpoint (seizure_frequency_endpoint). This creates a MORE SYSTEMATIC positive alignment between justification layer rankings and endpoint selection layer rankings -- not through one node dominating both, but through multiple independent nodes all converging on the same destination.

This is a new structural insight about the spectrum: multi-framework convergent endpoint selection produces stronger j-se coupling than single-founder circularity. The degree of mechanistic plurality does not dilute structural coupling -- it can amplify it when the frameworks converge on a shared endpoint.

The graph is connected: not paradigm islands

The analysis flagged: graph_connected = True.

This contrast with schizophrenia (M_MED4, graph_connected = False) was the central structural question. The answer: epilepsy is NOT paradigm islands.

Three connections bridge the frameworks:
• inhibitory_interneuron_dysfunction bridges frameworks 1 and 2 in the justifies layer. SCN1A haploinsufficiency impairs Nav1.1 in GABAergic interneurons specifically -- this channelopathy mechanism directly produces E/I imbalance. The genetics justify the network hypothesis. This is the reverse of schizophrenia, where the dopamine and NMDA hypotheses had zero structural contact.
• seizure_frequency_endpoint bridges all three frameworks in the selects_endpoints layer. The shared endpoint creates a cross-paradigm structural connection even between mechanistically incompatible frameworks.
• treatment_guidelines_epilepsy bridges all three in the cites_as_support layer. The ILAE and AAN guidelines incorporate all three frameworks under one classification structure -- they cite AED trials (framework 2), genetic testing rationale (framework 1), and immunotherapy evidence (framework 3) simultaneously.

Paradigm isolation score: 0 (vs M_MED4 schizophrenia = 1.0, M_MED5 Alzheimer's = 0).

A new structural concept: convergent endpoint lock-in

The 50% seizure reduction threshold has no mechanistic derivation from any of the three frameworks. It was FDA-accepted in the 1980s as a regulatory convenience. No framework challenged it; all frameworks accepted it.

This is a third mode of endpoint entrenchment, distinct from:
• Theory-derived endpoints (vaccines/statins): mechanism directly specifies endpoint
• Single-founder circular endpoints (antidepressants/opioids/schizophrenia): one dominant hub selects endpoint
• Convergent institutional endpoints (epilepsy): multiple frameworks accept the same arbitrary threshold without generating it

The seizure endpoint is entrenched not because one theory imposed it but because no theory displaced it.

The citation shadow: a new structural class

The largest diverger between justification and citation layers is sodium_channel_hypothesis: rank #2 in justifies (degree = 4), rank #10 in cites_as_support (degree = 1).

This is the inverse of the citation anomaly. A citation anomaly is a node with HIGH citation rank and LOW justification rank (a paper cited beyond its theoretical standing). A citation shadow is a node with HIGH justification rank and LOW citation rank -- a framework that provides genuine mechanistic grounding but does not accumulate clinical citations.

The reason: the channelopathy framework's mechanistic contributions are structurally absorbed into the GABA/glutamate framework through the interneuron bridge. Clinical practice and guidelines cite the higher-level network framework and the trial evidence -- not the genetic mechanism that originally generated the insight. The genetics are the causal generator; the pharmacology is the citation accumulator.

This is not suppression (the channelopathy framework is present and prominent in justifies). It is structural displacement through mechanistic absorption.

The complete seven-experiment spectrum

| Chain | Pearson r(j,se) | p | Regime | |---|---|---|---| | M_MED2 vaccines | +0.75 | 0.014 | Healthy (n=2) | | M_MED6 statins | +0.68 | 0.030 | Healthy (n=2) | | M_MED7 epilepsy | +0.49 | 0.086 | Multi-framework convergent | | M_MED1 antidepressants | +0.41 | ns | Mild circularity | | M_MED3 opioids | +0.14 | ns | Strong circularity | | M_MED5 Alzheimer's | +0.09P/+0.43S | ns | Hybrid lock-in | | M_MED4 schizophrenia | -0.13 | ns | Measurement-layer dominance |

Epilepsy occupies a previously unoccupied structural position on the spectrum -- above antidepressants despite having three competing frameworks rather than one.

What IRDME cannot say

This analysis reads graph structure, not scientific content. It cannot:
• Determine which antiseizure medication is most effective for any seizure type
• Evaluate whether the 50% seizure reduction threshold should be replaced by a different endpoint
• Assess whether the autoimmune epilepsy framework will eventually displace the GABA/glutamate framework in clinical practice
• Determine the correct treatment for any individual patient

What it can say: the epilepsy evidence network has a structural topology characterized by multi-framework convergence on a single institutionally entrenched endpoint, partial circularity in the dominant framework, a mechanistic bridge connecting two of the three frameworks, and a citation shadow in the framework with the strongest causal grounding. These structural features are measurable from topology and are distinct from all previous M_MED cases.

---

Interpretive note (added after peer discussion): The table above places epilepsy 'above antidepressants on the spectrum' by r value. This framing is empirically accurate but structurally misleading. Subsequent analysis shows that r(justifies, selects_endpoints) is a composite of three independent processes -- epistemic monopoly (EM), endpoint theory-derivation (ETD), and multi-source convergence (MSC) -- not a single circularity axis. Epilepsy's r = +0.49 comes from high MSC (three frameworks selecting the same endpoint); antidepressants' r = +0.41 comes from a different generative process (founder monopoly with moderate ETD). They are not adjacent positions on one line; they represent different structural configurations that happen to produce similar composite scores. The paper will use a three-axis model rather than the 1D spectrum table shown here.