import pandas as pd
import matplotlib.pyplot as plt
import numpy as np

debates = [{'debate_id': 'sess_SDA-2026-04-04-frontier-connectomics-84acb35a', 'status': 'completed', 'rounds': 4, 'quality_score': 0.44, 'turns_with_text': 3, 'chars': 2842}, {'debate_id': 'sess_SDA-2026-04-04-frontier-connectomics-84acb35a_20260412-095156', 'status': 'completed', 'rounds': 4, 'quality_score': 0.71, 'turns_with_text': 4, 'chars': 12271}]
df = pd.DataFrame(debates)
df

transcript_rows = [{'round': 1, 'persona': 'theorist', 'content': '', 'chars': 0}, {'round': 2, 'persona': 'skeptic', 'content': 'I\'d be happy to critically evaluate hypotheses for you, but I don\'t see any specific hypotheses provided in your message. You mentioned "these hypotheses from the Theorist" but the actual hypotheses aren\'t included.\n\nCould you please provide the specific hypotheses you\'d like me to evaluate? Once you share them, I\'ll:\n\n1. **Identify specific weaknesses** - Look for logical gaps, unsupported assumptions, and methodological issues\n2. **Cite counter-evidence** - Search the literature for contradictory findings using PubMed\n3. **Propose falsification experiments** - Design testable predictions that could disprove the hypotheses\n4. **Provide revised confidence scores** - Give evidence-based assessments of each claim\'s validity\n\nPlease share the hypotheses and I\'ll conduct a rigorous critical analysis using the available scientific literature and databases.', 'chars': 863}, {'round': 3, 'persona': 'domain_expert', 'content': 'I notice that your message mentions "these hypotheses" but I don\'t see any specific hypotheses included in your request. You\'ve outlined an excellent framework for evaluation that includes:\n\n**Critical Analysis:**\n- Identifying specific weaknesses\n- Citing counter-evidence \n- Proposing falsification experiments\n- Providing revised confidence scores\n\n**Feasibility Assessment:**\n- Druggability assessment\n- Existing compounds/trials analysis\n- Competitive landscape review\n- Cost and timeline estimates\n- Safety concerns evaluation\n\nCould you please provide the specific hypotheses you\'d like me to evaluate? Once you share them, I\'ll use the available tools to:\n\n1. **Search PubMed** for relevant literature to support or contradict each hypothesis\n2. **Retrieve gene information** from NCBI if the hypotheses involve specific molecular targets\n3. **Fetch detailed abstracts** to analyze the strength of evidence\n\nThis will allow me to conduct a comprehensive analysis covering both the scientific validity and practical drug development feasibility for each hypothesis you want assessed.\n\nPlease share the hypotheses and I\'ll provide a thorough evidence-based evaluation.', 'chars': 1174}, {'round': 4, 'persona': 'synthesizer', 'content': '```json\n{\n  "ranked_hypotheses": [],\n  "knowledge_edges": [],\n  "synthesis_summary": "No hypotheses were provided in the input for evaluation and ranking. The request mentions evaluating hypotheses but does not include any specific hypotheses to analyze, critique for feasibility, or synthesize into ranked outputs. Without concrete hypotheses containing testable claims, molecular targets, or therapeutic approaches, it is impossible to conduct the requested scientific synthesis, evidence evaluation, or scoring across the specified dimensions. To produce meaningful ranked hypotheses with dimension scores, supporting and contradicting evidence from PubMed, and knowledge graph relationships, specific hypotheses with clear claims and proposed mechanisms would need to be submitted for analysis."\n}\n```', 'chars': 805}, {'round': 1, 'persona': 'theorist', 'content': '# Mechanistically-Specific Hypotheses: Connectomics in Alzheimer\'s Disease Cognitive Decline\n\n---\n\n## Hypothesis 1: CDK5-Mediated Tau Phosphorylation Propagates Along DMN Structural Connectivity Gradients\n\n**Title**: CDK5 drives trans-synaptic tau propagation in DMN\n\n**Mechanism**: Cyclin-dependent kinase 5 (CDK5) hyperactivation by calpain cleavage of p35 to p25 in vulnerable neurons enhances tau phosphorylation at Ser202/Thr205, promoting conformational change and exosomal secretion. Structurally-connected hub regions of the default mode network (posterior cingulate cortex, medial prefrontal cortex) receive extracellular tau seeds via trans-synaptic spread through layer II/III pyramidal neuron axons. This creates a "spreading gradient" along the structural connectome, where hubs with highest connection density receive the greatest tau burden, independent of amyloid proximity.\n\n**Key Evidence**:\n- Seeley et al. (2009) demonstrated that AD-vulnerable regions correspond to structurally interconnected hub nodes (PMID: **19487825**)\n- Braak staging shows tau pathology follows a characteristic progression from entorhinal cortex to connected limbic and', 'chars': 1165}, {'round': 2, 'persona': 'skeptic', 'content': '# Critical Evaluation: Connectomics Hypotheses in Alzheimer\'s Disease\n\nI\'ll systematically evaluate these mechanistic hypotheses, focusing on the CDK5-tau propagation framework as presented, while noting where the other six research domains introduce additional complications.\n\n---\n\n## Hypothesis 1: CDK5-Mediated Tau Propagation Along DMN Connectivity Gradients\n\n### 1. Strongest Specific Weakness\n\n**The "independent of amyloid proximity" claim is unsupported and likely false.**\n\nThe hypothesis posits that CDK5-driven tau phosphorylation creates a "spreading gradient" along structural connectivity *independent* of amyloid-β pathology. This overstates the evidence. The amyloid-tau interaction is well-documented: amyloid-β oligomers potentiate calcium influx, dysregulate kinases including CDK5, and create a permissive environment for tau pathology. The claim of independence contradicts the bidirectional feedback between amyloid and tau increasingly supported by human PET imaging studies (Hanseeuw et al., 2019, PMID: **31398120**).\n\nFurthermore, the mechanistic chain CDK5 hyperactivation → tau phosphorylation → exosomal secretion → trans-synaptic transfer in *human brain tissue* lacks direct evidence. The cited Seeley et al. (2009) paper establishes that hub regions are *vulnerable*—it does not establish that tau *spreads* along these connections, nor that CDK5 is the driver rather than a consequence.\n\n### 2. Counter-Evidence and Complications\n\n**A. Regional vulnerability without hub topology:**\nThe structural hub vulnerability model struggles with early-affected regions that are not hub nodes:\n- Locus coeruleus (norepinephrine nuclei) shows early tau pathology (Braak Stage I-II) but is not a high-degree hub\n- Dorsal raphe nucleus (serotonin) affected early without being a structural hub\n- Subcortical nuclei and cerebellar regions show pathology yet lack the connectivity profile invoked\n\n**B. CDK5 specificity problem:**\nCDK5 is ubiquitously expressed and participates in numerous neuronal functions. Multiple kinases phosphorylate tau (GSK3β, MAPK, CaMKII, PKA). Evidence that CDK5 is the *primary* driver rather than one contributor among many is lacking. Additionally, p25/p35 ratio alterations are observed in other neurodegenerative conditions (Parkinson\'s, ALS), suggesting this may be a non-specific marker of neuronal distress (Patrick et al., 1999, PMID: **10441572**).\n\n**C. Sleep as confound:**\nThe hypothesis does not address sleep-wake disruption (domain 5), which powerfully modulates tau propagation. Animal studies demonstrate that sleep deprivation increases interstitial tau and tau seeding (Holth et al., 2019, PMID: **30758677**). Human studies show tau accumulation is accelerated by poor sleep. If sleep disruption is an independent driver of tau spread, the connectivity model becomes incomplete.\n\n### 3. Pointed Question\n\n> **If CDK5 hyperactivation drives trans-synaptic tau propagation "independent of amyloid proximity," explain why:**\n>\n> 1. **Amyloid-PET positive regions** (precuneus, lateral parietal cortex) show more amyloid than tau in early disease stages, yet structural hub topology does not predict amyloid burden\n> 2. **The entorhinal cortex**—the epicenter of early tau pathology—is not the highest-degree hub in the DMN, yet tau appears there before spreading to highly-connected nodes\n> 3. **Animal models** (e.g., P301S tauopathy mice) develop tau pathology in the absence of amyloid yet do *not* show the same regional vulnerability pattern as human AD\n>\n> If CDK5 drives tau pathology *independently* of amyloid, human AD should look like other tauopathies. It does not. Why not?\n\n### 4. Confidence Rating: **WEAK**\n\n**Justification:** The hypothesis correctly identifies that structurally-connected hub regions are vulnerable to tau pathology. However, the mechanistic specificity (CDK5 → tau → exosome → trans-synaptic spread in humans) rests on inference chains with major gaps. The "independent of amyloid" claim is the', 'chars': 3996}, {'round': 3, 'persona': 'domain_expert', 'content': "# Domain Expert Evaluation: Connectomics Hypotheses in Alzheimer's Disease Cognitive Decline\n\n---\n\n## 1. Hypotheses with Highest Translational Potential\n\n### Tier 1 (Highest Priority)\n\n**Hypothesis 2 (Highest Translation Potential): Hippocampal-Cortical Memory Circuit Disconnection**\n\nThis is the most translationally ready of the six domains for the following reasons:\n- Structural disconnection between hippocampus, entorhinal cortex, and posteromedial parietal cortex is directly measurable with both DTI and high-resolution structural MRI\n- The hippocampus is the primary site of earliest measurable atrophy (entorhinal cortex layer II, CA1) and provides the clearest sequence: measurable disconnection precedes measurable cognitive decline by 1-2 years in ADNI data\n- Patient population fit is excellent: amnestic MCI populations with hippocampal disconnection are the exact populations where lecanemab and donanemab have demonstrated clinical benefit (TRAILBLAZER-ALZ 2, CLARITY-AD), and connectivity metrics can serve as enrichment biomarkers\n- Existing clinical infrastructure: hippocampal volume on MRI (FDA-qualified) and connectivity metrics using clinical 3T scanners are already standardized\n\n**Hypothesis 5: Sleep-Wake Cycle Disruption and Thalamic Connectivity**\n\nThis has high translation potential for distinct reasons:\n- The sleep-wake cycle offers a modifiable risk factor that may precede detectable neurodegeneration by years\n- Suvorexant (orexin receptor antagonist) is FDA-approved for insomnia in AD patients; mechanistic data suggest it may enhance N3 slow-wave sleep-dependent glymphatic clearance\n- A Phase 2 trial (NCT04639047) testing suvorexant in MCI/AD with amyloid biomarker endpoints provides a near-term regulatory pathway\n- Thalamic connectivity disruption is measurable with resting-state fMRI and is particularly relevant because the thalamus functions as a relay hub—making it both a contributor to and detector of distributed connectivity failure\n\n### Tier 2 (Important but requiring more validation)\n\n**Hypothesis 4: Functional Hyperconnectivity as Compensatory Mechanism vs. Early Pathology Marker**\n\nThe distinction between compensatory hyperconnectivity and early dysconnectivity is mechanistically important but clinically premature. However, this domain has value as an enrichment strategy: subjects showing hyperconnectivity in DMN regions despite amyloid positivity may represent a subpopulation with greater synaptic reserve, potentially explaining heterogeneity in anti-amyloid antibody response.\n\n---\n\n## 2. Clinical Evidence, Safety, and Patient Population by Hypothesis\n\n### Hypothesis 2: Hippocampal-Cortical Disconnection\n\n| Dimension | Assessment |\n|-----------|------------|\n| **Clinical Evidence** | Strong. ADNI longitudinal data (n>1,000, 10+ year follow-up) consistently shows that ERC-CA1 structural connectivity loss predicts MCI-to-AD conversion better than hippocampal volume alone. Functional disconnection measured by rs-fMRI correlates with Word List Recall performance (r≈0.45). Tau PET (Braak", 'chars': 3063}, {'round': 4, 'persona': 'synthesizer', 'content': '{\n  "ranked_hypotheses": [\n    {\n      "rank": 1,\n      "title": "Hippocampal-Cortical Memory Circuit Disconnection as Primary Driver of AD Cognitive Decline",\n      "mechanism": "Synaptic disconnection between hippocampus, entorhinal cortex, and posteromedial parietal cortex progressively disrupts memory consolidation networks, measurable via DTI tractography before overt atrophy.",\n      "target_gene": "Not applicable (circuit-level)",\n      "confidence_score": 0.85,\n      "novelty_score": 0.55,\n      "feasibility_score": 0.80,\n      "impact_score": 0.90,\n      "composite_score": 0.77,\n      "testable_prediction": "Longitudinal DTI tracking of uncinate fasciculus and parahippocampal cingulum will predict MCI-to-AD conversion with >75% accuracy in asymptomatic carriers of APOE-ε4.",\n      "skeptic_concern": "Correlation between structural disconnection and cognitive decline may not establish causality; upstream molecular drivers remain unspecified."\n    },\n    {\n      "rank": 2,\n      "title": "CDK5-Mediated Tau Propagation Along DMN Structural Connectivity Gradients",\n      "mechanism": "CDK5 hyperactivation in hub neurons drives tau phosphorylation and trans-synaptic spread through connected networks, creating a vulnerability gradient weighted by structural connectivity density.",\n      "target_gene": "CDK5",\n      "confidence_score": 0.60,\n      "novelty_score": 0.75,\n      "feasibility_score": 0.50,\n      "impact_score": 0.85,\n      "composite_score": 0.67,\n      "testable_prediction": "Conditional CDK5 knockout in layer II/III pyramidal neurons of PCC will reduce phospho-tau burden in connected medial PFC regions in 3xTg-AD mice at 12 months.",\n      "skeptic_concern": "The \'independent of amyloid proximity\' claim is unsupported; amyloid-tau synergistic toxicity likely modulates propagation kinetics and regional vulnerability."\n    },\n    {\n      "rank": 3,\n      "title": "Functional Hyperconnectivity as Compensatory Mechanism vs. Early Pathology Marker",\n      "mechanism": "Early-stage hyperconnectivity within DMN and between DMN and frontal regions compensates for accumulating tau burden, but transitions to hypoconnectivity as synaptic failure ensues, creating a biphasic trajectory.",\n      "target_gene": "Not applicable (network-level)",\n      "confidence_score": 0.65,\n      "novelty_score": 0.70,\n      "feasibility_score": 0.70,\n      "impact_score": 0.75,\n      "composite_score": 0.69,\n      "testable_prediction": "rs-fMRI longitudinal monitoring in preclinical AD subjects will demonstrate hyperbolic connectivity trajectory: hyperconnectivity at baseline transitions to hypoconnectivity 18-24 months before cognitive symptom onset.",\n      "skeptic_concern": "Hyperconnectivity findings are highly heterogeneous across studies; may reflect cohort-specific effects rather than universal AD trajectory."\n    }\n  ],\n  "consensus_points": [\n    "Structural disconnection measured by DTI reliably tracks cognitive decline progression and is measurable before atrophy",\n    "The hippocampus and DMN hub regions represent the most vulnerable nodes in AD-related connectome disruption",\n    "Network-level dysfunction precedes and predicts regional atrophy in AD progression"\n  ],\n  "dissent_points": [\n    "Whether tau propagation operates independently of amyloid-β or requires amyloid as a priming event remains unresolved (Skeptic challenges Theorist\'s independence claim)",\n    "Functional hyperconnectivity represents beneficial compensation vs. pathological hypersynchrony remains debated"\n  ],\n  "debate_summary": "The debate converges on hippocampal-cortical disconnection as the most translationally actionable hypothesis for AD staging, while mechanistic debates center on whether CDK5-tau propagation requires amyloid co-factors. Expert consensus prioritizes circuit-level measurable biomarkers over molecular propagation models for near-term clinical application, though both frameworks offer complementary windows into AD progression when integrated with longitudinal connectomics."\n}', 'chars': 4047}]
turns = pd.DataFrame(transcript_rows)
if len(turns):
    summary = turns.groupby('persona').agg(turns=('round','count'), chars=('chars','sum')).sort_values('chars', ascending=False)
    display(summary)
else:
    print('No transcript turns with parseable structure')

if len(turns):
    persona_counts = turns.groupby('persona')['chars'].sum().sort_values()
    fig, ax = plt.subplots(figsize=(8, 4))
    persona_counts.plot(kind='barh', ax=ax, color='#607d8b')
    ax.set_xlabel('Transcript characters')
    ax.set_title('Debate contribution volume by persona')
    ax.grid(axis='x', alpha=0.25)
    plt.tight_layout(); plt.show()
else:
    print('No transcript text available for plotting')

excerpts = [{'round': 2, 'persona': 'skeptic', 'excerpt': 'I\'d be happy to critically evaluate hypotheses for you, but I don\'t see any specific hypotheses provided in your message. You mentioned "these hypotheses from the Theorist" but the actual hypotheses aren\'t included.  Could you please provide the specific hypotheses you\'d like me to evaluate? Once you share them, I\'ll:  1. **Identify specific weaknesses** - Look for logical gaps, unsupported assumptions, and methodolo', 'chars': 863}, {'round': 3, 'persona': 'domain_expert', 'excerpt': 'I notice that your message mentions "these hypotheses" but I don\'t see any specific hypotheses included in your request. You\'ve outlined an excellent framework for evaluation that includes:  **Critical Analysis:** - Identifying specific weaknesses - Citing counter-evidence  - Proposing falsification experiments - Providing revised confidence scores  **Feasibility Assessment:** - Druggability assessment - Existing com', 'chars': 1174}, {'round': 4, 'persona': 'synthesizer', 'excerpt': '```json {   "ranked_hypotheses": [],   "knowledge_edges": [],   "synthesis_summary": "No hypotheses were provided in the input for evaluation and ranking. The request mentions evaluating hypotheses but does not include any specific hypotheses to analyze, critique for feasibility, or synthesize into ranked outputs. Without concrete hypotheses containing testable claims, molecular targets, or therapeutic approaches, it', 'chars': 805}, {'round': 1, 'persona': 'theorist', 'excerpt': "# Mechanistically-Specific Hypotheses: Connectomics in Alzheimer's Disease Cognitive Decline  ---  ## Hypothesis 1: CDK5-Mediated Tau Phosphorylation Propagates Along DMN Structural Connectivity Gradients  **Title**: CDK5 drives trans-synaptic tau propagation in DMN  **Mechanism**: Cyclin-dependent kinase 5 (CDK5) hyperactivation by calpain cleavage of p35 to p25 in vulnerable neurons enhances tau phosphorylation at ", 'chars': 1165}, {'round': 2, 'persona': 'skeptic', 'excerpt': '# Critical Evaluation: Connectomics Hypotheses in Alzheimer\'s Disease  I\'ll systematically evaluate these mechanistic hypotheses, focusing on the CDK5-tau propagation framework as presented, while noting where the other six research domains introduce additional complications.  ---  ## Hypothesis 1: CDK5-Mediated Tau Propagation Along DMN Connectivity Gradients  ### 1. Strongest Specific Weakness  **The "independent o', 'chars': 3996}, {'round': 3, 'persona': 'domain_expert', 'excerpt': "# Domain Expert Evaluation: Connectomics Hypotheses in Alzheimer's Disease Cognitive Decline  ---  ## 1. Hypotheses with Highest Translational Potential  ### Tier 1 (Highest Priority)  **Hypothesis 2 (Highest Translation Potential): Hippocampal-Cortical Memory Circuit Disconnection**  This is the most translationally ready of the six domains for the following reasons: - Structural disconnection between hippocampus, e", 'chars': 3063}, {'round': 4, 'persona': 'synthesizer', 'excerpt': '{   "ranked_hypotheses": [     {       "rank": 1,       "title": "Hippocampal-Cortical Memory Circuit Disconnection as Primary Driver of AD Cognitive Decline",       "mechanism": "Synaptic disconnection between hippocampus, entorhinal cortex, and posteromedial parietal cortex progressively disrupts memory consolidation networks, measurable via DTI tractography before overt atrophy.",       "target_gene": "Not applica', 'chars': 4047}]
pd.DataFrame(excerpts)

edges = []
if edges:
    pd.DataFrame(edges)
else:
    print('No knowledge graph edges were recorded for this debate-only analysis')